Method and system for content classification

• US 9,916,381 B2
• Filed: 12/30/2008
• Issued: 03/13/2018
• Est. Priority Date: 12/30/2008
• Status: Active Grant

• Current Assignee: Telecom Italia S.p.A.
• Original Assignee: Telecom Italia S.p.A.
• Inventors: Antonelli, Fabrizio, Geymonat, Marina, Mana, Dario, Simeoni, Rossana, Candan, Selcuk Kasim, Cataldi, Mario, Di Caro, Luigi, Sapino, Maria Luisa
• Primary Examiner(s): Thomas, Ashish K
• Assistant Examiner(s): CHOJNACKI, MELLISSA M
• Application Number:
  US13/142,834
• Publication Number:
  US 20110264699A1
• Time in Patent Office: 3,360 Days
• Field of Search: None
• CPC Class Codes: G06F 16/353 into predefined classes

Abstract

A processing method for classification of contents in a domain that can be represented through a taxonomy includes generating a first digital mathematical representation of the taxonomy; generating a second digital mathematical representation of text documents different from the contents and containing keywords; processing the first and second digital mathematical representations for enriching the taxonomy, by associating keywords of the text documents with the first digital mathematical representation; generating a third digital mathematical representation of the contents; and processing the first digital enriched mathematical representation and third mathematical representation for classifying the contents in the enriched taxonomy.

Claims

1. A processing method for classification of contents in a domain represented through a taxonomy, comprising:

generating a first digital mathematical representation of a pre-existing version of the taxonomy, wherein the taxonomy represents a hierarchical relationship of elements in a category which includes nodes of specific elements that fall under a node of a broader element, the nodes existing independently of the contents themselves and representing at least one concept;

generating a second digital mathematical representation of text documents different from said contents and comprising keywords, the text documents including at least one journalistic publication written about the contents;

processing the first and second digital mathematical representation to enrich the taxonomy, from an initial state to an enriched state, by associating keywords of the text documents with the first digital mathematical representation, such that at least one keyword is extracted from one of the text documents to provide a new label to be associated with a node in the enriched taxonomy, wherein the at least one keyword is present in the text documents but not present in the initial taxonomy, and wherein enriching the first digital mathematical representation with keywords of the text documents comprises;

implementing a similarity function in order to classify the context documents to the taxonomy by associating the text documents with concepts of the taxonomy;

selecting representative keywords associated with said context documents to be used for enrichment; and

vectorial summing of vectors corresponding to nodes in the taxonomy and vectors of selected keywords associated with the context documents, in accordance with factors taking into account representative statistical significance, to generate enriched vectors for the nodes in the taxonomy;

generating a third digital mathematical representation of the contents from text descriptors of the contents which are separate and different from the text documents and the taxonomy; and

processing a first enriched digital mathematical representation and third digital mathematical representation for classifying the contents in the enriched taxonomy,

wherein the at least one keyword that is present in the text documents but not present in the initial taxonomy is associated with another keyword within the text documents that is the same as the node in the initial taxonomy,

wherein;

said first digital mathematical representation comprises weights indicating a contribution to a node of the taxonomy supplied by a plurality of concepts of the taxonomy, and a plurality of weights for at least one node in the enriched taxonomy is different than a plurality of weights for the same at least one node in the pre-existing version of the taxonomy;

said second digital mathematical representation comprises respective weights associated with keywords of the text documents, each indicating a relevance of a respective keyword;

said third digital mathematical representation comprises additional weights associated with additional keywords of the contents;

each indicating a corresponding relevance of a respective additional keyword, and

generating the first digital mathematical representation comprises;

defining a propagation matrix by adjacency associated with the taxonomy and representative of contributions of concepts toward other concepts of the taxonomy;

defining a matrix of concept vectors representative of the taxonomy;

defining a matrix of propagation effect depending upon the matrix by adjacency of the matrix of concept vectors; and

calculating a matrix representing the taxonomy obtained from the matrix of propagation by adjacency from the matrix of propagation effect and from the matrix of concept vectors.

2. The method according to claim 1, further comprising:

storing in a database data representative of said contents and data indicative of the relative classification in the taxonomy.

3. The method according to claim 1, wherein:

generating the first digital mathematical representation comprises generating first keyword vectors representative of concepts of taxonomy;

generating the second digital mathematical representation comprising generating second keyword vectors representative of text documents;

generating the third digital mathematical representation comprises generating third keyword vectors representative of contents to be classified; and

enriching the taxonomy comprises processing the first and second keyword vectors in order to associate keywords of the second vectors to concepts of the first vectors.

4. The method according to claim 1, wherein generating the first digital mathematical representation comprises determining said weights by means of an iterative calculation carried out starting from descriptive propagation coefficients of a degree of propagation between concepts of the taxonomy.

5. The method according to claim 1, wherein said matrix of propagation by adjacency is defined by a plurality of propagation coefficients of keys from a child node of a taxonomy to a parent node of the taxonomy.

6. The method according to claim 1, wherein generation of the first digital mathematical representation of the taxonomy comprises application of a concept propagation method.

7. The method according to claim 1, further comprising:

defining at least one further taxonomy;

calculating a respective further digital mathematical representation of said at least one further taxonomy;

processing the first digital mathematical representation and a respective further digital mathematical representation determining semantic relationships between concepts of the taxonomy and said at least a further taxonomy; and

implementing a method of content classification to the taxonomy as related to said at least a further taxonomy.

8. The method according to claim 1, wherein:

implementing a similarity function comprises evaluating a scalar product between the vectors corresponding to nodes in the taxonomy and the vectors of selected keywords associated with the context documents.

9. The method according to claim 1, wherein updating first vectors comprises:

interpreting the first vectors as a query to a document database, and an extension of a concept or a set of documents returned by the query and recognized as significant; and

updating the first vectors in order to take into account weights of keywords present in documents returned by the query and recognized as significant.

10. The method according to claim 9, wherein updating the first vectors is performed through a relevance feedback method by associating the first keyword vectors with interrogation queries of text documents.

11. The method according to claim 1, wherein content classification processing comprises:

implementing a respective similarity function between the first digital mathematical representation resulting from enrichment and the third digital mathematical representation of the contents.

12. The method according to claim 1, wherein:

said domain comprises one or more of the following domains;

cinematic genres, geographical locations, historic events, art, sport, and artists;

said contents are cinematic contents; and

said text documents are newspaper articles.

13. A computer system of classification of contents in a domain represented by means of a taxonomy comprising:

circuitry configured to;

generate a first digital mathematical representation of a pre-existing version of the taxonomy, wherein the taxonomy represents a hierarchical relationship of elements in a category which includes nodes of specific elements that fall under a node of a broader element, the nodes existing independently of the contents themselves and each representing at least one concept;

generate a second digital mathematical representation of text documents different from said contents and comprising keywords, the text documents including at least one journalistic publication written about the contents;

enrich the taxonomy, from an initial state to an enriched state, by processing the first and second digital mathematical representation and associate keywords of the text documents with the taxonomy, such that at least one keyword is extracted from one of the text documents to provide a new label to be associated with a node in the enriched taxonomy, wherein the at least one keyword is present in the text documents but not present in the initial taxonomy, and wherein enriching the first digital mathematical representation with keywords of the text documents comprises;

implementing a similarity function in order to classify the context documents to the taxonomy by associating the text documents with concepts of the taxonomy;

selecting representative keywords associated with said context documents to be used for enrichment; and

vectorial summing of vectors corresponding to nodes in the taxonomy and vectors of selected keywords associated with the context documents, in accordance with factors taking into account representative statistical significance, to generate enriched vectors for the nodes in the taxonomy;

generate a third digital mathematical representation of the contents to be classified from text descriptors of the contents which are separate and different from the text documents and the taxonomy; and

process the first and third digital mathematical representation for classifying the contents in the enriched taxonomy,

wherein the at least one keyword that is present in the text documents but not present in the initial taxonomy is associated with another keyword within the text documents that is the same as the node in the initial taxonomy,

wherein;

said first digital mathematical representation comprises weights indicating a contribution to a node of the taxonomy supplied by a plurality of concepts of the taxonomy, and combination of a plurality of weights for at least one node in the enriched taxonomy is different than a combination of a plurality of weights for the same at least one node in the pre-existing version of the taxonomy based on the generation of the enriched vectors for the nodes in the taxonomy;

said second digital mathematical representation comprises respective weights associated with keywords of the text documents, each indicating a relevance of a respective keyword;

said third digital mathematical representation comprises additional weights associated with additional keywords of the contents;

each indicating a corresponding relevance of a respective additional keyword; and

generating the first digital mathematical representation comprises;

defining a propagation matrix by adjacency associated with the taxonomy and representative of contributions of concepts toward other concepts of the taxonomy;

defining a matrix of concept vectors representative of the taxonomy;

defining a matrix of propagation effect depending upon the matrix by adjacency of the matrix of concept vectors; and

calculating a matrix representing the taxonomy obtained from the matrix of propagation by adjacency from the matrix of propagation effect and from the matrix of concept vectors.

14. The system according to claim 13, further comprising:

a database containing data representing said contents and a relative classification thereof, the database being consultable in order to identify at least one content through a link to another content associated with a keyword of said text documents.

15. The system according to claim 13, wherein the circuitry is further configured to:

apply similarity function between the first and second digital mathematical representations in order to classify context documents in said taxonomies; and

select representative keywords associated with said context documents and being used for enrichment.

16. The system according to claim 15 wherein:

the circuitry is configured to;

generate first vectors of keywords representative of concepts of the taxonomy;

generate second keyword vectors representative of concepts of text documents;

enrich the first and second keyword vectors and to associate keywords of the second vectors to concepts of the first vectors; and

generate third keyword vectors representative of contents to be classified.

17. A non-transitory computer readable medium that stores a program, which when executed by a computer, causes the computer to perform a processing method for classification of contents in a domain being represented through a taxonomy, the comprising:

generating a first digital mathematical representation of a pre-existing version of the taxonomy, wherein the taxonomy represents a hierarchical relationship of elements in a category which includes nodes of specific elements that fall under a node of a broader element, the nodes existing independently of the contents themselves and each representing at least one concept;

generating a second digital mathematical representation of text documents different from said contents and comprising keywords, the text documents including at least one journalistic publication written about the contents;

processing the first and second digital mathematical representation to enrich the taxonomy, from an initial state to an enriched state, by associating keywords of the text documents with the first digital mathematical representation, such that at least one keyword is extracted from one of the text documents to provide a new label to be associated with a node in the enriched taxonomy, wherein the at least one keyword is present in the text documents but not present in the initial taxonomy, and wherein enriching the first digital mathematical representation with keywords of the text documents comprises;

implementing a similarity function in order to classify the context documents to the taxonomy by associating the text documents with concepts of the taxonomy;

selecting representative keywords associated with said context documents to be used for enrichment; and

vectorial summing of vectors corresponding to nodes in the taxonomy and vectors of selected keywords associated with the context documents, in accordance with factors taking into account representative statistical significance, to generate enriched vectors for the nodes in the taxonomy;

generating a third digital mathematical representation of the contents from text descriptors of the contents which are separate and different from the text documents and the taxonomy; and

processing a first enriched digital mathematical representation and third digital mathematical representation for classifying the contents in the enriched taxonomy,

wherein the at least one keyword that is present in the text documents but not present in the initial taxonomy is associated with another keyword within the text documents that is the same as the node in the initial taxonomy,

said first digital mathematical representation comprises weights indicating a contribution to a node of the taxonomy supplied by a plurality of concepts of the taxonomy, and combination of a plurality of weights for at least one node in the enriched taxonomy is different than a combination of a plurality of weights for the same at least one node in the pre-existing version of the taxonomy based on the generation of the enriched vectors for the nodes in the taxonomy;

said second digital mathematical representation comprises respective weights associated with keywords of the text documents, each indicating a relevance of a respective keyword;

said third digital mathematical representation comprises additional weights associated with additional keywords of the contents;

each indicating a corresponding relevance of a respective additional keyword; and

generating the first digital mathematical representation comprises;

defining a propagation matrix by adjacency associated with the taxonomy and representative of contributions of concepts toward other concepts of the taxonomy;

defining a matrix of concept vectors representative of the taxonomy;

defining a matrix of propagation effect depending upon the matrix by adjacency of the matrix of concept vectors; and

calculating a matrix representing the taxonomy obtained from the matrix of propagation by adjacency from the matrix of propagation effect and from the matrix of concept vectors.