Context Multi-information Fusion Network Model Based on Human-machine Dialogue Context Multi-information Fusion Network

Abstract

With the continuous deepening of human-computer interaction technology, the related technologies of human-computer dialogue have also been developed rapidly. At the same time, which dialogue matching based on retrieval has been used on a large scale. Although generative dialogue is also developing continuously, generative dialogue Dialogue is extremely unstable and it is easy to generate meaningless sentences. Therefore, retrieval dialogue is takes the lead in the research of human-computer dialogue practice, and the fusion of contextual semantic information is particularly important and challenging. So, in order to solve the difficulty of capturing context semantics in retrieval dialogue, we proposes a recurrent convolutional neural network model (MFRCNN) for context multi-information fusion, and we tested our model on the text dialogue question answer matching data, the experimental results show that the model based on our proposed has stronger adaptability than the previous model and accuracy. For contextual information fusion semantic representation performance, the MFRCNN model is stronger than the RCNN model, with 1.4% higher accuracy and 0.12 higher F1-score. From the comparison of the experimental results of all models, we can conclude that the MFRCNN network proposed in this paper has the best effect among all the compared models, with an accuracy rate of 80.2% and an F1-score of 0.816.