Query Feedback Based Twitter Search Optimization

In recent years, twitter has become a popular social media for people to communicate with each other. It can provide large scale of real-time information, such as news, hot topic, comments and so on. Twitter search has many different aspects when compared to traditional search engines. Twitter search can provide more fresh and sociable contents. The demand for twitter search is increasing dramatically. How to optimize twitter search results is crucially important. Relevance feedback technique, can do query expansion and increase search performance largely. This paper focuses on twitter dataset, explores the relevance techniques and proposes a ranking model based on relevance feedback. The mainly works are as follows:Firstly, this paper proposes a refined RF algorithm based on relevance model and increase the performance of query expansion. Besides, we use word activation force based RF algorithm, generate word net and find the expansion words that topic words activate.Secondly, by incorporating diverse sources of features, learning to rank has been widely used in real-time Twitter search, where users tend to acquire the most relevant and fresh information. We investigate how the results of pseudo relevance feedback can be used to better estimate the tweet’s relevance. In particular, we propose a combined learning to rank framework that integrates a linear ranking model with query feedback and multiple tweet features. Through evaluation using the TREC2011-2013Microblog track topics, we thoroughly evaluate the effectiveness of our approach. Results show that our combined learning to rank approach could strongly outperform over the general LTR method. Moreover, our study highlights the further use of hyperlinks in tweets and emphasis its importance. Finally, we design and realize a relevance feedback twitter search system.