Multi-index dialogue data cleaning model
Xianxin Ke, Jiaojiao Bai, Wen Lei, Bin Cao · 2019
We proposed a multi-index dialogue data cleaning model. Our model has two distinctive characteristics: (1) it fuses multiple indices to build the model; (2) it applies two attention mechanisms when dealing with information of questions and answers, making the model's cleaning process more in line with human thinking process. The comparison experiment with the dual encoder model shows that the proposed architecture has higher accuracy. Moreover, the weight of the attention mechanism is in line with the human intuitive understanding and can highlight the key information of the sentence.