Scripts as source of information to contextual video advertising

• Raoni Simões Ferreira USP
• Karla Okada Intituto Nokia de Tecnologia
• Marco Cristo UFAM
• Klessius Berlt UFAM
• Edleno Silva de Moura UFAM
• David Fernandes UFAM

Resumo

The wide availability of online video content in the Internet has changed the way users interact with TV. In fact, TV users often watch the news, movies and TV series through the Web. Along with this change, new advertisement of products and services have emerged. The success of contextual advertising in the Web, mainly related to finding advertising keywords on page content, has motivated us to apply them to video. Textual evidence extracted from videos may provide good contextualization source for advertising. In this work, we evaluate the usefulness of movies scripts as a source of information for contextual advertising in video. We adopted a machine learning-based strategy to finding keywords and advertising. We studied not only features proven be useful in earlier studies, as well as novel features proposed and derived from the ad collection. We evaluate the impact of using scripts both in finding keywords and in finding relevant ads. The results indicate that the studies features were more useful for finding keywords than ads. Features derived from the ad collection performed consistently well in finding keywords. We also observed that the best keywords are found in the script section which describes the characters' actions and scenarios than in the one which describes the dialogues.