Expansion of Video Categories Through Link and Users Graphs

Boris Okun
Russia, Yandex
For problems such as search and recommendation of video content to users it is quite important to have reliable algorithms for categorical classification. Text-based approaches to such problems are often insufficient due to a lack of textual information about documents. A content analysis-based approach may be more effective but it is very resource-intensive.
As the categorical classification problem is similar to adult video detection it is reasonable to try similar approaches. In current work we apply two different graph-based algorithms to expand text-based classification. The first algorithm, which has shown its efficiency for adult video detection, uses a bipartite users-documents graph and naive Bayes approach to calculate the probability that a document belongs to a particular category. The second algorithm is a modification of the link graph-based algorithm described in [1]. We apply these algorithms to the following categorical markups: Games, Movies, Music, News, Movie Trailers, HowTo.