Automating Viewers' Side Annotations on TV Drama from Internet Bulletin Boards

This paper proposes a method for creating the viewers' side annotations that reflect viewers' attentions on TV dramas. Internet bulletin boards are filled with large amount of viewers' dialogues concerning TV programs. Our approach is to extract the viewers' attention embedded in these dialogues and express them in the form of the graphical structures called attention graphs. The attention graphs act as viewers' side annotations. Viewers' side annotations assist in providing viewers with hints to locate their favorite scenes in full-length TV programs. In general, Internet bulletin boards are described without any particular order and are expressed in a poor grammatical manner. Our approach is to statistically recognize the notable frequencies of the words which represent viewers' attentions out from such unstructured documents. Attention graphs are the outcome of this approach. Attention graphs are evaluated by three types of tests. The test results demonstrate that attention graphs sufficiently act as viewers' side annotations, in terms of pointing out which scene the viewers pay attention to and clarifying how deeply viewers are impressed by the scenes.