An Encoding Mechanism for Seq2Seq based Multi-Turn Sentimental Dialogue Generation Model
Zhiqiang Ma, Baoxiang Du, Ji Shen, Rui Yang, Jianxiong Wan · Procedia Computer Science · 2020
For the multi-turn dialogue generation task, Seq2Seq model lacks representation of sentimental information and dialogue context in the encoder, resulting in generating sentimentally poor and contextually independent responses. To address these problems, in this paper, we propose an encoding mechanism for Seq2Seq based multi-turn sentimental dialogue generation model. To make the Seq2Seq model contain sentiment, we build sentimental word vectors and sentimental sentence vectors to strengthen the sentimental information representation. To make the generated responses more context sensitive, we build semantic vector and contextual vector. Experiments show that Seq2Seq based multi-turn sentimental dialogue generation model with our proposed methods can generate more sentimental, contextually relevant and high-quality responses.