A Web Recommendation System based on Individual Preference Estimated from Twitter
Takeshi Toda, Mizuho Sawada · 2014
This paper proposes a web recommendation system that estimates and dynamically updates individual preference with twitter, in order to reduce web search effort. The proposed system gathers personal comments on twitter, extracts objectpredicate pairs by text analysis, and ranks the objects with weighting of the paired predicates in accordance with a prepared predicate-point dictionary such as “like so much (+5 points)” and “hate (-5 points).” We implemented the proposed system on a server in our laboratory using Twitter API for getting comments on Twitter, Yahoo API! for the text analysis and Bing API for the web search. In an experiment, we evaluated recall and precision of the objects ranking obtained by the proposed system. We also evaluated a precision of web page recommendation searched by top-ranked object. From the experimental result, the proposed web recommendation system provided higher relevance ratio