Knowledge Construction for the Broadcasting Content by Using Audience Oriented Data

This paper proposes a novel method to construct the content knowledge from audience generated data. The proposed method gathers audience data related with the broadcasting content and extracts images and meaningful keywords for knowledge construction. The knowledge graph is constructed with extracted entities based on their positions in the data and became the content knowledge by connecting images in the knowledge graph to the certain frame in the content. Through the matching process, it excludes noise images like icon and banner. The distance with timeline locality matches images and frames based on singular feature for scenes in the content. Finally, ORB is adopted to compare both images and frames with complex visual information. In experiments, for four episodes in a Korean TV series, the proposed method shows 72.3% of precision and 97.8% of recall in filtering while 0.737 of F-measure in matching. The constructed content knowledge is applied in face recognition to show the feasibility. By adopting the content knowledge, the performance of baseline method is increased up to 59%.