System and method for collaborative ranking of search results employing user and group profiles derived from document collection content analysis

• US 6,327,590 B1
• Filed: 05/05/1999
• Issued: 12/04/2001
• Est. Priority Date: 05/05/1999
• Status: Expired due to Term

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Grasso, Antonietta, Chidlovskii, Boris, Glance, Natalie S.
• Primary Examiner(s): Breene, John
• Assistant Examiner(s): RAYYAN, SUSAN F
• Application Number:
  US09/305,435
• Time in Patent Office: 944 Days
• Field of Search: 707/5, 707/10, 707/3
• US Class Current: 707/734
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 16/9536 Search customisation based on social or collaborative filtering
• CPC Class Codes: G06F 16/9538 Presentation of query results
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

A system for ranking search results obtained from an information retrieval system includes a search pre-processor, a search engine and a search post-processor. The search preprocessor determines the context of the search query by comparing the terms in the search query with a predetermined user context profile. Preferably, the context profile is a user profile or a community profile, which includes a set of terms which have been rated by the user, community, or a recommender system. The search engine generates a search result comprising at least one item obtained from the information retrieval system. The search post-processor ranks each item returned in the search result in accordance with the context of the search query.

Claims

1. A system for ranking search results obtained from an information retrieval system, comprising:

a document collection comprising a set of documents associated with a user u;

a user profiler for constructing a user profile P^u=(t_i, w^u_i) comprising a set of terms {t_i} extracted from each document in the document collection with their weights w^u_i;

a search pre-processor, responsive to a search query q, for determining a search query profile P^q=(t^q_i,w^q_i) in accordance with the user profile, wherein t^q_i comprise the query terms having term weight, w^q_i;

a search engine, responsive to the search query, for generating a search result comprising at least one document obtained from the information retrieval system; and

a search post-processor, for each document d returned in response to the search query q, for generating a document profile, P^d=(t^d_i, w^d_i) and for evaluating the search document rank in accordance with;

${\mathrm{relevance}}_{q,d}=\frac{\underset{i}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>w_i^d\times <br><br>w_i^u·<br><br>w_i^q}{w_d}$

where w^d_i is the weight of term t in the response document d, w^u_i is the weight of term t in the user context profile, w^q_i is the weight of the term t in the query q, and w_d is the vector length “

projected”

on the context profile and evaluated as $w_d=\sqrt{\underset{t}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>{\left(w_i^u<br><br>w_i^d\right)}^2}.$

2. The system of claim 1, wherein the user is a member of a community.

3. The system of claim 1, wherein the user is an expert member of a community.

4. The system of claim 1, wherein the search post-processor, for comparing each search document with the documents in the document collection and, responsive to a match, for assigning a relevance thereto.

5. A system for ranking search results obtained from an information retrieval system, comprising

a search pre-processor, responsive to a search query, for determining a context of the search query in accordance with a predetermined user context profile;

a search engine, responsive to the search query, for generating a search result comprising at least one item obtained from the information retrieval system; and

a search post-processor, responsive to the search result, for ranking the item in accordance with the context of the search query;

a context profiler for generating a context profile; and

a document collection comprising a set of documents;

wherein the context profiler comprises a community profiler for constructing a term-weight vector for the community, the term-weight vector being extracted from each document in the document collection and the community comprising a plurality of users u;

wherein the community profile P^c=(t_i, w^c_i) comprises the set of terms {t_i} with their weights w^c_i for each of the individual users u in the community;

wherein each member u in the community is assigned a weight α

_u and for the whole community;

Σ

_uα

_u=1.

6. The system of claim 5, wherein the weight of term t_i in the community profile is evaluated as:

w^c_i=Σ

_uα

_uw_i^u.

7. The system of claim 5, wherein at least one user of the community is an expert having a weight w^exp_i for each profile term t_i.

8. A system for ranking search results obtained from an information retrieval system, comprising:

a search pre-processor, responsive to a search query, for determining a context of the search query in accordance with a predetermined user context profile;

wherein the predetermined user context profile comprises a community profile, P^c=(t_i, w^c_i) where w^c_i=Σ

_uα

_uw_i^u, is the weight of t_i in a profile of user u in the community and α

_u is the weight of each user in the community;

wherein the search pre-processor determines a profile of the query P^q=(t^q_i,w^q_i) in accordance with the predetermined user profile, wherein t^q_i comprise the query terms having term weight, w^q_i;

a search engine, responsive to the search query, for generating a search result comprising at least one item obtained from the information retrieval system; and

a search post-processor, responsive to the search result, for ranking the item in accordance with the context of the search query;

wherein the search post-processor evaluates each item d in the search result and generates a document profile P^d=(t^d_i, w^d_i) for each item d, where t^d_i is the profile term and w^d_i the weight of each term.

9. The system of claim 8, wherein the community profile includes an expert having a user profile, which contains weight w^exp_i for each profile term t_i.

10. The system of claim 9, wherein the search post-processor determines

the relevance of each item d in the search result in accordance with: ${\mathrm{relevance}}_{q,d}=\frac{\underset{i}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>w_i^d\times <br><br>w_i^{\exp }·<br><br>w_i^q}{w_d}$

where w^q_i is the weight of the term t in a query q, and w_d is the vector length “

projected”

on the context profile and evaluated as $w_d=\sqrt{\underset{t}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>{\left(w_i^{\exp }<br><br>w_i^d\right)}^2}.$

11. A method of creating a user context profile for use in ranking search results obtained from an information retrieval system, comprising:

providing a document collection comprising a plurality of relevant documents;

assigning a rating to each of the documents in the document collection to generate a document profile P^d=(t^d_i, w^d_i) for each document in the collection;

constructing a weight-term vector, wherein the weight-term vector includes a portion of the set of terms {t^d_i} with their weights w^d_i to form a user profile P^u=(t_i, w^u_i;

for each document d returned in response to a search query q, generating a document profile, P^d=(t^d_i, w^d_i); and

evaluating the search document rank in accordance with;

${\mathrm{relevance}}_{q,d}=\frac{\underset{i}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>w_i^d\times <br><br>w_i^u·<br><br>w_i^q}{w_d}$

where w^d_i is the weight of term t in the response document d, w^u_i is the weight of term t in the user context profile, w^q_i is the weight of the term t in the query q, and w_d is the vector length “

projected”

on the context profile and evaluated as $w_d=\sqrt{\underset{t}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>{\left(w_i^u<br><br>w_i^d\right)}^2};<br><br>\mathrm{and}$

and

updating the user context profile using the highest ranked item returned from the search query.