Toffee - Semantic Media Search Using Topic Modeling and Relevance Feedback

Mikko Koho, Erkki Heino, Arttu Oksanen, Eero Hyvönen
2018 International Semantic Web Conference  
This paper considers relevance feedback [1, Ch. 5] search on the Web. Here the information need and query cannot be formulated in the outset-a typical situation in many search situations-but gets refined through making a series of queries and by evaluating the results in between. As an instance of such search the following problem setting is considered: since 1981, The Finnish engineering trade unions TEK and TFiF have given the yearly Finnish Engineering Award 3 to a "notable engineering or
more » ... hitectural work which has remarkably advanced technical competence in Finland". Would it be possible to devise a search system that could help the award committee members in finding out award winning candidates from the news and other materials on the Web? This paper presents and demonstrates the first results of our research on creating such a search service. The novel idea in the proposed approach is to combine implicit and explicit feedback methods [6] by using topic modeling [2] for extracting topics from the search results. Extracted topics and user feedback are used to generate new search keywords, which then guides the iterative search process. The developed search prototype Toffee is designed to work especially with Finnish language content, but can handle documents in any language.
dblp:conf/semweb/KohoHOH18 fatcat:rbfs3llcd5dxpnpvfrpnj3kdqe