Presentation Retrieval Method Considering the Scope of Targets and Outputs

We proposed the UPRISE (unified presentation slide retrieval by impression search engine) to retrieve scenes matched with given keywords from a video recording a lecture using presentation slides. To distinguish the most suitable scene for the keywords from scenes using reused or backtracked slides, we proposed a number of impression indicators based on the position, duration and context. The combined indicator is provide higher precision than the traditional TF-IDF. Moreover, our approach can still be improved by combining with some specificity factor like IDF. However, it is important to choose some appropriate range of calculating the factor. In this paper, we propose a new specificity factor ISFP (inverse scene frequency in a presentation) to indicate the appearance ratio of a keyword in a presentation corresponding to a lecture. We also have to consider scope of output. Sometimes, it is preferable to provide scenes from beginning the topic as output. To adjust the scope of output dynamically, we propose a new concept "output range". We evaluate the ISFP and scope varying method using actual lecture presentations. The experimental results indicate that the proposed method is effective especially for keywords appeared less frequently