融合指针网络的新闻文本摘要模型

本文针对实际党建领域中的新闻标题进行自动生成,提出了一种融合指针网络的自动文本摘要模型-Tri-PCN.相比于传统基于编码器-解码器框架的自动文本摘要模型,党建新闻标题生成模型还需要满足(1)从更长的文本序列提取特征;(2)保留关键的党建信息.针对党建新闻比普通文本摘要任务面临更长文本序列问题,论文使用Transformer模型在解码阶段提取多层次全局文本特征.针对党建新闻标题生成过程中需要保留关键的党建信息,论文引入指针生成网络模型的复制机制在新闻标题生成时可以直接从新闻文本中复制关键词信息.实验采用ROUGE值作为评测指标,结果表明本文提出的Tri-PCN模型在党建新闻领域自动文本摘要任务上效果明显优于基准模型,比其他模型具有更好的效果.

News Text Summarization Model Integrating Pointer Network

In this paper,the headline of party building news in the field of actual party building is automatically generated,and an automatic text summarization model-Tri-PCN is proposed.Compared with the traditional automatic text summarization model based on the encoder-decoder framework,the party-building news headline generation model also needs to satisfy;(1) extract features from morelonger text sequences;(2) retain key party-building information.Aiming at the problem that the party construction news faces a longer text sequence than the ordinary text summary task,the paper uses the Transformer model to extract global features of multi-level text during the decoding stage.In view of the need to retain key party building information during the process of generating party news headlines,the paper introduces a replication mechanism of pointer generation network model to effectively copy key information directly from news texts when generating news headlines.The experiment uses the ROUGE value as the evaluation index.The results show that the Tri-PCN model proposed in this paper is significantly better than the benchmark model in the task of automatic summary of news headlines in the party building field,and has better results than other models.