Attention-based BiLSTM Model for Answer Extraction in Question Answering System

Answer extraction is the core technology of question answering (QA) system. Most of the previous works were based on syntactic features and knowledge base, which highly required the language knowledge. Accordingly, we proposed an attention-based bidirectional long short-term memory (BiLSTM) neural network model for answer extraction to automatically extract and analyze semantic features. The model remembers information in both forward and backward sequences to extract more complete semantic features. The attention mechanism assigns appropriate attention weights according to the semantic similarity of question-answer pairs, which enables the network to focus on the significant information and raises the probability of selecting the correct answer. We also study the prevailing semantic similarity calculation methods and choose the most suitable cosine similarity as the calculation criterion. Experimental results demonstrate that the proposed model outperforms the compared models.