Target-Oriented Content and Sentiment Analysis

Target-oriented analysis aims at mining the targeted information towards specified targets (objects of users' interest). There are many real-world tasks in the data mining and natural language processing (NLP) areas. This thesis focuses on two specific tasks, target-oriented content analysis, and target-oriented sentiment analysis. The first task is to generate target-specific topics for focused analysis, which is also quite helpful not only for computer science but also health science and social science. For this task, we developed a new target-focused generative model, which can distill the topics relevant to the given target. The second task is to infer the target-specific sentiment in review data. We proposed several alternative target-sensitive supervised learning solutions. Empirical results demonstrate the effectiveness of our approaches for both tasks. My further works on these two tasks are to use the idea of lifelong machine learning (LML) for performance enhancement. The intuition is that, when a system/learner performs tasks continuously, we want it to utilize the knowledge obtained from the past to help future tasks. To achieve this goal, we proposed two lifelong learning models for content analysis and sentiment analysis respectively. Experimental results show the usefulness of the LML solutions for both tasks.