Method, apparatus and computer program for information retrieval

• US 7,930,281 B2
• Filed: 01/04/2008
• Issued: 04/19/2011
• Est. Priority Date: 01/05/2007
• Status: Expired due to Fees

• Current Assignee: KDDI Corporation
• Original Assignee: KDDI Corporation
• Inventors: Hoashi, Keiichiro, Ishizaki, Hiromi, Sugaya, Fumiaki
• Primary Examiner(s): Rones; Charles
• Assistant Examiner(s): ORTIZ DITREN, BELIX M
• Application Number:
  US11/969,477
• Publication Number:
  US 20080168059A1
• Time in Patent Office: 1,201 Days
• Field of Search: 707/705, 707/999.1, 707/999.003
• US Class Current: 707/705
• CPC Class Codes: G06F 16/634 Query by example, e.g. query by humming
• CPC Class Codes: G06F 16/683 using metadata automatically derived from the content

Abstract

A query building module 6 includes a query DB 61 for managing a query vector of a piece of music, which matches a user preference, a similarity calculating unit 62 for calculating a similarity between the query vectors, which are registered in the query DB 61, a query vector integrating unit 63 for integrating a plurality of query vectors similar to each other into one piece of preference information, and a query updating unit 64 for newly registering the integrated new query vector to the query DB 61 and deleting the integrated two similar query vectors from the query DB 61, thereby updating the query DB 61, and builds a query Q in which all of the query vectors having high similarity are integrated and is an aggregation of the query vectors not similar to each other.

Claims

1. An information retrieval device for retrieving information matching a user preference from an aggregation of retrieval target, comprising:

vectorizing means for generating a feature vector of each piece of the retrieval target;

clustering means for clustering each piece of the retrieval target into a plurality of clusters based on the feature vector thereof;

representative vector generating means for generating a representative vector of each of said clusters;

preference importing means for urging the user to input the preference regarding the retrieval target and importing input preference information;

query building means for building a query based on said input preference information;

retrieval target narrowing means for comparing the representative vector of each of said clusters and the query, thereby narrowing the clusters to be retrieved into at least one cluster of which similarity is high;

retrieval means for comparing the feature vector of retrieval target belonging to said retrieval target cluster and said query, thereby extracting the retrieval target of which similarity is high; and

retrieval result outputting means for presenting a result of said retrieval to the user.

2. The information retrieval device according to claim 1, wherein said clustering means comprises:

means for clustering a part of the retrieval target into a plurality of clusters based on the feature vector thereof, and

means for comparing the representative vector of each cluster and the feature vector of each piece of retrieval target other than said part and deciding a belonging cluster of each piece of retrieval target other than said part.

3. The information retrieval device according to claim 1, wherein said query building means comprises:

means for generating a query vector representing a feature of each piece of input preference information;

means for calculating a similarity of each query vector; and

integrating means for integrating a plurality of query vectors similar to each other into one query vector,

wherein an aggregation of the query vectors not similar to each other becomes the query.

4. The information retrieval device according to claim 3, wherein said retrieval target narrowing means selects the retrieval target clusters for each query vector of the query, and

said retrieval means retrieves retrieval target similar to each query vector of the query from the retrieval target clusters obtained by narrowing based on the each query vector.

5. The information retrieval device according to claim 1 provided with means for normalizing the similarity between the feature vector of each piece of retrieval target belonging to each retrieval target cluster and the query based on a retrieval target distribution in each retrieval target cluster, wherein

said retrieval means extracts retrieval target of which similarity is high from said normalized similarity.

6. An information retrieval method for retrieving target matching a user preference from an aggregation of pieces of retrieval target, comprising the steps of:

generating a feature vector of each piece of the retrieval target;

clustering each piece of the retrieval target into a plurality of clusters based on the feature vector thereof;

generating a representative vector of each of said clusters;

urging a user to input the user preference regarding the retrieval target and importing input preference information;

building the query based on said input preference information;

comparing the representative vector of each of said clusters and the query, thereby narrowing the clusters to be retrieved into at least one cluster of which similarity is high;

comparing the feature vector of the retrieval target belonging to said retrieval target cluster and said query, thereby extracting the retrieval target of which similarity is high; and

presenting a result of said retrieval to the user.

7. The information retrieval method according to claim 6, wherein said step of building the query comprises:

generating the query vector representing the feature of each piece of input preference information;

calculating the similarity of each query vector; and

integrating a plurality of query vectors similar to each other into one query vector,

wherein an aggregation of the query vectors not similar to each other becomes the query.

8. A non-transitory computer readable medium with a program stored thereon for causing a computer to retrieve information matching a user preference from an aggregation of retrieval target by executing the steps of:

generating a feature vector of each piece of the retrieval target;

clustering each piece of the retrieval target into a plurality of clusters based on the feature vector thereof;

generating a representative vector of each of said clusters;

urging a user to input the user preference regarding the retrieval target and importing input preference information;

building a query based on said input preference information;

comparing the representative vector of each of said clusters and the query, thereby narrowing the clusters to be retrieved into at least one cluster of which similarity is high;

comparing the representative vector of retrieval target belonging to said retrieval target cluster and said query, thereby extracting the retrieval target of which similarity is high; and

presenting a result of said retrieval to the user.

9. The computer readable medium according to claim 8, wherein said step of building the query comprises:

generating the query vector representing a feature of each piece of input preference information;

calculating the similarity of each query vector; and

integrating a plurality of query vectors similar to each other into one query vector,

wherein an aggregation of the query vectors not similar to each other becomes the query.