| 摘 要: 现有文本聊天机器人在对话过程中存在单一匹配传统知识库生成回复语句和采用通用性情感回复语句的问题。文章研究的系统将解决上述问题作为切入点，在基于Seq2Seq(Sequence-To-Sequence，序列到序列)模型的基础上引入注意力机制，以产生贴合实际的生成式回复，避免单一匹配知识库生成回复语句。构建TextCNN-BiLSTMSelfAttention情感分类模型，获取对话文本的情感特征和情感类别，并进一步在对话过程中引入情感监督实现对话过程中的情感响应与回复，从而避免产生通用性情感回复语句。结果表明，该系统有效提高了文本聊天机器人的回复语句质量。 |
| 关键词: 情感监督；Seq2Seq；注意力机制；深度学习 |
|
中图分类号: TP311.1
文献标识码: A
|
|
| Design and Implementation of Text Sentiment Chatbot System based on Deep Learning |
|
SHANGGUAN Xin, LV Junyu, ZHANG Huanyu, LIU Lijun
|
(School of Jinshen, Nanjing Audit University, Nanjing 210023, China )
phare_wy2023@163.com; 18851828906@163.com; 929143401@qq.com; 79471622@qq.com
|
| Abstract: The existing text chatbots have the problems of generating reply statements by matching single traditional knowledge base and adopting universal emotion reply statements in the process of dialogue. Aiming at these problems, this paper proposes a system and introduces the attention mechanism based on the Seq2Seq (Sequence-To-Sequence) model, so as to generate the actual generative response and avoid generating reply statements from a single matching knowledge base. The TextCNN-BiLSTM-SelfAttention sentiment classification model is constructed to obtain sentiment features and categories of the dialogue text, and sentiment supervision is further introduced in the dialogue process to realize the sentiment response and reply, so as to avoid the universal sentiment response statements. The results show that the proposed system effectively improves the quality of the text chatbots' reply statements. |
| Keywords: sentiment supervision; Seq2Seq; attention mechanism; deep learning |
