Aspect Based Sentiment Analysis on Review Data

Speaker Bio

Wei Xue received his B.S. and M.S. degrees in Computer Science from Zhejiang University, China in 2008 and 2011, respectively. Wei is a Ph.D. candidate at Florida International University’s School of Computing and Information Sciences. He is part of the Knowledge Discovery Research Group (KDRG) led by Dr. Tao Li. His research interest includes natural language processing and text mining.

Description

With proliferation of user-generated reviews, new opportunities and challenges arise. The advance of Web technologies allows people easily access a large amount of reviews of produces and services online. Knowing what others like and dislike becomes increasingly important for their decision making during shopping. The retailers also care more than ever about online reviews, because the vast pool of reviews enable them to monitor reputations and collect feedback efficiently. However, the average human reader will find difficult times in identifying and summarizing fine-grained sentiments buried in the opinion-rich resources. The traditional sentiment analysis, which focuses on the overall sentiments, fails to uncover the sentiments over the aspects of the reviewed entities.

In this dissertation, we study the research problem of Aspect Based Sentiment Analysis (ABSA), whose mission is to reveal the aspect-dependent sentiment information of review text. ABSA can be decomposed into three tasks: aspect extraction and classification, target expression detection, and aspect sentiment prediction. We propose several machine learning approaches based on topic models and neural networks. First, to extract the aspects discussed in reviews and detect aspect sentiments, we propose several topic models, which fully utilize aspect ratings. Second, we propose a multi-task learning model based on long-short term memory and convolutional neural network for aspect category classification and target expression detection. Third, for aspect-level sentiment classification, we propose an improved tree-structured long-short term memory using attention mechanism.