Beyond Relevancy: Multimedia Analytics for Exploratory Video Search
Project: Research
Researcher(s)
Description
The existing search techniques are mostly relevancy driven. Specifically, given a query,the underlying engine seeks to sort the documents in descending order of relevancy tothe query, followed by list-wise presentation of the results. Such strategy is deemed tobe effective, if the user has a clear search goal (e.g., top-100 kindergartens in HongKong) and expects an answer in the first page of results. For a complex query (e.g., HongKong education reform since 1997), however, it becomes the user’s responsibility toexplore the rank list to reveal the multi-faceted information underneath the query. Theexploration of search results can be highly difficult for users whose purpose are to learnand investigate. In practice, the exploration of search results is more complicatedbecause of two facts: the search intent is not static but may switch gradually when moreinformation is exposed; the search results are not purely text-based but mixed with avariety of multimedia information such as videos and images.This project targets for a multimedia-based exploratory search system beyond relevancyseeking and list-wise presentation. The focus of studies will be on addressing fourresearch problems in exploratory search. First, the search is recall-oriented. Thepresented information in the first page of results should be as complete, relevant andless redundant as possible, for an overview of multiple facets in a query beforenavigating the results in the next page. Second, the relationship among search results ofdifferent media should be exposed for efficient comprehension of information. Third, theindex engine should continuously capture the time-evolving nature of potential searchtopics for displaying of up-to-date event relationship without information delay. Finally,there is no absolute measure of relevancy ranking among the search results. Theselection of results to display is not only based on content relevancy, but also contextssuch as user interaction to make sense of the presented results.This project will contribute by researching a solution that integrates user (throughmodeling search behavior), data (through multimedia analysis) and space (throughanalytical presentation) for exploratory search of Internet videos. The solution aims forrecall-oriented techniques in providing multi-perspective information access andnavigation; and utility-driven strategies that emphasize of how the result of multimediaanalysis could be utilized by users (e.g., through visualization and recommendation) forsubsequence search. These two peculiarities basically mark the difference between theproposed works in this project and the conventional goal-driven precision-orientedsearch techniques.Detail(s)
Project number | 9042053 |
---|---|
Grant type | GRF |
Status | Finished |
Effective start/end date | 1/01/15 → 3/06/19 |
- multimedia search,multimedia analytics,video browsing,cross-media analysis,