Sugato Chakrabarty

I Know What You Feel: Analyzing the Role of Conjunctions in Automatic Sentiment Analysis

We are interested in finding how people feel about certain topics. This could be considered as a task of classifying the sentiment: sentiment could be positive, negative or neutral. In this paper, we examine the problem of automatic sentiment analysis at sentence level. We observe that sentence structure has a fair contribution towards sentiment determination, and conjunctions play a major role in defining the sentence structure. Our assumption is that in presence of conjunctions, not all phrases have equal contribution towards overall sentiment. We compile a set of conjunction rules to determine relevant phrases for sentiment analysis. Our approach is a representation of the idea to use linguistic resources at phrase level for the analysis at sentence level. We incorporate our approach with support vector machines to conclude that linguistic analysis plays a significant role in sentiment determination. Finally, we verify our results on movie, car and book reviews.

Application of ontology guided search for improved equipment diagnosis in a vehicle assembly plant

In the body shop of an automobile assembly plant, having access to correct and timely diagnostic information is very important for solving equipment and tooling maintenance problems. Variation Reduction Adviser (VRA) is an internal General Motors (GM) system that contains information related to the problems encountered in process, their root cause and possible solutions. This paper presents our work on ontology based diagnosis, where, a thesaurus (which is a simple form of ontology) has been used for retrieving diagnostic information. A thesaurus has been developed from existing problem descriptions and their solutions written in a natural language (such as English). A systematic methodology has been developed for the creation of a thesaurus. The results of ontology based diagnostic information retrieval have been compared with ‘exact match’ information retrieval.