Predicting Current User Intent with Contextual Markov Models

摘要: In many web information systems like e-shops and portals predictive modeling is used to understand user intentions based on their browsing behavior. User behavior inherently sensitive various contexts. Identifying such relevant contexts can help improve the prediction performance. this work, we propose a formal approach in which context discovery process defined as an optimization problem. For simplicity assume concrete yet generic scenario considered be secondary label of instance that either known from available contextual attribute (e.g. location) or induced training data novice vs. expert user). ideal case, objective function problem has analytical form enabling us design algorithm solving directly. An example with Markov models, typical for behavior, shows derived provides useful mathematical insights Experiments real-world use-case show discover allowing significantly models.