Question Answer Re-Ranking using Syntactic Relationship

With the arrival of the World Wide Web, the tremendous increase in textual data has encouraged the development of such platforms where a user can answer a question or ask a question in natural language. Community Question Answering (CQA) based websites play a significant role in the rise of the Social Web. These systems are designed to answer complex user queries effectively. In this study, a system has been proposed to solve the problem of re-ranking relevant answers to community questions by considering the syntactic structures between them using Tree Kernels i.e. Partial Tree Kernels (PTK), SubTree Kernels (STK), and SubSet Tree Kernels (SSTK). For this purpose, various experiments were conducted to achieve maximum accuracy and mean average precision score. The results were compared with an already existing state-of-art system and with a system using standard information retrieval similarity measures including cosine similarity, BM25, Levenshtein distance, and Jaccard coefficient. Results show the superior performance of tree kernels over compared baseline similarity measures.