2020 IASA - FIAT/IFTA Joint Conference

NHK preserves over 1 million titles of programs and 8 million news footages, which are used not only for rebroadcasting but also as materials for new program production. Until now, the only way to find video from these vast amounts of content was to do an exact keyword search, and in preparation for that, we had to manually input metadata to each   content. However, there are problems such as work labor for annotation and deterioration of retrieval accuracy caused by the fluctuation of the notation. To solve these problems, we have developed various search functions using AI engines. In this presentation, I will introduce "Similar Content Search" and "Free Text Search" using concept search technology, and the automatic annotation using face recognition technology.