| There is a prominent issue around internet with the rapid growth of the online information.lt is getting harder for users to find the appropriate information they need. Some of the canonical approaches like classified catalogue, portal website, search engine cannot solve this problem gracefully, thus comes the personalized recommendation technique. Personalized recommender system is one type of super Business Intelligence system based on big data mining; it analyzes user’s history activity log, filters out those information this user don’t really care, then present with user the most possible interested information thus improving his user experience;Similar challenges appeared in media channel like online news, newly emerged news sites start to use personalized news recommendation system, and this is a hot area for research as well. There are four major types of personal recommendation algorithms including Content Based recommendation, Collaborative Filtering, Web Based recommendation and Hybrid recommendation. Solving some of the common pain points of these algorithms like cold start, low precision&recall, over personalization will greatly improve the research and application of recommender system;This thesis summarized recent research result around personal recommender system, piloting the new approaches of using social signal to improve quality of recommendation result, the major task and result including:(1) Combine the social signal with user history log information;(2) Combine user action in social media like forward, thumb up with user’s tag;(3) Make better recommendation result based on better user interest understanding;(4) Based on above idea, and considering real-time, data volume, precision and recall, implement a personalized news recommendation system;The result of this thesis providing a new way of improving the accuracy, cold start, novelty of result, some additional work like fine-tuned granularity and expanded user signal sources can further been explored. |
PDF Full Text Request