基于改进编解码器和情感词典的对话生成模型

Dialogue generation model based on improved encoder-decoder and emotion dictionary

针对现有对话模型生成的回复语句缺乏情感共鸣、拟人效果不够理想的问题,提出一种基于改进编解码器和情感词典的对话生成模型。利用AgSeq2Seq模型对语料库进行训练并构建高质量对话生成系统,结合情感词典识别输入语句的情绪特征并计算回复语句的情感值,基于情绪对比机制根据不同的情感特征生成相应的拟人回复。实验结果表明,相对传统的对话生成模型,提出模型可以主动识别用户情绪,生成更加合乎逻辑、适应语境的回复,实现拟人程度更高的情感对话过程。

Aiming at the problem that the response sentences generated using existing dialogue models lack emotional resonance and the personification effect is not ideal, a dialogue generation model based on improved encoder-decoder and emotion dictionary was proposed. The AgSeq2Seq model was used to train the corpus and construct a high-quality dialogue generation system. The emotion dictionary was combined to identify the emotional characteristics of input sentences and calculate the emotional value of reply sentences. Based on the emotion comparison mechanism, the corresponding anthropoid responses were generated according to different emotional characteristics. Experimental results show that compared with the traditional dialogue generation model, the proposed model can actively identify user emotions, generate more logical and context-based responses, and achieve a higher degree of anthropomorphic emotional dialogue process.