Question Answering Chatbot using Deep Learning with NLP

In spite of the number of techniques, models and datasets, Question Answering is still an exacting problem because of the issues in understanding the question and extracting the correct answer. It refers to creating platforms that when given a question in a natural language by humans, can automatically answer it. While many information retrieval chatbots achieve the task, recently, deep learning has earned a lot of attention to question answering due to its capability to learn optimal representation for the given task. This paper aims to build a closed domain, factoid Question Answering system. We recruit NLP methods of pattern matching and information retrieval to create an answer candidate pool. Before scoring similarities between the question and answers, we map them into some feature space. Our approach solves this task through distributional representations of the words and sentences wherein encodings store their lexical, semantic, and syntactic aspects. We use a convolutional neural network architecture to rank these candidate answers. Our model learns an optimal representation for the input question and answer sentences and a matching function to relate each such pair in a supervised manner from training data. Our model does not require any manual feature engineering or language sensitive data; hence can be extended to various domains. Training and testing on TREC QA, a Question Answering dataset, showed very promising metrics for our model.