A Semantic Text Similarity Model for Double Short Chinese Sequences
Tang Shancheng, Bai Yunyue, Ma Fuyu · 2018
Semantic Text Similarity plays a major role in natural language processing. In recent years, researchers have paid considerable attention to Semantic Text Similarity. Some breakthroughs have been made in English, but there are two disadvantages when these models are applied to Chinese: Single sequence models don't consider semantic ambiguity such as polysemy, synonym; these models don't consider that Chinese stop words are important for Chinese word segmentation, voice analysis, semantic understanding. Firstly, in order to overcome the first problem, we proposed the double short text sequences model that has two identical LSTM (Long Short-Term Memory) processing two text sequences at the same time. Secondly, in order to overcome the second problem, according to the characteristics of Chinese, we used the Chinese semantic similarity data sets designed by experts to train and test the model, and retained the stop words in the model training process. Finally, the proposed model was compared with the Semantic Text Similarity model based on CNN (Convolution Neural Network) and the Baidu Semantic Text Similarity model. The results show that the model is greater than the previous two in terms of accuracy, recall rate and so on, and the generalization ability is improved also.