硬约束限制的情感文本生成方法研究

预训练语言模型在情感文本的生成任务中取得了良好效果，但现有情感文本生成方法多使用软约束的方式控制文本整体的情感属性，缺乏单词和短语级别的硬性控制。为解决以上问题，提出硬约束限制下的情感文本生成方法。首先使用方面情感分析技术提取句子的方面词、情感词并判断情感极性；之后，选择目标情感的方面词和情感词作为预训练语言模型的硬约束输入来重建完整句子，其中，设计了一种新的单词权重计算方法，旨在使模型优先生成重要单词。实验结果表明，该方法生成的句子不仅具有方面级情感，在文本质量和多样性的评价指标上也有显著提高。

Affective text generation with hard constraints

Pre-trained language models have achieved good results in the task of sentiment text generation, but existing sentiment text generation methods mostly use soft constraints to control the sentiment attributes of the text as a whole and lack hard control at the word and phrase level. To solve the above problems, this paper proposed sentiment text generation methods under hard constraints. Firstly, it extracted aspect and sentiment words of the sentence and judged sentiment polarity using aspect sentiment analysis techniques. After that, it selected the aspect and sentiment words of the target sentiment as hard-constrained inputs to the pre-trained language model to reconstruct the complete sentence, in which designed a new word weight calculation method, aiming to make the model generate important words in priority. The experimental results show that the sentences generated by this method not only have aspect-level sentiment, but also have significant improvement in the evaluation indexes of text quality and diversity.