U.S. patent application number 14/508264 was filed with the patent office on 2015-03-26 for system and method for ranking search results within citation intensive document collections.
This patent application is currently assigned to LexisNexis, a division of Reed Elsevier Inc.. The applicant listed for this patent is LexisNexis, a division of Reed Elsevier Inc.. Invention is credited to Harry R. Silver, Ling Qin Zhang.
Application Number | 20150088875 14/508264 |
Document ID | / |
Family ID | 42981764 |
Filed Date | 2015-03-26 |
United States Patent
Application |
20150088875 |
Kind Code |
A1 |
Zhang; Ling Qin ; et
al. |
March 26, 2015 |
System and Method For Ranking Search Results Within Citation
Intensive Document Collections
Abstract
Systems and methods facilitate a search and identify documents
and associated metadata reflecting content of the documents. In one
implementation, a method receives a query comprising a set of
search terms, identifies a stored document in response to the
query, and determines a score value for the retrieved document
based on a similarity between one or more of the query search terms
and metadata associated with the identified document. The method
locates the identified document in a citation network of baseline
query results, the citation network comprising a first set of
documents that cite to the identified document and a second set of
documents cited to by the identified document. The method further
determines a new score value of the identified document as a
function of the score value and a quantity and a quality of
documents within the first and second set of documents.
Inventors: |
Zhang; Ling Qin;
(Springboro, OH) ; Silver; Harry R.; (Shaker
Heights, OH) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
LexisNexis, a division of Reed Elsevier Inc. |
Miamisburg |
OH |
US |
|
|
Assignee: |
LexisNexis, a division of Reed
Elsevier Inc.
Miamisburg
OH
|
Family ID: |
42981764 |
Appl. No.: |
14/508264 |
Filed: |
October 7, 2014 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
13403253 |
Feb 23, 2012 |
8886638 |
|
|
14508264 |
|
|
|
|
12385668 |
Apr 15, 2009 |
8150831 |
|
|
13403253 |
|
|
|
|
Current U.S.
Class: |
707/726 |
Current CPC
Class: |
G06F 16/3334 20190101;
G06F 16/24578 20190101; Y10S 707/942 20130101; G06F 16/382
20190101; G06F 16/93 20190101 |
Class at
Publication: |
707/726 |
International
Class: |
G06F 17/30 20060101
G06F017/30 |
Claims
1. A computerized method for calculating a normalized activity
score value to rank an identified document, the method comprising:
determining a number of times an identified document was cited in a
subject matter community of the identified document; determining a
probability distribution that individual documents within the
subject matter community are cited a variable number of times by
other individual documents in the subject matter community;
calculating a probability function by performing a regression on
the probability distribution; calculating the activity score value
according to an activity score function formulated as an inverse of
the probability function; storing in computer memory a ranking of
the identified document based on the activity score value; and
outputting on a display device, a report reflecting the ranking of
the identified document.
2. The method of claim 1, further comprising weighting the activity
score value by an age of the identified document.
3. The method of claim 1, wherein the subject matter community
comprises a jurisdiction of a legal community.
4. The method of claim 1, wherein the calculating the activity
score further comprises: generating an activity score vector;
calculating an individual activity score value for dimensions of
the activity score vector, wherein at least one of the dimensions
is associated with the number of times the identified document was
cited in a subject matter community of the identified document;
weighing at least one of the individual activity score values by an
age of the identified document; and adding the individual activity
score values to obtain a total activity score value.
5. The method of claim 4, wherein the dimensions of the activity
score vector comprise at least one of U.S. Supreme Court cites, law
review articles, treatises, signal, or treatment in addition to the
dimension associated with the number of times the identified
document was cited in a subject matter community of the identified
document.
6. The method of claim 1, wherein the activity score value of the
identified document is normalized in relation to other documents in
the subject matter community.
7. A non-transitory computer-readable medium comprising program
instructions, which, when executed by a processor, cause the
processor to perform a method for calculating a normalized activity
score to rank an identified document, the method comprising:
identifying a stored document; determining a number of times the
identified document was cited in a subject matter community of the
identified document; determining a probability distribution that
individual documents within the subject matter community are cited
a variable number of times by other individual documents in the
subject matter community; calculating a probability function by
performing a regression on the probability distribution;
calculating the activity score value according to an activity score
function formulated as an inverse of the probability function;
storing in computer memory a ranking of the identified document
based on the activity score value; and outputting on a display
device, a report reflecting the ranking of the identified
document.
8. The non-transitory computer-readable medium of claim 7, the
method further comprising weighing the activity score value by an
age of the identified document.
9. The non-transitory computer-readable medium of claim 7, wherein
the subject matter community comprises a jurisdiction of a legal
community.
10. The non-transitory computer-readable medium of claim 9, wherein
the calculating the activity score further comprises: generating an
activity score vector; calculating an individual activity score
value for dimensions of the activity score vector, wherein at least
one of the dimensions is associated with the number of times the
identified document was cited in a subject matter community of the
identified document; weighing at least one of the individual
activity score values by an age of the identified document; and
adding the individual activity score values to obtain a total
activity score value.
11. The non-transitory computer-readable medium of claim 10,
wherein the dimensions of the activity score vector comprise at
least one of U.S. Supreme Court cites, law review articles,
treatises, signal, or treatment in addition to the dimension
associated with the number of times the identified document was
cited in a subject matter community of the identified document.
12. The non-transitory computer-readable medium of claim 10,
wherein the activity score value of the identified document is
normalized in relation to other documents in the subject matter
community.
Description
TECHNICAL FIELD
[0001] Systems and methods consistent with disclosed embodiments
display and rank a set of retrieved documents in response to a
query.
BACKGROUND INFORMATION
[0002] Conventional search tools return a list of documents in
response to a search query. The documents from the list may be
ranked according to their relevance to the search query. For
example, highly relevant documents may be ranked higher than, and
may be displayed in a list above, documents of a lesser relevance.
This allows a user to quickly and conveniently identify the most
relevant documents retrieved in response to the query.
[0003] Some conventional search tools allow a user to perform a
query using natural language. For example, LexisNexis.RTM. uses
Freestyle.TM. to enable users to submit query terms associated with
a case or legal concept. The search tool then returns a ranked list
of legal documents matching the query terms. The search tool may
rank the legal documents based upon a number of times the query
terms appear in the legal document. For example, a term "patent"
may occur in a first document 50 times, and may occur in a second,
similarly sized document, 10 times. If the user entered a query for
"patent," the search tool would deem the first document to be more
relevant than the second document because it includes the term
"patent" more times. In this instance, frequency and size are used
to determine ranking. Therefore, the search tool would assign the
first document a higher ranking than the second document.
[0004] With more complex queries, search tools may use word vectors
when comparing a query with a document. Generally, a vector can be
represented as a line segment with a direction and a magnitude. In
a two-dimensional space, a two dimensional vector V=[x, y] can be
graphed with a start point at the origin (0,0) of the graph and an
endpoint at a coordinate (x, y) of the graph. A similarity between
any two vectors in the two dimensional space can be determined by
calculating the cosine of the angle .theta. between the two
vectors.
[0005] However, vectors can theoretically be defined across any
number of dimensions n, such that V=[x, y, . . . n]. While it is
not possible to graphically model vectors over 3 dimensions, it is
still possible to perform mathematical operations on these
multidimensional vectors. For example, it is possible to determine
an angle .theta. between two vectors that are defined over 3
dimensions, and to determine the similarity between those two
vectors by calculating the cosine of the angle .theta..
[0006] Word vectors can be used to model any string of words, such
as a document or a natural language query. The vectors can be
defined according to a number of concepts in the English language.
For example, if a modern thesaurus includes 1000 concepts, then
each word vector would include 1000 dimensions. In other words,
V=[x, y, . . . n] where n=1000. Each dimension in the vector would
correspond to a unique one of the 1000 concepts, and a number in
any particular dimension of the vector is the number of times that
the concept corresponding to that dimension occurred in the query
or document.
[0007] The following example shows a comparison between a document
and a query using word vectors. The concepts from this example can
also apply to a comparison between any two sets of words, such as
between two documents. Table 1 illustrates an exemplary set of
concepts along with words related to each concept.
TABLE-US-00001 TABLE 1 Concept Definitions Concept No. Words 1 the,
a 2 attractive, nice, beautiful 3 rose, carnation, pansy 4 white,
pink, purple
[0008] Table 2 illustrates an exemplary set of word strings, along
with words included in each word string.
TABLE-US-00002 TABLE 2 Documents Word String. Text Document the
nice, attractive white rose Query the beautiful carnation
[0009] Table 3 illustrates a vectorization of the document and the
query from Table 2 using the concepts from Table 1.
TABLE-US-00003 TABLE 3 Vectorization Word String Vector
Categorization Document [1, 2, 1, 1] [the; nice, attractive; rose;
white] Query [1, 1, 1, 0] [the; beautiful; carnation; null]
[0010] The dimensions from the vectors in Table 3 correspond to the
concepts set forth in Table 1, such that dimension 1 of each vector
corresponds to concept 1, dimension 2 corresponds to concept 2, and
so on. Accordingly, the document includes one term from concept 1
("the"), and so a "1" is assigned to dimension 1 of its vector. The
document includes two terms from concept 2 ("nice" and
"attractive"), and so a "2" is assigned to dimension 2 of its
vector. The remaining dimensions in the document vector, as well as
the dimensions for the query vector, are filled in this manner.
[0011] Once the document vector and query vector are calculated in
this example, it is possible to mathematically determine the angle
.theta. between them. Therefore, it is also possible to determine
the similarity between the query and the document by calculating
the cosine of the angle .theta. between their respective word
vectors. This similarity value can be compared with the similarity
value of the same query with a different document. In this way, the
search tool may rank the documents depending on their similarity
with respect to the query. Phrase vectors may also be used in
addition to, or instead of word vectors.
[0012] This technique may not be the best indicator of relevance.
For one thing, it relies fundamentally on the frequency of terms
within a particular class. It also ignores other factors that may
be important in determining relevance and ranking.
[0013] Accordingly, there is a need to improve the ranking of
search results in response to a query.
SUMMARY
[0014] In accordance with one embodiment, there is provided a
computer implemented method for facilitating a search and
identification of documents and associated metadata reflecting
content of the documents stored in a memory device. The method
involves receiving a query comprising a set of search terms,
identifying a stored document in response to the query, determining
a score value for the retrieved document based on a similarity
between one or more of the query search terms and metadata
associated with the identified document, and locating the
identified document in a citation network of baseline query
results. The citation network may include a first set of documents
that cite to the identified document and a second set of documents
cited to by the identified document. The method further involves
determining a new score value of the identified document as a
function of the score value and a quantity and a quality of
documents within the first and second set of documents, ranking the
identified document based on the new score value, and outputting on
a display device, a report reflecting the ranking of the identified
document.
[0015] In accordance with another embodiment, there is provided a
computer-readable medium comprising program instructions, which,
when executed by a processor, cause the processor to perform a
method for facilitating a search and identification of documents
and associated metadata reflecting content of the documents stored
in a memory device. The method involves receiving a query
comprising a set of search terms, identifying a stored document in
response to the query, determining a score value for the identified
document based on a similarity between one or more of the query
search terms and metadata associated with the identified document,
and locating the identified document in a citation network of
baseline query results. The citation network may include a first
set of documents that cite to the identified document and a second
set of documents cited to by the identified document. The method
further involves determining a new score value of the identified
document as a function of the score value and a quantity and a
quality of documents within the first and second set of documents,
ranking the identified document based on the new score value, and
outputting on a display device, a report reflecting the ranking of
the identified document.
[0016] In accordance with another embodiment, there is provided a
computer system, including memory and at least one processor for
facilitating a search and identification of documents and
associated metadata reflecting content of the documents stored in a
memory device. The system includes a processor receiving a query
comprising a set of search terms, and identifying a stored document
in response to the query. The system also includes an IR score
generating component determining a score value for the identified
document based on a similarity between one or more of the query
search terms and metadata associated with the identified document.
The system also includes a citation network of baseline query
results comprising a first set of documents that cite to the
identified document and a second set of documents cited to by the
identified document. The system also includes a citation component
locating the identified document in the citation network, and
determining a new score value of the identified document as a
function of the score value and a quantity and a quality of
documents within the first and second set of documents. The system
also includes a display device displaying a report reflecting a
ranking the identified document based on the new score value.
[0017] In accordance with yet another embodiment, there is provided
a computerized method for calculating an activity score value to
rank an identified document. The method involves identifying a
stored document, determining a number of times the identified
document was cited in a subject matter community of the identified
document, determining a probability distribution that individual
documents within the subject matter community are cited a variable
number of times by other individual documents in the subject matter
community, calculating the activity score value according to a
probability that the individual documents in the subject matter
community are cited at least the number of times the identified
document was cited in the subject matter community, and storing in
computer memory a report reflecting a ranking of the identified
document based on the activity score value.
[0018] In accordance with still yet another embodiment, there is
provided a computer-readable medium comprising program
instructions, which, when executed by a processor, cause the
processor to perform a method for calculating an activity score
value to rank an identified document. The method involves,
identifying a stored document, determining a number of times the
identified document was cited in a subject matter community of the
identified document, determining a probability distribution that
individual documents within the subject matter community are cited
a variable number of times by other individual documents in the
subject matter community, calculating the activity score value
according to a probability that the individual documents in the
subject matter community are cited at least the number of times the
identified document was cited in the subject matter community, and
storing in computer memory a report reflecting a ranking of the
identified ranking based on the activity score value.
[0019] It is to be understood that both the foregoing general
description and the following detailed description are exemplary
and explanatory only and are not restrictive of the disclosed
embodiments, as claimed.
BRIEF DESCRIPTION OF THE DRAWINGS
[0020] The accompanying drawings, which are incorporated in and
constitute a part of this specification, illustrate various
disclosed embodiments. In the drawings:
[0021] FIG. 1 includes a block diagram of system components for
ranking search results in accordance with one embodiment;
[0022] FIG. 2 includes a block diagram of components that may be
used in ranking search results in accordance with one
embodiment;
[0023] FIG. 3 includes a flow diagram illustrating a ranking of
search results in accordance with one embodiment;
[0024] FIG. 4 includes a flow diagram illustrating a calculation of
an IR score using metadata in accordance with one embodiment;
[0025] FIG. 5 includes a citation network illustrating citation
relationships among baseline documents in accordance with one
embodiment;
[0026] FIG. 6 includes a flow diagram illustrating a weighting of
an IR score using citations in accordance with one embodiment;
[0027] FIG. 7 includes a flow diagram illustrating an activity
score calculation for a legal document in accordance with one
embodiment;
[0028] FIG. 8 shows a linear graph used to illustrate the
relationship of number of case cites of a particular case to
activity score in accordance with one embodiment;
[0029] FIG. 9 shows a graph used to illustrate a probability of
cases within a subject matter community are cited a variable number
of times in accordance with one embodiment; and
[0030] FIG. 10 includes an exemplary results page generated in
response to a query of a legal database in accordance with one
embodiment.
DETAILED DESCRIPTION
[0031] Disclosed systems and methods may facilitate a search and
identify documents and associated metadata reflecting the content
of the documents. For example, disclosed embodiments may rank
search results within citation intensive document collections. To
do so, disclosed embodiments may identify a stored document in
response to a query and determine a score value for the document
based on a similarity between one or more of query search terms and
metadata associated with the document. Disclosed embodiments may
further locate the document in a citation network of baseline query
results. For example, the citation network may include a first set
of documents that cite to the document and a second set of
documents that are cited to by the identified document. Disclosed
embodiments may determine a new score value for the document as a
function of the score value and a quantity and a quality of
documents within the first and second set of documents.
[0032] Reference will now be made in detail to disclosed
embodiments, examples of which are illustrated in the accompanying
drawings. Wherever possible, the same reference numbers will be
used throughout the drawings to refer to the same or like
parts.
[0033] FIG. 1 is an exemplary system 100, consistent with a
disclosed embodiment. Although a specific number of components are
depicted in FIG. 1, any number of these components may be provided.
Furthermore, the functions provided by one or more components of
system 100 may be combined or separated. Moreover, the
functionality of any one or more components of system 100 may be
implemented by any appropriate computing environment.
[0034] With reference to FIG. 1, system 100 may include device 102,
network 112, and database 114. Device 102 may be used for
performing computing tasks, such as legal research and other types
of research. For example, device 102 may be a desktop computer,
laptop computer, or other mobile device. Device 102 may include
central processing unit (CPU) 104, memory 106, user interface 108,
and/or I/O unit 110.
[0035] CPU 104 may execute computer program instructions to perform
various processes and methods. CPU 104 may read the computer
program instructions from memory 106 or from any computer-readable
medium. Memory 106 may include random access memory (RAM) and/or
read only memory (ROM) configured to access and store information
and computer program instructions. Memory 106 may also include
additional memory to store data and information, and/or one or more
internal databases (not shown) to store tables, lists, or other
data structures. Moreover, user interface 108 may access user data,
such as a user supplied query. In some embodiments, user interface
108 may be separate from device 102. User interface 108 may also
include a visual display. Furthermore, I/O unit 110 may access data
over network 112.
[0036] Network 112 in system 100 may facilitate communications
between the various devices in system 100, such as device 102 and
database 114. In addition, device 102 may access legacy systems
(not shown) via network 112, or may directly access legacy systems,
databases, or other network applications. Network 112 may be a
shared, public, or private network, may encompass a wide area or
local area, and may be implemented through any suitable combination
of wired and/or wireless communication networks. Furthermore,
network 112 may comprise a local area network (LAN), a wide area
network (WAN), an intranet, or the Internet.
[0037] Database 114 in system 100 may include an organized set of
documents 116 and/or information about documents 116. Documents may
be associated with any subject matter, such as legal, scientific,
financial, and/or political. Information about documents 116 may
include data about the documents (e.g., metadata), for example,
data from a Shepard's.RTM. database, which is applicable to legal
documents. Moreover, data about the documents (e.g., metadata) may
be included within the documents themselves. For example, for legal
cases, the metadata may include a listing of core terms and
headnotes. The listing of core terms may capture the content of the
document. Moreover, the headnotes may be descriptive text blocks
located with the document. Headnotes may briefly summarize the
content of the baseline document. The headnotes may include
headnotes from LexisNexis.RTM.. The documents 116 may pertain to
any subject and, accordingly, information about documents 116 may
pertain to or relate to information that is associated with the
documents 116.
[0038] Moreover, although shown as separate components in FIG. 1,
database 114 and device 102 may be combined. For example, device
102 may include one or more databases in addition to or instead of
database 114. Database 114 may also be distributed over multiple
databases.
[0039] FIG. 2 shows memory 200, which may include components used
for ranking search results. Memory 200 may be similar to memory 106
from FIG. 1. Memory 200 may include initial result component 202,
baseline component 204, IR (information retrieval) score generating
component 206, citation component 208, and activity score component
210. These components may be implemented in hardware, software, or
firmware, or a combination.
[0040] Initial result component 202 may calculate an initial result
set of initial documents in response to a query. The initial result
set may be a list of the initial documents that represents a first
cut at identifying documents satisfying the query. However, in some
cases, the initial result set may not include important documents
that are relevant to the query. In these situations, the initial
result may be augmented to include additional documents.
[0041] Baseline component 204 may augment the initial result set to
include additional documents that are relevant to the query. In
particular, baseline component 204 may determine additional
documents that are frequently cited by the initial documents from
the initial result set. These additional documents may be added to
the initial result set to create a baseline result. Baseline
documents may be documents within the baseline result. The baseline
documents may include the initial documents and the additional
documents. Alternatively, the baseline documents may include only
the initial documents without any additional documents. After the
baseline documents are identified, they may then be ranked
according to their relevance with respect to the query.
[0042] IR score generating component 206 may retrieve metadata from
within the baseline documents themselves (e.g., within database
114). The metadata may describe the baseline documents, for example
by including core terms and/or headnotes. IR score generating
component 206 may compare the metadata of the baseline document to
the query. Accordingly, IR score generating component 206 may also
generate an IR score for the baseline documents using the metadata.
The IR score may represent a similarity between the baseline
documents and the query.
[0043] Citation component 208 may weight an IR score of a
particular baseline document according to citations from other
baseline documents (i.e., in-cites) and/or citations to other
baseline documents (i.e., out-cites). Activity score component 210
may generate an activity score to further weight the IR scores
according to how well-known the scored case is within the legal
community as a whole. Alternatively, the activity score may not
weight the IR score, and may be used independently from the IR
score.
[0044] FIG. 3 illustrates the operations of an exemplary method 300
for ranking search results. Method 300 may be executed by CPU 104,
alone or in conjunction with other components. In block 302, CPU
104 may receive a query, for example, a search query for documents.
The query may be received from a user via user interface 108 or may
be automatically generated. In block 304, CPU 104 may execute
initial result component 202 to calculate an initial result set of
the query, for example, by identifying a list of initial documents
that are generally relevant to the query. The initial documents may
be identified according to a frequency with which query terms occur
within the initial documents, the proximity with which query terms
occur within the initial documents, and/or other factors. The
initial documents may reside in database 114 and may be associated
with documents 116.
[0045] In block 305, CPU 104 may execute baseline component 204 to
calculate a baseline result with baseline documents. The baseline
documents may include the initial result documents and additional
documents. The additional documents may be documents that are
relevant to the query, but are not included with the initial result
set. The additional documents may be documents that are frequently
cited by the initial documents. Moreover, the additional documents
may be documents that are cited, more than a threshold number of
times, by the initial documents.
[0046] In block 306, CPU 104 may execute IR score generating
component 206 to calculate an information retrieval (IR) score for
each of the baseline documents using metadata. As discussed, the
metadata may include data describing the baseline documents, and
may be located within the baseline documents themselves and/or in
database 114. Specifically, metadata from each of the baseline
documents may be compared with the query to calculate the IR score.
Accordingly, the IR score may represent a similarity between each
of the baseline documents and the query.
[0047] In block 308, CPU 104 may execute citation component 208 to
weigh the IR score of the baseline documents using citation
information. An IR score for a baseline document may be increased
if it is cited by other baseline documents (i.e., in-cites) and/or
if it cites to other baseline documents (i.e., out-cites).
[0048] In block 310, CPU 104 may execute activity score component
210 to weigh the IR score of the baseline documents using an
activity score. An IR score for a baseline document may be
increased if it is famous and well known in the legal community as
a whole. Alternatively, the activity score may not be used to
weight the IR score, and may be used independently from the IR
score.
[0049] In block 312, CPU 104 may rank the baseline documents
according to weighted IR scores. Alternatively, or in addition, CPU
104 may rank the baseline documents according to the activity
score. In block 314, CPU 104 may cause the baseline documents to be
displayed in accordance with their rank. For example, higher ranked
baseline documents may be displayed higher on a list than lower
ranked baseline documents. One of ordinary skill will recognize
that any blocks 302-314 may be omitted and/or executed in any
order.
[0050] FIG. 4 is a flow diagram 400 of the operations involved in
calculating an information retrieval (IR) score using metadata. In
accordance with one implementation consistent with a disclosed
embodiment, flow diagram 400 may be a process implemented by an IR
score generating component 206.
[0051] In block 402, IR score generating component 206 may receive
a baseline result, for example, from baseline component 204. As
discussed, the baseline result may identify baseline documents that
are relevant to or that meet a query. A baseline document within
the baseline result may be located in database 114. Moreover, the
baseline document may include metadata portions that describe
contents of the baseline document. The metadata portions may
include a listing of core terms and headnotes. The listing of core
terms may capture the content of the baseline document. Moreover,
the headnotes may be descriptive text blocks located with the
baseline document. Headnotes may briefly summarize the content of
the baseline document. The headnotes may include headnotes from
LexisNexis.RTM..
[0052] In block 404, IR score generating component 206 may access a
baseline document in the baseline result, for example, from
database 114. In block 406, IR score generating component 206 may
access a metadata portion located within the accessed baseline
document. The metadata portion may include the core terms or may
include the headnotes.
[0053] In block 408, IR score generating component 206 may
calculate a similarity between the query and the accessed metadata
portion. For example, a similarity may be calculated between the
query and the core terms of the metadata and/or between the query
and the headnotes of the metadata.
[0054] If calculating a similarity between the query and the
headnotes, suppose q represents a word vector from the query and H
represents a word vector from the headnotes. In this case,
similarity(H,q)=|H|*cos .theta. [Equation 1]
where .theta. is the angle between the word vectors H and q. In
other words, the similarity between the word vector q (for the
query) and the word vector H (for the headnotes) equals the
magnitude of the word vector H, multiplied by the cosine of the
angle .theta. between q and H.
[0055] Alternatively, or in addition, if calculating a similarity
between the query and the core terms, suppose q still represents
the word vector from the query and T represents a word vector from
core terms. In this case,
similarity(T,g)=|T|*cos .tau. [Equation 2]
where .tau. is the angle between the word vectors T and q. In other
words the, similarity between the word vector q (for the query) and
the word vector T (for the core terms) equals the magnitude of the
word vector T, multiplied by the cosine of the angle .tau., between
q and T.
[0056] As discussed, the metadata in the baseline document may
include both headnotes and core terms. In some embodiments, IR
score generating component 206 may retrieve and process the
headnotes and core terms separately. Accordingly, after the
similarity data is calculated for either the headnotes (according
to Equation 1), the core terms (according to Equation 2), or both,
IR score generating component 206 may determine, in block 410,
whether any metadata remains within the baseline document. If the
headnotes were already processed, then core terms may remain.
Alternatively, if the core terms were already processed, then the
headnotes may remain.
[0057] If metadata from the baseline document remains un-retrieved
and unprocessed (410--Yes), then process returns to block 406 to
retrieve and process the remaining metadata. If no metadata from
the baseline document remains un-retrieved and unprocessed
(410--No), then all metadata has been considered, and in block 412,
IR score generating component 206 may generate the IR score for the
baseline document according to the processing of the headnotes
(according to Equation 1) and core terms (according to Equation 2)
from the metadata.
[0058] For example, suppose similarity data for the core terms and
the headnotes, with respect to the query, are calculated according
to Equation 1 and Equation 2 above. IR score generating component
206 may then add and weight these values to generate the IR score
for a particular baseline document. For example, for a document d
that includes the core terms and headnotes,
IRScore(d,q)=w1*similarity(H,q)+w2*similarity(T,q) [Equation 3]
where w1 and w2 are parameter variables to weight the similarity
data, and may be determined experimentally. Thus, the IR score of
document d, with respect to query q, equals the similarity between
the headnotes of the document and the query on the one hand
(Equation 1), added to the similarity between the core terms of the
document and the query (Equation 2) on the other hand.
[0059] Next, in block 414, IR score generating component 202 may
determine whether any baseline documents from the baseline result
remain unscored by IR score generating component 206. If baseline
documents from the baseline result remain unscored (414--Yes), then
process returns to block 404 to process the remaining baseline
documents. If no baseline documents from the baseline result remain
unscored (414--No), then all baseline documents have been
considered and scored by IR score generating component 206.
Accordingly, in block 414, IR score generating component 206 may
send the IR score(s) to citation component 208. One of ordinary
skill will recognize that any blocks 402-416 may be omitted and/or
executed in any order.
[0060] FIG. 5 illustrates a citation network 500 including citation
relationships among baseline documents, in accordance with one
implementation consistent with a disclosed embodiment. Citation
network 500 may be stored in memory 106 and/or database 114.
Citation network 500 may represent a relationship among documents,
such as documents 116.
[0061] Citation network may include document d 501, in citation
sub-network 502, out citation sub-network 504, no citation
sub-network 506, and documents 1-10. In some embodiments, citation
network 500 may include only the baseline documents from the
baseline results that are retrieved in response to a query.
[0062] Document d 501 may be a particular baseline document from
the baseline results. In citation sub-network 502 may include
baseline documents that include a citation or reference to document
d 501. Specifically, baseline documents 1, 2, 3, and 4 within in
citation sub-network 502 cite to document d 501, as indicated by
the arrows pointing from baseline documents 1, 2, 3, and 4 to
document d 501. Thus, each of documents 1, 2, 3, 4 include a
reference to d 501.
[0063] Moreover, out citation sub-network 504 may include baseline
documents that are cited to by document d 501. Specifically,
baseline documents 4, 5, and 10, within out citation sub-network
504 are cited by d 501, as indicated by the arrows pointing from
document d 501 to baseline documents 4, 5, and 10. Thus, d 501
includes a references to each of documents 4, 5, and 10.
[0064] Furthermore, no citation sub-network 504 may include
baseline documents that have no direct relationship to document d
501. Specifically, baseline documents 6, 7, 8, and 9 within no
citation sub-network 506 have no direct citation relationship with
d 501, and thus, do not have any arrows showing a relationship
directly to document d 501. Thus, none of documents 6, 7, 8, and 9
include a reference to d 501, nor does d 501 include a reference to
any of documents 6, 7, 8, and 9.
[0065] As discussed, baseline documents represent a first cut of
retrieving relevant documents that are responsive to a query. To
that end, a "meta rank score" takes into account citation
relationships among the baseline documents to rank the baseline
documents according to their relevancy with respect to the query.
For example, baseline documents that have a large number of
relationships with other baseline documents may be particularly
relevant to the query. The citation relationships used to calculate
the meta rank score may be similar to those discussed in connection
with FIG. 5.
[0066] FIG. 6 includes a flow diagram 600 illustrating a use of
citation relationships for calculating the meta rank score. Flow
diagram 600 may be a process implemented by citation component 208.
Citation component 208 may use citation relationships specified in
citation network 500 to calculate the meta rank score.
[0067] In block 602, citation component 208 may receive an IR score
for a document d, for example, from IR score generating component
206. As discussed, the IR score generating component 206 may have
calculated the IR score according to metadata (including headnotes
and core terms). The IR score may now be used to calculate document
d's meta rank score in combination with the citation relationships
among baseline documents. As discussed, the meta rank score may be
an accurate indication of d's relevance to the query because it
takes into account citation relationships between d and other
baseline documents that were returned as baseline results to the
query.
[0068] Citation component 208 may use in-cites and out-cites in
calculating the meta rank score for document d. In-cites may relate
to baseline documents that cite to document d. For example,
in-cites were discussed in connection with in citation sub-network
502 from FIG. 5. Furthermore out-cites may relate to baseline
documents that document d cites to. For example, out-cites were
discussed in connection with out citation sub-network 504 from FIG.
5.
[0069] In block 604, citation component 208 may calculate the meta
rank score using in-cites to d. Thus, the documents from
in-citation network 502 may be used to calculate d's meta rank
score. For example, d's meta rank score may depend on a number of
baseline documents that cite to d. The meta rank score may equal
the IR score for d (e.g. as previously calculated in equation 3),
plus an added amount for each baseline document that cites to d.
Moreover, the relevance of the baseline documents that cite to d
may be also considered when calculating the meta rank score for d.
For example,
metaRankScoreIn ( d ) = IRScore ( d , q ) + t .di-elect cons. C ( d
) log 2 ( IRScore ( c , q ) ) [ Equation 4 ] ##EQU00001##
for C(d), a set of legal documents c that cite to d, for example
baseline documents within in citation sub-network 502. In other
words, a meta rank score for d is the IR score of d (the first term
of Equation 4), plus the log.sub.2 of the IR score of each of the
documents c that cite to d (the second term of Equation 4). In this
way, the second term of Equation 4 takes into account not only the
number of documents c that cite to d (by virtue of the summation),
but also the quality of documents c. The quality of documents c is
determined in Equation 4 according to the relevance of documents c
to the query q (by virtue of their IR score in the second term of
Equation 4).
[0070] Instead of using Equation 4 to calculate the meta rank score
for in-cites, the following equation may alternatively be used,
metaRankScoreIn ( d ) = IRScore ( d , q ) + c .di-elect cons. C ( d
) IRScore ( c , q ) * smetaSimilarity ( c , d ) [ Equation 5 ]
##EQU00002##
where new term smetaSimilarity(c,d) is a calculation directly
comparing the headnotes of each of documents c with the headnotes
from d, and also comparing the core terms of each of documents c,
with the core terms from d. Moreover, smetaSimilarity(c,d) can be
mathematically calculated by Jaccard measure as follows,
smetaSimilarity ( c , d ) = w 1 * Core ( c ) Core ( d ) Core ( c )
Core ( d ) + w 2 * HN ( c ) HN ( d ) HN ( c ) HN ( d ) [ Equation 6
] ##EQU00003##
where Core(c) and HN(c) are the set of core term and headnotes,
respectively, for documents c, and Core(d) and HN(d) are the set of
core term and headnotes, respectively, for d. In other words, an
additional way of measuring the quality of documents c that cite to
d, is to value the degree to which each document c is similar to d,
by comparing the headnotes and core terms of each. Specifically the
numerator of the first fraction of equation 6 is the number of core
terms common to both document c and d (exemplified by the
intersection operator .andgate.). The denominator of the first
fraction of equation 6 is the total number of distinct core terms
in document c and d (exemplified by the union operator .orgate..
Moreover, the numerator of the second fraction of equation 6 is the
number of headnotes common to both document c and d (exemplified by
the intersection operator .andgate.). The denominator of the second
fraction of equation 6 is the total number of distinct headnotes in
document c and d (exemplified by the union operator .orgate.). In
this way, smetaSimilarity(c,d) may be calculated using the in-cites
to d.
[0071] In addition to, or instead of, computing the meta rank score
using in-cites as in block 604, out-cites can be used in computing
the meta rank score. Thus, the documents from out-citation network
504 may be used to calculate d's meta rank score. Accordingly, in
block 606, citation component 208 may calculate the meta rank score
using out-cites from d. For example, d's meta rank score may depend
on a number of baseline documents cited by d. The meta rank score
may equal the IR score for d (e.g. as previously calculated in
equation 3), plus an added amount for each baseline document cited
by d. Moreover, the relevance of the baseline documents cited by d
are also considered when calculating the meta rank score. For
example,
metaRankScoreOut ( d ) = IRScore ( d , q ) + e .di-elect cons. E (
d ) IRScore ( e , q ) a - .beta. [ Equation 7 ] ##EQU00004##
for E(d), a set of legal documents e that are cited by d (for
example baseline documents within out citation sub-network 504),
and where .alpha. and .beta. may be experimentally determined, and
may be initially set as .alpha.=1/2 and .beta.=0. In other words, a
meta rank score for d is the IR score of d (the first term in
Equation 7), plus the IR score of each of the documents e cited by
d (the second term in Equation 7). As mentioned, the third term in
Equation 7, .beta., may be experimentally determined. This way, the
second term of Equation 7 takes into account not only the number of
documents e cited by d, but also the quality of documents e. The
quality of documents c is determined in Equation 7 according to the
relevance of the documents e to the query q (by virtue of their IR
score in the second term of Equation 7).
[0072] Instead of using Equation 7 to calculate the meta rank score
for out-cites, the following equation may be used,
metaRankScore ( d ) Out = IRScore ( d , q ) + e .di-elect cons. E (
d ) IRScore ( e , q ) * t .di-elect cons. T ( d ) IRScore ( t , q )
T ( d ) [ Equation 8 ] ##EQU00005##
for T(d), a set of topics included in d. In other words, a meta
rank score for d is the IR score of d, plus the IR score of each of
the documents e cited by d, weighted by an IR score of the topical
relevance of d. Moreover, IRScore(t,q) may be mathematically
calculated by Jaccard measure as follows:
IRScore ( t , q ) = Topic ( t ) Term ( q ) Topic ( t ) Term ( q ) [
Equation 9 ] ##EQU00006##
where Topic(t) and Term(q) may be the set of topics t in d, and the
set of terms in query q, respectively. In other words, the topical
similarity between d and q, divided by the total number of topics
in d (the last sum in equation 8) weights the IR score of each
document e that is cited by d (the first sum in equation 8).
Moreover, with respect to equation 9, the numerator is the total
number of distinct topics that are included in both topic t of
document d and query q (exemplified by the intersection operator
.andgate.). The denominator is the total number of distinct terms
that are included in topic t of document d or query q (exemplified
by the union operator .orgate.). In this way, the meta rank score
may be calculated using the out-cites from d.
[0073] After calculating the meta rank score using at least one of
in-cites from 504 and out-cites from 506, in block 608, citation
component 208 may send the meta rank for display. Alternatively, in
block 610, citation component 208 may send the meta rank to
activity score component 210 for further weighing. One of ordinary
skill will recognize that any blocks 602-610 may be omitted and/or
executed in any order.
[0074] FIG. 7 includes a flow diagram 700 illustrating an activity
score calculation for a document, such as a legal document. An
activity score may reflect the prominence of document d within its
subject matter community as a whole, such as a legal community.
Flow diagram 700 may be a process implemented by an activity score
component 210 to calculate the activity score of a document d. The
activity score component 210 may operate independently from a query
or other search.
[0075] In block 702, activity score component 210 may determine an
activity score vector for document d. The activity score vector may
be used to model the manner in which outside sources have treated,
commented on, or described document d. The activity score vector
may be constructed according to six parameters of document d. Each
parameter may correspond to a different outside source that has
treated document d. The six parameters may include case cites, U.S.
Supreme Court cites, law review articles, treatises, signal, and/or
treatment. The activity score vector as a whole, and each of the
parameters in particular, may provide an indication as to the
overall prominence of d in the subject matter community.
[0076] For example, parameter one of the activity score, i.e., case
cites, may relate to court cases in a particular subject matter
community that cite to d. The subject matter community may include
the legal community as a whole, and may not be limited to the
baseline search results. In some embodiments, the subject matter
community may be limited by a legal jurisdiction, such as a
particular state, or may include multiple jurisdictions. A number
of cases in the subject matter community that cite to d may be an
indicator of the prominence of d within the subject matter
community as a whole.
[0077] Parameter two of the exemplary activity score, i.e., U.S.
Supreme Court cites, may include U.S. Supreme Court cases that cite
to d. Because documents cited by the U.S. Supreme Court are
considered strong precedent, any U.S. Supreme Court cases that cite
d may be an indicator of the prominence of d.
[0078] Parameters three and four of the exemplary score vector,
i.e., law review articles and treatises that cite to d, may be
legal documents, which are not court cases, and that may indicate
the academic treatment of d. Accordingly, law review articles and
treatises that cite to d may also be an indicator of the prominence
of d within the subject matter community.
[0079] Parameters five and six of the activity score, i.e., signal
and treatment, may be retrieved from a Shepard's.RTM. database. The
signal may summarize the treatment of d within the legal community.
The treatment may include a number of positive and negative
treatments of d within the subject matter community.
[0080] More or fewer parameters may be used for calculating the
activity score. Moreover, the parameters disclosed herein are
exemplary only.
[0081] Next, in block 704, activity score component 210 may
calculate an individual activity score for each of the parameters
of the activity score. For example, individual activity scores may
be calculated for parameters one through four (case cites, U.S.
Supreme Court cites, law review articles, treatises), which
emphasis a number of documents, not limited to baseline documents,
that cite to d. Moreover, individual activity scores may also be
calculated for parameters five and six (i.e. signal and
treatment).
[0082] In block 706, activity score component 210 may adjust or
normalize at least one of the individual activity scores, for
example those corresponding to parameters one through four,
according to a probability. In block 708, activity score component
210 may calculate a total activity score by adding the adjusted or
normalized individual activity scores from parameters one through
six. In block 710, the activity score component 210 may adjust the
total activity score according to the age of d. In block 712, the
activity score component 210 may store the total activity score in
memory. One of ordinary skill will recognize that any blocks
702-712 may be omitted and/or executed in any order.
[0083] Specifically, there are several ways to individually score
parameters one through four as set forth in block 704. Parameters
one through four emphasize a number of documents in a subject
matter community as whole, not limited to baseline documents
related to a query, that cite to document d. For example, an
individual activity score for parameter one (representing case
cites), may simply equal a number of times that documents within
the subject matter community cite to d.
[0084] FIG. 8 illustrates such an example. FIG. 8 includes linear
graph 800 illustrating the relationship of number of case cites of
document d (on the x-axis) to its corresponding activity score (on
the y-axis). Thus, linear graph 8 illustrates one manner of
calculating an individual activity score for parameter one of d's
activity score. As shown, the individual activity score for
parameter one of d's activity score vector equals the number of
times that cases within the subject matter community cite to
document d.
[0085] However, linear graph 800 provides one way to calculate the
individual activity score of d for parameter one of the activity
score. In particular, linear graph 800 does not take into account
how often other documents in the subject matter community are cited
to. For example, with respect to parameter one which represents
case cites, legal document d.sub.1 on graph 800 is cited by 10
cases in the legal community, and legal document d.sub.2 is cited
by 20 cases in the legal community. Accordingly, for documents
d.sub.1 and d.sub.2, the difference in the individual activity
score (for parameter one) is 10. Moreover, legal document d.sub.3
is cited by 30 cases in the legal community. Accordingly, for
documents d.sub.2 and d.sub.3, the difference in the individual
activity score (for parameter one) is also 10.
[0086] However, the number of times that other documents are cited
to in the subject matter community should also be a factor in
determining activity scores for documents d.sub.1-d.sub.3. In this
way, a number of times that other documents in the subject matter
community are cited to may be used to adjust or normalize the
activity score of document d. For example, if there are 5 documents
in the subject matter community that are cited 10 times like
d.sub.1, and there are 100 documents in the subject matter
community that are cited to 20 times like d.sub.2, then the
difference in activity score between d.sub.1 and d.sub.2 should be
proportionately large. By contrast, if there are 110 other
documents in the subject matter community that are cited to 30
times like d.sub.3, then the difference in activity score between
d.sub.2 and d.sub.3 should be proportionately small. In other
words, the difference in individual activity score between d.sub.1
and d.sub.2 should be larger than the difference in individual
activity score between d.sub.2 and d.sub.3 in this example. In the
subject matter community, the difference between being cited 10
times and being cited 20 times is considerable, whereas the
difference between being cited 20 times and being cited 30 times is
not considerable. However, using the manner of calculating an
activity score in FIG. 8, the activity score difference between
d.sub.1 and d.sub.2 and between d.sub.2 and d.sub.3 are the same,
i.e. 10.
[0087] Moreover, since parameters two to four of the activity score
vector are conceptually similar to parameter one, a graph similar
to linear graph 800 may not provide the optimal way to calculate
the activity score for parameters two to four of d's activity
score.
[0088] One way of taking into account the other documents in the
subject matter community to determine parameter one of the activity
score (case cites to d), is to use a probability. For example, if
it is known that document d is cited 10 times in the legal
community, then instead of using 10 as a basis for the activity
score, the probability that d is cited 10 times may be used. Using
the probability enables other documents in the subject matter
community to be considered for the purposes of adjustment or
normalization. That is,
Pr(X=x) [Equation 10]
which is the probability that X (the number of times d is cited)
equals x, which is 10 in this example. This probability
distribution may not be optimal because the data set may change
dynamically such that additional cases are added to the subject
matter community that may cite to d. Therefore, alternatively, the
probability that d is cited 10 times or more may be calculated.
That is,
Pr(X.gtoreq.x) [Equation 11]
which is the probability that X (the number of times d is cited) is
greater than or equal to x, which is 10 in this example.
Probability distributions from Equations 10 and/or 11 may be used
to adjust or normalize the activity score of d with respect to the
other documents in the subject matter community.
[0089] With further reference to equation 11, multiple values may
be used for X (the number of times that d is cited), instead of
only 10 in the previous example. When multiple values of X are
used, a probability is calculated for each of the multiple values.
The following table of results was generated for a particular court
case in California. The table reflects the number of times that the
particular court case was cited in the jurisdiction of California,
as well as a corresponding probability.
TABLE-US-00004 TABLE 4 Sample results Total_cites Entry (x) Pr(X
>= x) 1 14179 0.071392 2 13867 0.075854 3 13279 0.080316 4 13043
0.084778 5 12790 0.08924 6 12717 0.093702 7 12672 0.098164 8 12014
0.102626 9 11242 0.107088 10 11204 0.11155 11 11149 0.116012 12
10427 0.120474 13 10055 0.124936 14 9980 0.129398 14 9730 0.13386
15 9371 0.138322 16 9211 0.142784 17 8967 0.147246 18 8961
0.151708
[0090] The first column (entry) of Table 4 serves to provide a row
number for the data values, for the purposes of reference. The
second column of Table 4 (total cites) illustrates multiple values
of X for the particular court case. In other words, the second
column of Table 4 illustrates potential values for a number of
times that court cases in the jurisdiction of California cite to
the particular court case. The corresponding values in column three
(Pr(X>=x), illustrate a corresponding probability for each of
the values in column 1.
[0091] For example, the first data entry (entry 1) illustrates that
the probability of the particular court case being cited more than
14,179 times is 0.071392 (7.1392%). Accordingly, it is relatively
unlikely that the particular court case is cited more than 14,179
times in California. By contrasted, the last data entry (line 18)
illustrates that the probability of the particular court case being
cited more than 8,961 times is 0.151708 (15.1708%). Accordingly, it
is somewhat likely that the particular court case is cited more
than 8,961 times in California.
[0092] Once a set of data points (such as the ones in Table 4) is
calculated according to equation 11, it becomes necessary to
calculate a formula that accurately models the set of data points.
Accordingly, a regression may be performed on the data set
according to the following:
Pr(X<x)=ax.sup..alpha. [Equation 11.1]
where x is a number of cases citing document d, and a and .alpha.
are learned from a regression method. Moreover, x may correspond to
the values from column 2 of table 4. For exemplary purposes only,
the regression function for the data from Table 4 may be calculated
to be:
Pr(X<x)=12643x.sup.-1.1598. [Equation 11.2]
[0093] FIG. 9 shows graph 900 that illustrates a relationship
between a probability of cases within a subject matter community
that are cited a variable number of times and its activity score.
Graph 900 includes an activity score value or a probability value
on the y-axis, and a number of cases cited on the x-axis.
Probability distribution curve 902 shows an exemplarily probability
distribution over a set of documents in the subject matter
community. Probability function 902 uses equation 11 to illustrate
the probability (y) that a given document is cited greater than or
equal to certain number of times (x), in the subject matter
community.
[0094] In this example, probability distribution curve 902 may be
based on the data points illustrated in Table 4. Moreover,
probability distribution curve 902 may follow equation 11.2, which
is the equation calculated (by performing a regression) to
represent the data points from Table 4.
[0095] For example, for a document d.sub.4 that is cited 320 times,
the Pr(X.gtoreq.320)=15, i.e. the probability that d.sub.4 is cited
more than 320 times is 15%. For a document d.sub.5 that is cited
620 times, the Pr(X.gtoreq.620)=7, i.e. the probability that
d.sub.5 is cited more than 620 times is 7%. For a document d.sub.6
that is cited 920 times, the Pr(X.gtoreq.920)=4, i.e. the
probability that d.sub.6 is cited more than 920 times is 4%. Thus,
probability distribution curve 902 takes into account the frequency
with which other documents in the legal community are cited to, and
therefore, can be used to formulate the activity score. However,
probability distribution curve 902 is downward sloping, such that
the probability (y) decreases as the number of cited cases (x)
increases. By contrast, the activity score should increase as the
number of cased cited increases, while still taking into account
other documents in the legal community.
[0096] Therefore, an activity score may be formulated as an inverse
of equation 11.1. Accordingly, an equation for the activity score
may be:
Score ( x ) = k ( a x - .alpha. + 1 ) p [ Equation 11.3 ]
##EQU00007##
in which k and p are constants decided by application needs,
p<1, and a and .alpha. are learned from a regression method (as
was the case with equation 11.1)
[0097] In FIG. 9, activity score curve 904 illustrates an
individual activity score from equation 11.3. Accordingly, while
probability distribution curve 902 decreases and converges to zero
as the number of times cited (x-axis) increases, the activity score
curve 904 is formulated to increase and converge at a maximum value
k>0 as the number of times cited (x-axis) increases. In this
example, activity score curve 904 converges to a value near 50.
[0098] Moreover, for a document d.sub.4 that is cited 320 times,
the individual activity score is 37. For a document d.sub.5 that is
cited 620 times, the individual activity score is 42. For a
document d.sub.6 that is cited 920 times, the individual activity
score is 45. In this example, the individual activity scores
increase to reflect the importance of additional citations, yet
increase at a diminishing rate to reflect the decreased likelihood
of documents being cited to a larger and larger number of
times.
[0099] FIGS. 8 and 9 are illustrative of parameter one of the
activity score (case cites). Similar concepts, including equations
11.1 and 11.3, may be used to calculate parameters two through four
(U.S. Supreme Court cites, law review articles, and treatises). As
discussed, individual activity scores for parameters one though
four may be added to individual activity scores for parameters five
and six (signal and treatment), to calculate a total activity score
for a document d.
[0100] Accordingly, the total activity score for d may be
represented as,
ActivityScore ( d ) = W ( age ) i = 1 4 w i Score ( x i ) + Score (
x 5 ) + Score ( x 6 ) [ Equation 12 ] ##EQU00008##
where w.sub.i is the weight or score distribution and x.sub.i is
the total number of citing references for activity score vector
dimension i. For example, i=1 corresponds to the first dimension in
the activity score vector, which is citing cases. Moreover, i=2
corresponds to the second factor in the activity score vector,
which is U.S. Supreme Court Cases, and so forth.
[0101] The first term in Equation 12, W(age) may increase the
activity score of legal documents that are younger or more recent.
For example, a 30 year old legal document which is cited to 5 times
receives a lower activity score than a 1 year old legal document
which is also cited to 5 times. The factor W(age) may be calculated
according to the following:
W ( age ) = k log 2 ( age ) m + 1 [ Equation 13 ] ##EQU00009##
where k and m are constants which may be dynamically decided based
on application needs. In this way, younger cases may be promoted to
have higher activity scores.
[0102] The next term in Equation 12 (the summation), adds
individual activity scores for each of dimensions one through four
in the activity score vector. The individual activity score for
each of dimensions one through four may use an inverse of a
probability distribution to take into account the frequency with
which other documents in the legal community are cited to, such as
exemplified in FIG. 9.
[0103] The next term in Equation 12 is an activity score of a
signal associated with d (corresponding to dimension five of the
activity score vector). This activity score may be decided
semantically by application needs.
[0104] The next term in Equation 12 is activity score of a
treatment associated with d (corresponding to dimension six of the
activity score vector). The activity score of the treatment may be
defined according to a number of positive and/or negative
treatments of d. The activity score of the treatment may be
calculated according to the following,
Score ( x 6 ) = k ( a ( x + 1 ) - .alpha. * 100 + 1 ) p where , [
Equation 14 ] x = ( P - N ) 2 P + N [ Equation 15 ]
##EQU00010##
where k is positive when P>N, and k is negative when P<N, and
k=0 when P=N. Moreover, variables a, a, and p may be learned from a
regression method (as was the case with equation 11.1).
[0105] As set forth above, the terms in Equation 12 have been
described. These terms may be used in calculating a total activity
score for d. The total activity score for d may be used to weight a
meta rank score for d. Alternatively, the total activity score of d
may independent from the meta rank score of d, and may be displayed
separately.
[0106] FIG. 10 illustrates an exemplary results page 1000, which
may be generated in response to a query of a legal database, such
as database 114, in accordance with disclosed embodiments. Results
page 1000 includes a ranked list of documents, as well as
corresponding data about the ranking for each document. Each row is
an entry and each entry corresponds to a document. For example,
entry 1001 corresponds to a stored document determined to be
responsive to a query. The same is true for the remaining entries
in page 1000. The order of ranking in results page 1000 may be
determined according to an IRScore, meta rank score, and/or
activity score, among other factors, for each document. Moreover,
results page 1000 may include a series of columns 1002-1026 that
describe the ranked list of documents.
[0107] Column 1002 may include an original rank for each displayed
document, before the IR Score, meta rank score, and/or activity
score may be used to rank the displayed documents. For example, the
original rank corresponding to entry 1001, shown in column 1002, is
"7." Accordingly, column 1002 may be used to illustrate a change in
ranking between prior ranking systems and systems consistent with
disclosed embodiments.
[0108] Column 1004 may include a number of documents within a
baseline set that are cited to by each displayed document
(out-cites). For example, the number of documents that are cited to
by the document corresponding to entry 1001, as shown in column
1004, is "0." Moreover, column 1006 may include a number of
documents within the baseline set that cite each displayed document
(in-cites). For example, the number of documents that cite to by
the document corresponding to entry 1001, as shown in column 1006,
is "16." Out-cites and in-cites were previously discussed with
respect to FIGS. 5 and 6.
[0109] Column 1008 may include an IR Score, which illustrates a
similarity between metadata of the displayed documents and the
query. For example, the IR score corresponding to entry 1001, as
shown in column 1008, is "17.0." IR Score from column 1008 may be
calculated according to Equation 3 (which incorporates Equations 1
and 2). Column 1010 may include a first meta rank score for each
displayed document calculated according to in-cites, out-cites, and
the IR Score. For example, the first meta rank score corresponding
to entry 1001, as shown in column 1010, is "44.0." First meta rank
score from column 1010 may be calculated according to a combination
of equations 4 and 7 (e.g., by adding the results of Equations 4
and 7).
[0110] Moreover, column 1012 may include a second meta rank score
for each displayed legal document, also calculated according to
in-cites, out-cites, and the IR Score. For example, the second meta
rank score corresponding to entry 1001, as shown in column 1012, is
"66.0." Second meta rank score from column 1012 may be calculated
according to a combination of equations 5 (which incorporates
Equation 6) and 8 (which incorporates Equation 9). For example, the
results of Equations 5 and 8 may be added to arrive at the second
meta rank score.
[0111] First meta rank score and second meta rank score may both
correspond to a meta rank score. First meta rank score and second
meta rank score may each be the result of different methods by
which to calculate a meta rank score according to an IR Score,
in-cite, and/or out-cites.
[0112] Column 1014 may include a signal (e.g. a Shepard's.RTM.
signal) for each of the displayed documents. For example, the
signal corresponding to entry 1001, as shown in column 1014, is an
upward arrow signifying positive treatment. Column 1016 may include
an activity score for each of the displayed legal documents. For
example, the activity score corresponding to entry 1001, as shown
in column 1016, is "73." The activity score from column 1014 may be
calculated according to equation 12, which incorporates equations
13-15.
[0113] Column 1018 may include a first new ranking by combining the
first meta rank score with the activity score. For example, the
first new ranking corresponding to entry 1001, as shown in column
1018, is "58." Column 1020 may include a second new ranking by
combining the second meta rank score with the activity score. For
example, the second new ranking corresponding to entry 1001, as
shown in column 1020, is "80." First new ranking and second new
ranking may be the results of different methods to rank the
displayed legal documents using the IR Score, meta rank score,
and/or activity score.
[0114] Column 1022 may include case names or other identifiers for
the displayed documents. For example, the case name corresponding
to entry 1001, as shown in column 1022, is "Case 2." Column 1024
may include citations for each of the displayed documents. For
example, the citation corresponding to entry 1001, as shown in
column 1024, is "Cite 2." Column 1026 may include a date associated
with each of the displayed documents, for example, a date decided.
For example, the date corresponding to entry 1001, as shown in
column 1026, is "Date 2."
[0115] The foregoing description has been presented for purposes of
illustration. It is not exhaustive and is not limiting to the
precise forms or embodiments disclosed. Modifications and
adaptations will be apparent to those skilled in the art from
consideration of the specification and practice of the disclosed
embodiments. For example, the described implementations include
software, but systems and methods consistent with the disclosed
embodiments be implemented as a combination of hardware and
software or in hardware alone. Examples of hardware include
computing or processing systems, including personal computers,
servers, laptops, mainframes, micro-processors and the like.
Additionally, although aspects of the disclosed embodiments are
described as being stored in memory, one skilled in the art will
appreciate that these aspects can also be stored on other types of
computer-readable media, such as secondary storage devices, for
example, hard disks, floppy disks, or CD-ROM, or other forms of RAM
or ROM, USB media, DVD, or other optical drive media.
[0116] Computer programs based on the written description and
disclosed methods are within the skill of an experienced developer.
The various programs or program modules can be created using any of
the techniques known to one skilled in the art or can be designed
in connection with existing software. For example, program sections
or program modules can be designed in or by means of .Net
Framework, .Net Compact Framework (and related languages, such as
Visual Basic, C, etc.), Java, C++, HTML, HTML/AJAX combinations,
XML, or HTML with included Java applets. One or more of such
software sections or modules can be integrated into a computer
system or existing e-mail or browser software.
[0117] Moreover, while illustrative embodiments have been described
herein, the scope of any and all embodiments having equivalent
elements, modifications, omissions, combinations (e.g., of aspects
across various embodiments), adaptations and/or alterations as
would be appreciated by those in the art based on the present
disclosure. The limitations in the claims are to be interpreted
broadly based on the language employed in the claims and not
limited to examples described in the present specification or
during the prosecution of the application, which examples are to be
construed as non-exclusive. Further, the blocks of the disclosed
routines may be modified in any manner, including by reordering
blocks and/or inserting or deleting blocks. It is intended,
therefore, that the specification and examples be considered as
exemplary only, with a true scope and spirit being indicated by the
following claims and their full scope of equivalents.
[0118] Other embodiments will be apparent to those skilled in the
art from consideration of the specification and practice of the
embodiments disclosed herein. It is intended that the specification
and examples be considered as exemplary only, with a true scope and
spirit being indicated by the following claims.
* * * * *