Automatic Generation of News Commentary on Social Media Based on Self-Attention Mechanism

Abstract

With development for social media, the technology of automatic comment generation for social media news is expected to generate great social and commercial value; so, this technology has attracted more and more attention from industry and academia. The automatic generation of social media news comments refers to the use of current popular technologies such as deep learning, NPL, and data mining to build high-performance algorithms to give machines the same ability to understand and express language as humans and to imitate human language habits to comment on social media news. Automatic generation of social media news comments is a challenging and understudied task in natural language processing. This work addresses the problem that social media news texts are too long and have complex contextual information, and existing work does not model the content structure of news well and proposes a topic encoding model. First, topic extraction and construction of news texts are carried out using the Latent Dirichlet Allocation (LDA) topic model to convert unstructured news texts into topic-sentence pairs. Second, a topic encoder is designed based on the self-attention mechanism, and topic-sentence pairs are input into the topic encoder. It goes through the sequence embedding module, the sentence encoding module, and the topic encoding module in turn to obtain the topic hidden state sequence. Finally, this paper designs a topic attention mechanism to generate diverse comments by giving different attention to each topic in the decoding part. Comprehensive and systematic experiments verify the effectiveness of this work and improve the performance of automatic generation of social media news comments.