Deep Learning Approaches for Answer Selection in Question Answering System for Conversation Agents

The conversation agent acts as core interfaces between a system and user in answering users queries with proper responses. Question answering system acquires an important role in the information retrieval field. The deep learning approach enhances the accuracy in answering complex questions. As outcome, the user is receiving the precise answer instead of large document collections. The aim of this paper is to develop a model with deep learning approach for improving answer selection process which supports more relevant answer displaying by conversation agents. To achieve this, word2vector used for word representation and biLSTM attentive model is used for training, testing and disclosure play precise answer. Question type is identified using POS-tagger based Question Pattern analysis (T-QPA) model. The knowledgebase is created from the bench mark datasets bAbI Facebook (simple QA tasks), TREC QA, Yahoo! Answer, Insurance QA dataset. The proposed framework is built by embedding of questions and answers based on bidirectional long short-term memory (biLSTM) attentive models. The similarity between questions and answers has been measured by semantic and cosine similarity. The proposed model reduces the search gap in extracting among user queries and answer sentences in the education domain. The system results are evaluated with the standard metrics MAP, Top 1 accuracy, F1- Score for the answer selection.