A variational selection mechanism for article comment generation

Jiamiao Liu

doi:10.1016/j.eswa.2023.121263

2024-03-01

PRE

DOI：10.1016/j.eswa.2023.121263

原文链接

原文求助

摘要

En 中文

Article comment generation is a new and challenging task in natural language processing, which has recently received much attention from researchers. In the article comment generation, there are apparent distinctions and different perspectives on the comments of a single article. However, current researches ignore the one-to-many relationship between articles and comments, resulting in a lack of diversity and coherence in generated com-ments. To solve this problem, a variational selection mechanism model (VSMM) is proposed in our research. In this model, we construct a Gaussian mixture prior network to capture a richer latent space and generate comments with more diversity and informativeness. At the same time, VSMM maps latent variables into different semantic spaces through the selection mechanism to capture one-to-many relationships. Then we introduce a discrimi-nator to distinguish whether the selected content is consistent with the reference comment content, thus improving the coherence of the generated comments. In addition, a hierarchical encoder with attention is introduced in the VSMM model, which can effectively solve the problem of long document encoding. Further-more, we propose a multi-category article comment dataset to align closely with practical applications. Exper-iments on three datasets demonstrate that VSMM outperforms existing state-of-the-art comment generation methods in terms of diversity in single and multiple comment generations. Moreover, VSMM can generate fluent, diverse, and coherent comments on multi-category and topic-rich datasets.

Keyword:

Article comment generation

Variational selection mechanism

Recurrent neural network

Hierarchical encoding