Abstract: This paper proposes task-oriented content-based recommendation for cross service recommendation. The proposed method has two features: one is that task-based features are automatically mined from the web, second is that it estimates user's intention on task, which means "what user wants to do" and "what problem user has" by task-based profile representation. We evaluate improvement of recommendation accuracy by user evaluation, in which we collect ratings on variety of contents (i.e. mobile web content, TV programs, restaurants, sightseeing spots, and hotels) from 1,859 people and conduct cross validation. In an experiment, the combination of task-based profile representation and term-based representation yields a 17.7% improvement in MAE (Mean Absolute Error) compared to term-based profile representation only or domain (content category) based profile representation only.
Keywords: Content-based recommendation, user profiling, task, ontology learning
