Research on Chinese Text Summarization Based on Core Word Attention Mechanism

The neural sequence-to-sequence model provides a feasible new method for abstract text summarization.Because the traditional Seq2Seq neural network model only judges the importance of each vocabulary based on the word meaning information of the vocabulary itself and the current input of the decoder, they can easily reproduce the factual details inaccurately and often repeat themselves. However, we know that the more relevant a certain vocabulary is to the core words of the original text, the higher its importance will be. Therefore, this article proposes a summary generation that integrates the core word attention mechanism based on the Seq2Seq model based on the attention and coverage mechanism. The model (Fusion Core Word Attention Mechanism Model, FCWAM) integrates the core word information of the original vocabulary, that is, the probability information of the word being assigned to each core word of the original text, as a kind of prior information into the traditional attention mechanism, thereby Improve the decoder’s attention to the important information of the original text, and finally use the pointer generation network to alleviate the problem of unregistered words that may appear in the model. The experimental results based on the LCSTS data set show that the addition of core word information improves the accuracy of the generated abstract