U.S. patent number 6,718,368 [Application Number 09/325,111] was granted by the patent office on 2004-04-06 for system and method for content-sensitive automatic reply message generation for text-based asynchronous communications.
This patent grant is currently assigned to General Interactive, Inc.. Invention is credited to V. A. Shiva Ayyadurai.
United States Patent |
6,718,368 |
Ayyadurai |
April 6, 2004 |
**Please see images for:
( Certificate of Correction ) ** |
System and method for content-sensitive automatic reply message
generation for text-based asynchronous communications
Abstract
A process and system is disclosed for automatically and
adaptively retrieving information from a database and transmitting
reply messages based on the content of a received message, such as
an e-mail, CGI form, or facsimile document. In one step of the
process, an enhanced e-mail editor user interface is employed to
display the original message, the system's adaptive analysis
results, and the automatically generated reply message so that a
user may review the reply, modify the analysis results, and modify
the reply message. The process then learns the changes indicated,
and updates the adaptive analysis steps which analyze the received
messages. In so doing, future reply messages are increasingly more
accurate and correct. After the is adaptive algorithms have been
sufficiently trained through the user intervention and correction
step, the process and system disclosed is capable of fully
autonomous reply generation and transmission.
Inventors: |
Ayyadurai; V. A. Shiva
(Belmont, MA) |
Assignee: |
General Interactive, Inc.
(Cambridge, MA)
|
Family
ID: |
32030483 |
Appl.
No.: |
09/325,111 |
Filed: |
June 3, 1999 |
Related U.S. Patent Documents
|
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
Issue Date |
|
|
323312 |
Jun 1, 1999 |
|
|
|
|
Current U.S.
Class: |
709/206; 704/1;
706/12; 706/45; 706/47; 707/999.001; 709/204; 709/207; 715/205 |
Current CPC
Class: |
H04L
51/12 (20130101); Y10S 707/99931 (20130101); Y10S
707/99936 (20130101) |
Current International
Class: |
G06F
15/16 (20060101); G06F 015/16 () |
Field of
Search: |
;709/206,207,204
;345/733 ;370/260 ;379/93.01 ;707/1,6 ;706/12,45,47 ;704/1
;715/500.1,531 |
References Cited
[Referenced By]
U.S. Patent Documents
Primary Examiner: Coulter; Kenneth R.
Assistant Examiner: Nguyen; Hai V.
Attorney, Agent or Firm: Burns Doane Swecker & Mathis,
LLP
Parent Case Text
CROSS-REFERENCE TO RELATED APPLICATIONS
CLAIMING BENEFIT UNDER 35 U.S.C. 120
This application is a continuation-in-part of U.S. application Ser.
No. 09/323,312, filed on Jun. 1, 1999, by V. A. Shiva
Ayyadurai.
FEDERALLY SPONSORED RESEARCH AND DEVELOPMENT STATEMENT
This invention was not developed in conjunction with any Federally
sponsored contract.
MICROFICHE APPENDIX
Not applicable.
Claims
What is claimed is:
1. A system for automatic generation and reply to messages in an
asynchronous communications system, comprising: a message reception
means for inputting a first electronic text message into the
system; a message natural language filtering and modeling means
receiving tile text message from the message reception means, said
filtering and modeling means generating a plurality of assigned
message characteristics according to natural language analysis of a
body portion of the text message, said assigned characteristics
indicating factors, including one or more of customer attitude,
customer profile, customer education level, and issues presented,
upon which a reply will depend, said natural language analysis
comprising performing feature extraction to produce a plurality of
output signals relating to one or more of keyword frequencies, word
co-occurrence statistics, a dimensionally-reduced representation of
the keyword frequencies, phoneme frequencies, and structural
pattern statistics for any of sentences, paragraphs, and pages; and
a message reply composition means receiving said assigned message
characteristics, and having an interface for communicating to a
reply phrase library database using database queries, and having a
proposed text-based reply electronic message output, said message
reply composition means performing the following steps to create a
grammatically complete, comprehensible and relevant reply
electronic message output: retrieving at least one phrase from said
phrase library database for each assigned message characteristic
received from said filtering and modeling means; combining said
retrieved phrases into a grammatically complete, comprehensible and
relevant reply electronic text message; and outputting the reply
electronic text message.
2. The system for automatic generation and reply to messages in an
asynchronous communications system of claim 1, further comprising a
user interface means which receives said first and second
electronic text messages, said tags, and presents them via a
computer display.
3. The system for automatic generation and reply to messages in an
asynchronous communications system of claim 2, wherein said user
interface means further comprises a user input means for allowing a
user to change the text within said second electronic text
message.
4. The system for automatic generation and reply to messages in an
asynchronous communications system of claim 2, wherein said user
interface means further comprises a user input means for allowing a
user to change the tag values.
5. The system for automatic generation and reply to messages in an
asynchronous communications system of claim 4, further conspiring a
learning processing means for receiving said user changes to said
assigned characteristics via said user interface means, and for
automatically updating in real time adaptable parameters within
said filtering and modeling means without operator
intervention.
6. The system for automatic generation and reply to messages in an
asynchronous communications system of claim 1 further comprising
and electronic text message transmission means for transmitting
said second electronic text message to another destination.
Description
TECHNICAL FIELD OF THE INVENTION
This invention pertains to the arts of automatic analysis,
classification, characterization, routing and response to
text-based messages in electronic asynchronous messaging systems,
and especially to electronic mail and facsimile systems.
INCORPORATION BY REFERENCE
The above referenced application Ser. No. 09/323,312, filed on Jun.
1, 1999, by V. A Shiva Ayyadurai, is incorporated herein by
reference in its entirety, including drawings, and hereby is made a
part of this application.
BACKGROUND OF THE INVENTION
Electronic mail and facsimile ("fax") messaging have become
critical tools of everyday personal and business life. Most
corporations, government agencies, organizations, and institutions
have established fax numbers and e-mail addresses for a wide
variety of contact purposes, including requesting information such
as literature and office locations from the entity, requesting
investment information, requesting service on or technical support
for a product, reporting a product problem or failure, submitting
suggestions for products and service improvements, submitting
complimentary comments, and in some cases, carrying on dialogues
with personalities and celebrities associated with the entity. Fax
and email messaging have converged in electronic form, as messages
originating in the form of fax are commonly captured by computers
with fax/modem interfaces and optically converted to text files,
and as many services offer low cost fax message delivery via
e-mail-based interfaces.
Underlying the tremendous proliferation of fax and email are
several factors, including wide-spread availability of inexpensive
e-mail clients such as personal computers, and inexpensive fax
machines, and the development of common standards for exchange of
electronic text messages between computers, including RFC821 Simple
Mail Transfer Protocol ("SMTP") from the Internet Network
Information Center, and Recommendation X.400 from the International
Telecommunications Union ("ITU").
Consequently, corporations, government agencies, and other entities
which successfully promote the availability of their fax telephone
numbers and e-mail addresses can receive thousands to
tens-of-thousands of messages per day.
Traditionally, all of the electronic messages are received in a
general repository, or "mailbox", and reviewed by human agents for
their content, intent, at which time a determination of the correct
disposition of the email is made. This may involve sending the
author a standard reply, and/or copying or forwarding the e-mail to
one or more divisions, departments, or individuals within the
organization for further handling. In the later case where multiple
parties must be consulted, the consolidation of replies from all of
the parties can be cumbersome and overwhelming, given the volume of
messages to be handled.
For example, assume a company receives five thousand messages per
day. Further assume that on average each one of those messages
contains issues or requests that involve an average of 3
departments or individual. The original message must be read once
by the reviewing agent who forwards the message to the 3
individuals or departments. The receiving departments may read the
forwarded message one to three times per department before it
reaches the person who can respond. In total, 5,000 received
e-mails may result in up to 20,000 to 50,000 reviews of the
messages within the company. In many cases, the final recipient may
need to instigate a short dialogue over several message exchanges
with the author to ascertain exactly what the author needs or how
the author can be serviced. Consequently, a daily volume of 5,000
new messages may reasonably result in a total network volume and
work load of tens-of-thousands to a hundred-thousand messages per
day.
To deal with this volume of messages and to attempt to provide a
timely response to these messages, some systems have been developed
which allow the human reviewer to combine and create or select a
response from libraries of standard responses. For example, FIG. 1
shows a prior art method wherein a customer for a company creates
an initial message using an e-mail viewer and editor (1), and
transmits (2) that message into an e-mail routing system (3). The
corporation will receive (4) the message and an agent will read it
using another e-mail viewer and editor (5). As the agent manually
analyzes the message content (7), he or she may search (8) a file
or database (15) which contains text replies to the most commonly
asked questions. The agent then selects one or more of the text
replies, uses a "cut and paste" user interface function (11) to
compose the basis for an e-mail reply, composes the remainder of
the reply (that which is not drawn from the database or not
provided for by the database), and sends (12) that message back to
the customer via the e-mail routing system (3) using the e-mail
view and editor (5) send or reply function. Finally, the customer
may receive his or her reply by retrieving (14) the message from
the e-mail routing system (3), and reading it with his or her
e-mail viewer (1). This "cut and paste" method can reduce the
amount of labor required to respond to only a low percentage of the
messages received as the database will contain answers only to
commonly asked questions. Complicated questions and follow-up
messages may not be able to be handled by this method, so
ultimately, some percentage of these messages are not handled with
this method and are transferred or forwarded to one or more
recipients for manual disposition.
For example, a typical two-issue email is given in TABLE 1.
TABLE 1 Typical Two-Issue Email To: info@xyzcorp.com FROM:
john_customer@an_isp.net SUBJECT: Model 999 of your product and
your service centers MESSAGE: Please send me the technical specs
for your Model 999 widget, especially the standard battery life.
And, if it breaks, where is the closest service center to my home
town, Smithville, or do I have to ship it back to you for repair?
Thanks, John.
Using the method of FIG. 1, an agent would search the sample
responses for the response to the first issue, the request for
technical specifications, and would cut and paste the following
text into a new or reply e-mail, as shown in TABLE 2.
TABLE 2 First Cut and Paste for Reply To: john_customer@an_isp.net
FROM: info@xyzcorp.com SUBJECT: Re: Model 999 of your product and
your service centers MESSAGE: "The Model 999 typically operates for
up to 4 hours between battery charges".
Then, the agent would search for a response to the second issue,
the request for the nearest service center, and would paste text
into the message as shown in TABLE 3.
TABLE 3 First Cut and Paste for Reply To: john_customer@an_isp.net
FROM: info@xyzcorp.com SUBJECT: Re: Model 999 of your product and
your service centers MESSAGE: Dear John, The Model 999 typically
operates for up to 4 hours between battery charges. "Bob's Hardware
Store, 19 Main Street, Smithville, Telephone: 1-800-999-8888."
Finally, the agent would complete the composition by adding
connective text to complete a comprehendible message, as shown in
TABLE 4.
TABLE 4 Final Edit for Reply To: john_customer@an_isp.net FROM:
info@xyzcorp.com SUBJECT: Re: Model 999 of your product and your
service centers MESSAGE: Dear John, Thank you for your recent
message. To answer your first question, the Model 999 typically
operates for up to 4 hours between battery charges. As for your
second question, your closest authorized repair facility is Bob's
Hardware Store, 19 Main Street, Smithville, Telephone:
1-800-999-8888. I hope this has provided the answers you need, but
if not, please feel free to contact us again. Best Regards, Suzy,
Customer Service Agent 19.
If some of the issues presented by the initial message from the
customer are not provided for by the sample reply database, the
agent may need to manually forward the message to one or more
specialists or departments, finally merging their replies with
other sample replies.
FIG. 2 shows a slightly more advanced method in the prior art,
whereby the e-mail is received from the customer to an agent's
e-mail viewer and editor (5) in the same manner as described above
and shown in FIG. 1. However, a database of entire message
responses (22) is available for the agent to simply select to be
sent to the customer. The agent manually browses or searches the
library of standard replies (22) using an index (21), and the reply
message or messages is automatically sent based on the agent's
selection. Using this method, the time required to cut and paste a
new reply and to complete the message is eliminated as each reply
contained in the library is a complete reply in and of itself, and
the issues not provided for in the library may be handled
separately. For example, the two issues presented in the message of
TABLE 1 could be handled with two separate messages, as shown in
TABLE 5 and TABLE 6, both selected from the library of standard
complete replies.
TABLE 5 Reply to First Issue To: john_customer@an_isp.net FROM:
info@xyzcorp.com SUBJECT: Model 999 Specifications MESSAGE: Thank
you for your recent message. Here are the specifications for the
Model 999: Weight: 10 lbs. Safety: UL-approved Color: Safety Yellow
Battery Life: Standard 4 hour operation between charges, optional
extended life battery provides up to 10 hours operation. We
appreciate your interest in our products, Sincerely, the XYZ
Customer Response Center.
TABLE 6 Reply for Second Issue To: john_customer@an_isp.net FROM:
info@xyzcorp.com SUBJECT: Service Centers MESSAGE: Thank you for
your recent message. A current listing of service centers across
the country can be found on our website at:
www.xyzcorp.com/service_locations.htm
While this second method reduces the labor to generate the reply,
it has several disadvantages. The first of which is the fact that
the customer received multiple replies for a single message. The
second is that the messages miss the "human touch" and are
obviously "canned" because they contain no information which is
personal to the customer, such as addressing him by name. Also, the
second reply does not actually answer his question, but refers him
to the source for his answer. Finally, for any issues presented in
the message for which no standard reply is available in the
library, a third message reply may be sent to the customer by a
specialist.
Therefore, there exists a need in the art for an automated reply
system and method which intelligently composes a preliminary
response to a message which may contain one or more issues or
requests.
Further, there exists a need in the art for this automated reply
system to use conventional technology and techniques which find
practical application to the analysis of free-form written
speech.
Additionally, there exists a need in the art for this system to
allow a human agent to review, correct, modify, or cancel a
proposed reply.
Still further, there exists a need for this system to be capable of
autonomous message composition and replying without need for human
agent intervention.
BRIEF DESCRIPTION OF THE DRAWINGS
For a more complete understanding of the invention, the following
disclosure can be taken in conjunction with the presented
figures.
FIG. 1 shows one method of reviewing and composing replies to
electronic messages, and
FIG. 2 shows a semi-automated method, both of which are well known
within the art.
FIG. 3 depicts the system and method of the invention as disclosed
herein.
FIG. 4 shows a view of the user interface allowing a human agent to
review the original email, the proposed reply, and the tags.
SUMMARY OF THE INVENTION
An object of the invention disclosed herein is to provide an
automated system composing reply messages to electronic text-based
messages within an asynchronous communications system.
The system and method disclosed intelligently composes a
preliminary response to a message which may contain one or more
issues or requests. The system is comprised of conventional
technology and techniques which are practical applications of
analysis of free-form written speech.
The preliminary response is presented to a user interface, which
allows a human agent to review, correct, modify, or cancel a
proposed reply.
In an optimal embodiment or mode of operation, this system is
capable of autonomous message composition and replying without need
for human agent intervention.
DETAILED DESCRIPTION OF THE INVENTION
The following disclosure, when taken in conjunction with the
presented figures, sets forth the invention which meets the objects
of the invention set forth in the SUMMARY OF THE INVENTION.
Turning to FIG. 3, a customer may compose and send a message using
an e-mail viewer and editor (1), and may send (2) that message to
the system using an e-mail routing system (3), which are well known
within the art. Preferably, this is an SMTP e-mail servers, but can
also be a proprietary e-mail system such as IBM's Lotus Notes
system. Alternatively, the customer's message may originate as a
written document sent using a facsimile machine, which is then
subjected to optical character recognition ("OCR") and converted to
a text-based electronic message.
The message is then received (4) into a filter and modeler (30).
The reception means may be via file transfer over a network, such
as the Internet or a Local Area Network, or may be by interfacing
to a database in which the electronic messages are stored, such as
an ODBC interface.
The filter and modeler (30) performs language analysis and
characterization of the content of the text message, as described
in the parent patent application, yielding a tagged message (31)
which indicates the attitude of the customer (positive, negative,
or neutral), the issue(s) or request(s) presented in the message,
the profile of the customer and an estimate of his education level.
The techniques employed by the filter and modeler (30) are more
fully described in the aforementioned parent application. They
include feature extraction methods, which can be any one or
multiple methods of pattern recognition, such as keyword analysis,
morphology, natural language processing, thesauri, co-occurrence
statistics, syllabic analysis and word analysis. The feature
extraction methods result in several output signals, including
keyword frequencies, co-occurrence statistics, a
dimensionally-reduced representation of the keyword frequencies,
phoneme frequencies, structural pattern statistics for sentences,
paragraphs, and pages, estimated education level of the author
based on word choices and complexity of sentence structure, and
customer type.
This tagged message (31) is received by an automatic reply composer
(91) using the same transfer mechanisms described for the filter
modeler input (4). Based upon the tags, the automatic reply
generator (91) retrieves from a library of phrases (90) partial
replies or reply phrases (92) for each issue and request presented,
introductory phrases based on the attitude of the customer, and may
further re-select these replies based on educational level of the
customer. For example, TABLE 7 shows a simple, two-issue message
received from a customer.
TABLE 7 Example Initial Message To: customer_service@big_corp.com
FROM: unhappy_customer@an_isp.net SUBJECT: Please help! MESSAGE: I
have been trying to find a service location for my Model 999, which
has now been broken for several months. I just bought it last year,
and never expected it to break so soon. I live in Smithville. Also,
I never received my rebate in the mail. Please tell me when I can
expect that to arrive, too.
Based on the filter and modeler tag outputs, the phrases "please
help!", "for several months", "never expected", and "never"
indicate a general negative disposition of the customer. Further,
the filter identifies two issues, namely a request for the nearest
service location and a request for status on the rebate. So, a
preliminary message would be composed by searching the library and
concatenating text phrases into a whole message, such as the
process shown in TABLE 8.
TABLE 8 Search and Retrieval Process Steps intro_text =
Get_intro_phrase(negative) issue(1) = (service_locations,
smithville) issue(2) = (corp_policies, rebate) Reply_text =
intro_text For issue_index m = 1 to n: issue_text(m) =
Get_issue_reply(issue(1)) Reply_text = Reply_text + issue_text(m)
Next issue Reply_text = Reply_text + "If you have further
questions, please feel free to contact us again, or call us at
1-800-XYZ-CORP"
An example database of replies is given in TABLE 9, in a
comma-separated variable ("CSV") format, which is commonly known
within the art of database technologies, in which each carriage
return character <CR> signifies the end of a relational
record, and each relational record is made up of several fields
separated by a comma.
TABLE 9 Example Reply Phrase CSV Database Tag,Parm1,Parm2,Text
Reply Attitude,Negative,"We apologize for your
difficulties."<CR> Attitude,Neutral,"Thank you for your
message."<CR> Attitude,Positive,"We are pleased to hear from
you."<CR> Issue,corp_policies_refunds,"We do not offer
refunds."<CR> Issue,corp_policies_rebates,"All rebates will
be paid after the rebate period expiration date, August 15,
2001."<CR> Issue,corp_policies,hiring,"We are an equal
opportunity employer. Your message has been forwarded to our
staffing department."<CR> Issue,corp_policies,donations,"We
regularly support a number of charitable organizations. You message
has been forwarded to our community relations
department."<CR> Issue,service_centers smithville,"Your
nearest service center is Bob's Hardware, 19 Main Street,
Smithville."<CR> Issue,service_centers chicago"Your closest
service center is 1419 Enterprise Blvd., Schaumberg,
Il."<CR>
Using this example database and the pseudocode of TABLE 8, a reply
would be composed of the text reading:
TABLE 10 Concatenated Text Reply "We apologize for your
difficulties. Your nearest service center is Bob's Hardware, 19
Main Street, Smithville. All rebates will be paid after the rebate
period expiration date, August 15, 2001. If you have further
questions, please feel free to contact us again, or call us at
1-800-XYZ-CORP."
This proposed message is presented (32) to a user via an enhanced
user interface (33), with the tags and the original message, as
shown in FIG. 4. The user is able to view the original message (40)
preferably in a window with the typical left/right and up/down
navigation bars, similar to standard e-mail viewers and editors. A
second window for the proposed reply (41) allows the user to
compare the reply to the original message. A mouse-type cursor (42)
is preferably provided, which allows the user to select windows,
toggle options and the like, as is common with computer display
graphical user interfaces. A control panel (43) is provided in
which the user can see the associated tags, and which has at least
two options: APPROVE/SEND and is CHANGE. If the user wishes to
accept and send the message as proposed, he or she simply "clicks"
on the APPROVE button (45). If the system has incorrectly generated
a tag, such as the attitude, the user can click on the CHANGE
button (46), and then click on the tag to change it.
In either case, the user's selections are received (36) by the
learning process (37), as shown in FIG. 3. The learning process
updates (38) the adaptive reasoning and analysis logic of the
filter and modeler (30) such that future tag generation is more
precise.
When the message is ready to be sent, the user simply clicks on the
APPROVE button, and the message is transmitted (35) to the customer
via the e-mail routing system (3) such that he or she can receive
it (14) and view it using his or her e-mail viewer (1).
In an enhanced embodiment of the invention, the proposed replies
are automatically composed and transmitted (39) to the customer
without the step of manual review. A process of random or
statistical selection of the messages by a user can be incorporated
into the automatic reply composer such that a user can review and
correct proposed replies on a sample basis.
While the invention has been set forth with particular details to
the preferred embodiments, those who are skilled in the art will
recognize that a number of deviations and changes to the system and
method can be made without departing from the spirit and scope of
the invention.
* * * * *
References