Deep Learning for Text Matching: A Survey

Text matching is one of the crucial technology in the field of Natural Language Processing (NLP), and it has been applied in many tasks, such as textual similarity, information retrieval and question answering. The target of text matching is to model the relationship between two input texts. In this paper, we aim to give a survey on recent advance techniques of deep-learning based text matching methods. Specifically, depending on whether a model will first encode a sentence into a fixed-length vector without any incorporating from the other sentence, the existing studies can be categorized into two major categories: representation-based models and interaction-based models. The latter can be divided into two groups according to the interaction methods. In addition, we summarize the strengths and weaknesses of these methods to help beginners in this area to choose the appropriate model for their application. Finally, we make a conclusion by highlighting several directions and open problems which need to be further explored by the community in the future.