Dialogue Understanding and Generation of Sequence Template and Path Retrieval Based on Knowledge Enhancement

With the rapid development of deep learning technology, pre-trained language models have made breakthrough achievements in the field of dialogue systems, and these models have demonstrated their excellent ability to capture complex contextual relationships and semantic features. However, they still face challenges in understanding entities in cross-domain data; The sentences generated by the current dialogue model have some fluency, but the generated statements contain the problem of dialogue illusion. To solve these problems, this paper proposes a dialogue understanding and generation method of sequence template and path retrieval based on knowledge enhancement. The dialogue understanding part first selects the knowledge related to the text from the external knowledge, and constructs the knowledge into the graph and the text content as the input of the model. The attention mechanism is also used to evaluate knowledge. Finally, by using knowledge, lexeme and its lexeme category to build a generative template, the model can predict the lexeme entity on the template. The dialogue generation part mainly constructs the dialogue as a graph structure, learns the relationship information between nodes, enhances the use of knowledge, injects knowledge and dialogue to generate dialogue, and re-identifies and retrieves the knowledge in the generated dialogue. The implementation of this method enables the model to show better adaptability and superior performance in processing natural language processing tasks.