Mining Mobile Users’ Activities Based on Search Query Text and Context

Abstract

Mobile search market is growing very fast. Mining mobile search activities is helpful for understanding user preference, interest and even regular patterns. In previous works, text information contained by either search queries or web pages visited by users is well studied to mine search activities. Since rich context information (e.g., time, location and other sensor inputs) is contained in the mobile search data, it has also been leveraged by researchers for mining user activities. However, the two types of information were used separately. In this paper, we propose a graphical model approach, namely the Text and Context-based User Activity Model (TCUAM), which mines user activity patterns by utilizing query text and context simultaneously. The model is developed based on Latent Dirichlet Allocation (LDA) by regarding users’ activities as latent topics. In order to guide the activity mining process, we borrow some external knowledge of topic-word relationship to build a constrained TCUAM model. The experimental results indicate that the TCUAM model yields better results compared with text-only and context-only approaches. We also find that the constrained TCUAM model is more effective than the unconstrained TCUAM model.