On Improving a Microblog Ranking

Microblog ranking is a hot research topic in recent years. Most of the related works apply TF-IDF metric for calculating content similarity while neglecting their semantic similarity. And most existing search engines which retrieve the microblog list by string matching the search keywords is not competent to provide a reliable list for users when dealing with polysemy and synonym. Besides, treating all the users with same authority for all topics is intuitively not ideal. In this paper, a comprehensive strategy for microblog ranking is proposed. First, we extend the conventional TF-IDF based content similarity with exploiting knowledge from WordNet. Then, we further incorporate a new feature for microblog ranking that is the topical relation between search keyword and its retrieval. Author topical authority is also incorporated into the ranking framework as an important feature for microblog ranking. Gradient Boosting Decision Tree(GBDT), then is employed to train the ranking model with multiple features involved. We conduct thorough experiments on a large-scale real-world Twitter dataset and demonstrate that our proposed approach outperform a number of existing approaches in discovering higher quality and more related microblogs.