Tightening Data Analysis and Feature Extraction for Micro-blog Recommendation

Information explosion in micro-blog services brings bad experience to users. Therefore, approaches that leverage users' preferences in applications of messages filtering, recommendation and searching were proposed by scholars in recent years. In general, features extraction is critical process in applying these approaches to applications. However, current researches have been focused on finding better models on varied features, but ignored why these features were used. To answer this question, we make an intuitive assumption that directly applying the result of data analysis, especially using the result of data analysis as features in our proposal, might lead to better performance than general raw features. In this paper, we propose to use these new features in a naive approach and a learning to rank approach for application of messages recommendation in micro-blog service. The experiments by the two approaches over a large real-world data set, which compare performance of proposed new features and raw features, support our assumption.