USING MULTI-LAYER LSTMS FOR QUESTION RETRIEVAL

Abstract

Question retrieval is one of the important problems in the Community Question Answering system. The biggest challenge of this problem is the lexical gap between the words and phrases of the first and second question. Although there are many studies applied to this problem, the exploitation of multi-layer LSTM model has not been tested on this problem. In this paper, we exploit a multi-layer LSTM model applied to the problem of finding similar questions for the purpose of exploiting hidden semantics of sentences. The multi-layer LSTM model is capable of synthesizing semantics by multiple layers and exploits hidden semantics through many layers. Our model learned the semantics of sentences and improved the performance of finding question. The results show that the model with 3 layers gives the best results compared to the original LSTM model and other multi-layer models on the 2017 semeval dataset for the problem of finding similar questions.