Ranking expert responses and finding experts based on rank
View Patent ↗In a computerized social network, expert and user chat sessions are stored and rated probabilistically. Later user requests for information are met with an expert ranking, based on a balance of similarities between expert profile and questions; similarity between expert profile and prior chat sessions, and dynamically updated chat session ratings. New sessions can be rated automatically with reference to keywords distilled from past sessions responsive to user ratings—and based on session length.
1. A computer method comprising performing operations in at least one data processing device, the operations comprising:
maintaining, embodied in at least one computer readable medium, at least one database including expert profile information and historical chat session transcripts along with a respective quality indication for each transcript, the transcripts relating to past user inquiries of experts within a social network;
receiving a request for a new chat session including a question from a current user; and
deriving information relevant to the new chat session responsive to at least one of:
the quality indications; and
similarity between the request and text from chat session transcripts from the database, said deriving comprising, not necessarily in the following order:
comparing the question with text from the chat session transcripts to obtain first results;
comparing the question with the expert profile information to obtain second results; and,
combining the first and second results of together with the quality indications to yield a ranking of experts relevant to the question.
2. The method of claim 1 , wherein the request for the new chat session also includes a topic indication and the deriving operation comprises using the topic indication in selecting experts to rank.
3. The method of claim 1 , wherein combining comprises using a formula that includes a tunable parameter that balances the results of the first and second comparing.
4. The method of claim 3 , wherein the formula comprises:
Score
(
e
)
=
α
*
sim
(
profile
(
e
)
,
q
)
+
(
1
-
α
)
Σ
sim
(
session
(
e
,
u
,
i
)
,
q
)
*
[
p
(
e
,
u
,
i
)
-
0.5
]
where:
e is variable representing an expert;
q is a variable representing a question;
profile(e) is a function that retrieves a profile of an expert;
sim( ) is a function that calculates similarity of text;
u is a variable representing a past user, who engaged in a chat session;
i is an index running through the stored sessions; and
p( ) is a probabilistic score assigned by a ratings model.
5. The method of claim 1 , wherein the user and experts are members of a same organization.
6. The method of claim 1 , the operations further comprise dynamically updating the database responsive to new chat sessions.
7. The method of claim 1 , wherein the operations further comprise:
implementing a classification protocol with respect to the database, the classification protocol using one or more of
session length;
a first set of keywords from the database, the first set being associated with at least one session that has a positive quality indication; and
a second set of keywords from the data base, the second set being associated with at least one session that has a negative quality indication; and
upon completion of the new chat session, creating and storing a new quality indication for the new chat session based on the classification protocol.
8. The method of claim 1 , wherein quality indication is responsive to a binary classification model.
9. The method of claim 8 , wherein, for at least one chat session, quality indications are obtained from both chat session participants and the session is considered
positive if both quality indications are positive;
negative if both quality indications are negative; and
unrated if one quality indication is positive and the other negative.
10. A system comprising:
at least one medium for embodying machine readable data and machine performable program code;
at least one interface for communicating externally;
at least one processor adapted to perform operations responsive to the medium and interface, the operations comprising
maintaining, embodied in at least one computer readable medium, at least one database including expert profile information and historical chat session transcripts along with a respective quality indication for each transcript, the transcripts relating to past user inquiries of experts within a social network;
receiving a request for a new chat session including a question from a current user; and
deriving information relevant to the new chat session responsive to at least one of:
the quality indications; and
similarity between the request and text from chat session transcripts from the database, said deriving comprising, not necessarily in the following order:
comparing the question with text from the chat session transcripts to obtain first results;
comparing the question with the expert profile information to obtain second results; and,
combining the first and second results of together with the quality indications to yield a ranking of experts relevant to the question.
11. The system of claim 10 , wherein combining comprises using a formula that includes a tunable parameter that balances results of the first and second comparing.
12. The system of claim 10 , wherein the operations further comprise, not necessarily in the following order:
implementing a classification protocol with respect to the database, the classification protocol using one or more of
session length;
a first set of keywords from the database, the first set being associated with at least one session that has a positive quality indication; and
a second set of keywords from the data base, the second set being associated with at least one session that has a negative quality indication; and
upon completion of the new chat session, creating and storing a new quality indication for the new chat session based the classification protocol.
13. The system of claim 12 wherein
quality indication is responsive to a binary classification model; and
for at least one chat session, quality indications are obtained from both chat session participants and the session is considered
positive if both quality indications are positive;
negative if both quality indications are negative; and
unrated if one quality indication is positive and the other negative.
14. A computer program product for performing operations in at least one data processing device, the computer program product comprising:
a storage medium readable by a processing circuit and storing instructions run by the processing circuit for performing a method comprising:
maintaining, embodied in at least one computer readable medium, at least one database including expert profile information and historical chat session transcripts along with a respective quality indication for each transcript, the transcripts relating to past user inquiries of experts within a social network;
receiving a request for a new chat session including a question from a current user; and
deriving information relevant to the new chat session responsive to at least one of:
the quality indications; and
similarity between the request and text from chat session transcripts from the database, said deriving comprising, not necessarily in the following order:
comparing the question with text from the chat session transcripts to obtain first results;
comparing the question with the expert profile information to obtain second results; and,
combining the first and second results of together with the quality indications to yield a ranking of experts relevant to the question.
15. The product of claim 14 , wherein combining comprises using a formula that includes a tunable parameter that balances the first and second similarities.
16. The product of claim 14 , wherein the method further comprises, not necessarily in the following order:
implementing a classification protocol with respect to the database, the classification protocol using one or more of
session length;
a first set of keywords from the database, the first set being associated with at least one session that has a positive quality indication; and
a second set of keywords from the data base, the second set being associated with at least one session that has a negative quality indication; and
upon completion of the new chat session, creating and storing a new quality indication for the new chat session based the classification protocol.
17. The product of claim 14 wherein
quality indication is responsive to a binary classification model;
for at least one chat session, quality indications are obtained from both chat session participants and the session is considered positive if both quality indications are positive;
negative if both quality indications are negative; and
unrated if one quality indications is positive and the other negative.