Browse Prior Art Database

APPLICATION OF KOLMOGOROV COMPLEXITY TO TV CONTENT SIMILARITY

IP.com Disclosure Number: IPCOM000238349D
Publication Date: 2014-Aug-19
Document File: 8 page(s) / 59K

Publishing Venue

The IP.com Prior Art Database

Related People

Andre Surcouf: AUTHOR [+2]

Abstract

The similarities between programs and entities associated with contents (e.g., media programs), such as actors, etc., can be automatically extracted from a well defined set of corpuses. These techniques do not require any initial human intervention whatsoever, and can be fully automated. A new disruptive method is provided that automatically extracts semantic relations between arbitrary objects in a way that is background, domain, genre, and language independent. Different classification methods and tools to evaluate "distances" may be used.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 19% of the total text.

Page 01 of 8

APPLICATION OF KOLMOGOROV COMPLEXITY TO TV CONTENT SIMILARITY

AUTHORS:

Andre Surcouf Enzo Fenoglio

CISCO SYSTEMS, INC.

ABSTRACT

    The similarities between programs and entities associated with contents (e.g., media programs), such as actors, etc., can be automatically extracted from a well defined set of corpuses. These techniques do not require any initial human intervention whatsoever, and can be fully automated. A new disruptive method is provided that automatically extracts semantic relations between arbitrary objects in a way that is background, domain, genre, and language independent. Different classification methods and tools to evaluate "distances" may be used.

DETAILED DESCRIPTION

    In digital television (TV), it has always been a challenge to determine the level of "similarity" between two or more contents (programs). (The term "content(s)" and "program(s)" are used interchangeably herein.) This "similarity" is usually defined by comparing some attributes describing the contents (e.g., contents from the same director or having the same genre have some "similarity"). The more attributes two contents have in common the more they are "similar."

    In all cases, the "similarity" cannot be determined without going through some content metadata analysis. This method is unfortunately error prone since it does not work across different languages and since it is by definition highly dependent on the metadata quality and completeness. There are also other ways to attempt to guess "content similarity" based on contribution from different users. Knowing the similarity between programs can be used for different purposes, including but not limited to, recommendation, advertisement placement, data clustering, classification, multimedia content annotation.

Copyright 2014 Cisco Systems, Inc.
1


Page 02 of 8

    As an example, to achieve similar results one technique that is known to have bee used involves defining loads of categories and subcategories to have better recommendations. In this approach, programs from the same subcategory are declared "similar". There is also some adjacency between Netflix subcategories that is used to find programs which are "almost similar." People were paid to watch films and tag them with all kinds of metadata. This process is so sophisticated and precise that taggers receive complicated training document that teaches people how to rate movies on their sexually suggestive content, goriness, romance levels, and even narrative elements like plot conclusiveness.

    Defining this large number of subcategories, as well as grouping contents in those subcategories are manual processes because this information cannot be directly inferred from the metadata information provided alongside the content by the production studios.

    Presented herein are techniques for a totally different method to determine program similarities. This method does not require any detailed understanding of the metadata meaning and does not require any ma...