U.S. patent application number 11/106215 was filed with the patent office on 2006-10-19 for system and method for expedited resolution of inquiries for an automated agent.
Invention is credited to Subrata K. Das, Ea-Ee Jan.
Application Number | 20060233344 11/106215 |
Document ID | / |
Family ID | 37108477 |
Filed Date | 2006-10-19 |
United States Patent
Application |
20060233344 |
Kind Code |
A1 |
Das; Subrata K. ; et
al. |
October 19, 2006 |
System and method for expedited resolution of inquiries for an
automated agent
Abstract
An automated agent system and method includes an action
classifier which scores a representation of a portion of a user
inquiry based on a knowledge bank. A comparing device compares a
running score of the representation to a threshold to determine if
enough information has been collected to respond early to the
inquiry. An action mechanism provides an early response to the user
prior to the user completing the inquiry if the threshold has been
exceeded.
Inventors: |
Das; Subrata K.; (Granite
Springs, NY) ; Jan; Ea-Ee; (Ardsley, NY) |
Correspondence
Address: |
KEUSEY, TUTUNJIAN & BITETTO, P.C.
20 CROSSWAYS PARK NORTH, SUITE 210
WOOBURY
NY
11797
US
|
Family ID: |
37108477 |
Appl. No.: |
11/106215 |
Filed: |
April 14, 2005 |
Current U.S.
Class: |
379/265.02 |
Current CPC
Class: |
H04M 2201/40 20130101;
H04M 3/527 20130101 |
Class at
Publication: |
379/265.02 |
International
Class: |
H04M 3/00 20060101
H04M003/00 |
Claims
1. An automated agent system, comprising: an action classifier,
which scores a representation of a portion of a user inquiry based
on a knowledge bank; a comparing device which compares a running
score of the representation to a threshold to determine if enough
information has been collected to respond early to the inquiry; and
an action mechanism which provides an early response to the user
prior to the user completing the inquiry if the threshold has been
exceeded.
2. The system as recited in claim 1, wherein the knowledge bank
includes information based on a history of previous inquiries.
3. The system as recited in claim 1, further comprising a speech
recognizer to generate the representation.
4. The system as recited in claim 1, wherein the action mechanism
includes a barge-in module, which enables the system to barge-in on
a user's inquiry prior to its completion.
5. The system as recited in claim 1, further comprising a dialog
manager, which responds the user's inquiry.
6. The system as recited in claim 1, further comprising a module to
detect an end of speech for the user's inquiry.
7. The system as recited in claim 6, wherein the end of speech
module generates a final score, which is compared to a final
threshold, such that if the final score exceeds the threshold
action is made on the inquiry in a no barge-in mode, otherwise the
user is prompted to repeat the inquiry.
8. The system as recited in claim 1, wherein the running score is
continuously monitored by the comparing device to ensure an
earliest response to the user's inquiry.
9. An automated call agent system, comprising: a speech recognizer,
which receives speech information from a caller and decodes the
speech information into a representation of the speech information
as the speech information is being received; an action classifier,
which scores portions of the representation to the inquiry based on
a knowledge bank; a comparing device which compares a running score
of the portions of the representation to a threshold to determine
if enough information has been collected to barge-in on the user's
inquiry to respond early to the inquiry; and an action mechanism
which barges-in on the inquiry and provides an early response to
the user prior to the user completing the inquiry, if the threshold
has been exceeded.
10. The system as recited in claim 9, wherein the knowledge bank
includes information based on a history of previous inquiries.
11. The system as recited in claim 9, further comprising a dialog
manager, which responds the user's inquiry when the action
mechanism barges-in.
12. The system as recited in claim 9, further comprising a module
to detect an end of speech for the user's inquiry.
13. The system as recited in claim 12, wherein the end of speech
module generates a final score, which is compared to a final
threshold, such that if the final score exceeds the threshold
action is made on the inquiry in a no barge-in mode, otherwise the
user is prompted to repeat the inquiry.
14. The system as recited in claim 9, wherein the running score is
continuously monitored by the comparing device to ensure an
earliest response to the user's inquiry.
15. A method for responding early to a user inquiry in an automated
agent system, comprising the steps of: scoring a representation of
a portion of a user inquiry based on a knowledge bank; comparing a
running score of the representation to a threshold to determine if
enough information has been collected to respond early to the
inquiry; and responding to the user prior to the user completing
the inquiry.
16. The method as recited in claim 15, wherein the knowledge bank
includes information based on a history of previous inquiries, and
further comprising updating the knowledge bank after an action is
taken.
17. The method as recited in claim 15, further comprising
recognizing speech to generate the representation.
18. The method as recited in claim 15, wherein responding includes
barging in on a user's inquiry prior to its completion.
19. The method as recited in claim 15, further comprising detecting
an end of speech for the user's inquiry.
20. The method as recited in claim 19, wherein detecting the end of
speech includes generating a final score, which is compared to a
final threshold, such that if the final score exceeds the threshold
action is made on the inquiry in a no barge-in mode, otherwise the
user is prompted to repeat the inquiry.
21. The method as recited in claim 15, wherein comparing includes
continuously monitoring the running score to ensure an earliest
response to the user's inquiry.
22. A program storage device readable by machine, tangibly
embodying a program of instructions executable by the machine to
perform method steps for responding early to a user inquiry in an
automated agent system, as recited in claim 15.
Description
BACKGROUND
[0001] 1. Technical Field
[0002] The present invention relates to an automated agent system
and method and more particularly to a system and method, which
expedites resolution of inquiries in an automated agent.
[0003] 2. Description of the Related Art
[0004] A call center provides rapid resolution of a caller's
inquiry. In a traditional call center manned by human agents, both
the caller and the agent tend to barge in on each other during the
course of a normal conversation; this may serve to speed up the
process of problem resolution.
[0005] For example, the following exchange illustrates a call made
to a call center. The caller had just installed a software package
on his computer following the agent's instructions and inquired,
"So I need to start the . . . ", but before the caller could finish
the sentence, the agent barged in with the answer, "Yeah, you need
to take out the CD first and then restart . . . " and so on, since
the agent could guess what the caller was going to say and did not
want to wait for the caller to finish saying the full sentence.
[0006] This type of barge-in by the agent and the caller helps to
expedite the process of problem resolution in a traditional call
center manned by human agents.
[0007] In an automated call center, barge-in by a caller is often
permitted to speed up the process of problem resolution. For
example, an experienced caller familiar with the call center
functions may decide to bypass the initial greetings playback by
barging-in in the middle of the greetings with an inquiry. However,
the technology for a barge-in function by an automated agent does
not currently exist.
SUMMARY
[0008] An automated agent system and method includes an action
classifier which scores a representation of a portion of a user
inquiry based on a knowledge bank. A comparing device compares a
running score of the representation to a threshold to determine if
enough information has been collected to respond early to the
inquiry. An action mechanism provides an early response to the user
prior to the user completing the inquiry if the threshold has been
exceeded.
[0009] A method for responding early to a user inquiry in an
automated agent system includes scoring a representation of a
portion of a user inquiry based on a knowledge bank, and comparing
a running score of the representation to a threshold to determine
if enough information has been collected to respond early to the
inquiry. A response is made to the user prior to the user
completing the inquiry.
[0010] These and other objects, features and advantages will become
apparent from the following detailed description of illustrative
embodiments thereof, which is to be read in connection with the
accompanying drawings.
BRIEF DESCRIPTION OF DRAWINGS
[0011] The disclosure will provide details in the following
description of preferred embodiments with reference to the
following figures wherein:
[0012] FIG. 1 is a block diagram of an automated agent that
expedites responding to user inquiries in accordance with one
illustrative embodiment; and
[0013] FIG. 2 is a block/flow diagram for an illustrative
system/method, which expedites response time to a user inquiry.
DETAILED DESCRIPTION OF PREFFERED EMBODIMENTS
[0014] Illustrative embodiments of the present disclosure provide
systems and methods for barge-in operation by an automated agent to
expedite the problem resolution process. In a particularly useful
embodiment, when a caller initiates an inquiry, the automated agent
keeps recognizing the input speech in conjunction with a knowledge
bank and tries to make a reasonable guess regarding the caller's
intent. Information in the knowledge bank is updated continuously,
either from a call center management team, if they know of
prevalent problems, or it could be culled from caller inputs, if a
number of special issues dominate the calls.
[0015] As soon as the automated agent is confident, it barges in on
the caller with a response to the inquiry. Without this type of
barge-in method, the process of problem resolution is likely to be
slower, making the caller unhappy with the automated call center
experience.
[0016] The invention described in this disclosure enables barge-in
by an automated agent with the help of a knowledge bank to speed up
the process of caller's problem resolution. One advantage is that
the caller's problem is resolved faster, making the caller
satisfied with the call center experience.
[0017] It should be understood that the elements shown in the FIGS.
may be implemented in various forms of hardware, software or
combinations thereof. Preferably, these elements are implemented in
a combination of hardware and software on one or more appropriately
programmed general-purpose digital computers having a processor and
memory and input/output interfaces.
[0018] It should be further understood that the use of the word
inquiry includes any uttered speech from a user that can evoke a
response by the agent system described herein. Inquiry may include
a user question, a command, a comment, or a response as the case
may be.
[0019] Referring now to the drawings in which like numerals
represent the same or similar elements and initially to FIG. 1, an
automated call center 10 is shown in accordance with an
illustrative embodiment. Speech input 12 from a caller is processed
by a speech recognizer 14 and an action classifier 16 to determine
an action to be taken. The action classifier 16 is aided in its
task by a knowledge bank 18, which knows about a domain and a state
of the inquiry. The knowledge bank stores information about
historically asked questions or responses and may be trained in
accordance with the specific application(s) to be performed by
agent 10.
[0020] Action classifier 16 preferably scores using likelihoods or
probabilities that a particular response or inquiry will be posed
to agent 10. This calculation or determination may be performed in
a plurality of ways. For example, a likelihood that a portion of
the response is actually part of a known question is checked and a
score is computed. Known algorithms may be employed. An action
classifier score is output from the classifier 16 and checked
against a threshold in block 20.
[0021] In a conventional system this check would be performed only
after the end of the entire inquiry. If the final score falls below
a preselected confidence threshold, the caller is typically
prompted to repeat the query, as the conventional system was unable
to make a proper decision. The caller could then repeat the inquiry
or make a new one as needed and receive a reply from the automated
agent and so on. This suffers from unnecessary delay as described
above.
[0022] In accordance with embodiments of the present invention, the
speech input 12, which is processed by the speech recognizer 14 and
the action classifier 16 depends on the knowledge bank 18 to keep
track of the domain and state of the inquiry. When an action is
taken, the knowledge bank is updated to reflect the latest status
of a call and learn new responses.
[0023] Rather than waiting until the end of speech is reached, the
action classifier score output from classifier 16 is monitored
continuously in block 20. When the running score in block 20
exceeds a certain confidence threshold, the automated agent 10
permits a barge-in, even though the caller has not finished
speaking the whole inquiry. This includes that when the running
score exceeds the threshold, the automated agent 10 has enough
information to make a decision and respond to the caller.
[0024] As long as the running score is less than the threshold, the
system 10 keeps checking to see if the caller has finished
speaking. When the end of speech is detected in block 24, the rest
of the procedure may follow the paths as described herein. Thus,
the final score is compared against a confidence threshold in block
26. If the final score is below this threshold, the caller is
prompted to repeat the query in block 28. Otherwise, action is
taken in block 30 in a no barge-in mode.
[0025] In block 20, when the running score exceeds the threshold, a
barge-in action is taken in block 32.
[0026] The system 10 permits automated agent barge-in, and the
action classifier 16 responds in an appropriate manner in
accordance with the system's understanding of the user's inquiry,
expediting the problem resolution process.
[0027] In an illustrative example, a call center receives a call
and the user provides input (12) in the form of an inquiry, for
example, "Is tax included in the price of the-item?" The speech
recognizer 14 would begin recognizing the speech, and when "Is tax"
is recognized and input to the action classifier 16 using the
knowledge bank 18 would score these two words with a score of say
78%. In this example, the threshold in block 20 is 75%. Therefore,
the running score of 78% exceeds the threshold of 75%.
[0028] The action of barging-in would be performed in block 32,
followed by a response generated by a dialog manager 36 or action
classifier 16, which generates an appropriate response. For
example, in the present case, the response would be "No, tax is not
included in the price".
[0029] Referring to FIG. 2, a block/flow diagram illustratively
depicts a system/method for responding early to a user inquiry in
an automated agent system in accordance with illustrative
embodiments. In block 202, recognizing speech input by a caller in
portions as they are received is optionally performed. Text or
other input form may be provided instead.
[0030] In block 203, the speech representation portions from the
user inquiry are scored by the action classifier based on a
knowledge bank. The representation preferably includes recognized
or decoded speech, as it is being received. Faster decoding
improves efficiency of the present invention. The speech is
preferably decoded in portions so that an early inference or guess
may be made as to the inquiry, response or comment being made by
the user.
[0031] In block 204, a running score of the representation
(portions of the speech) is compared to a threshold to determine if
enough information has been collected to respond early to the
inquiry. This preferably includes periodically or continuously
monitoring the running score to ensure an earliest response to the
user's inquiry.
[0032] In block 206, if enough information has been received, prior
to the user completing the inquiry an appropriate response is made.
This may include, e.g., barging in on a user's inquiry prior to its
completion, in block 208.
[0033] In block 210, when an end of speech for the user's inquiry
is detected by conventional means, e.g., by monitoring the speech
energy, the final score from the action classifier is compared to a
threshold in block 211. This may occur if an earlier barge-in or
response cannot be made. Detecting the end of speech includes
generating a final action classifier score, which is compared to a
final threshold, such that if the final score exceeds the
threshold, response is made on the inquiry in block 212 in the no
barge-in mode. Otherwise, the user is prompted to repeat the
inquiry or give a new command or statement in block 214.
[0034] In block 216, the knowledge bank includes information based
on a history of previous inquiries, and is updated after an action
is taken. Updating the knowledge bank may be employed to increase
the response capabilities of the system.
[0035] Having described preferred embodiments of a system and
method system and method for expedited resolution of inquiries for
an automated agent (which are intended to be illustrative and not
limiting), it is noted that modifications and variations can be
made by persons skilled in the art in light of the above teachings.
It is therefore to be understood that changes may be made in the
particular embodiments disclosed which are within the scope and
spirit of the invention as outlined by the appended claims. Having
thus described aspects of the invention, with the details and
particularity required by the patent laws, what is claimed and
desired protected by Letters Patent is set forth in the appended
claims.
* * * * *