Generative multi-round chat dialogue method and system and computer-readable storage medium

Abstract

The invention discloses a generative multi-round chat dialogue method and system and a computer-readable storage medium. The method is divided into two stages: a stage 1, training a multi-round chat dialogue model by using dialogues of a corpus; and a stage 2, inputting a question to be answered, posed by a user, into the trained multi-round chat dialogue model, and outputting an actual answer. Bymining keywords in historical dialogues and introducing an attention mechanism into the model, all words in the historical dialogues are treated differently, which expands the role of the keywords inthe historical dialogues in generating answers. A width channel is used to predict keywords and broaden topics, a depth channel is used to predict the weights of the keywords in the historical dialogues so as to penetrate into the current topic, and the topic information obtained from the two parts is introduced into a decoder to assist in decoding. The guidance of the topic information effectively solves the problem of meaningless answers, and greatly reduces the number of meaningless answer.

Description

technical field [0001] The invention relates to a generative multi-round chat dialogue method, system and computer-readable storage medium. Background technique [0002] Dialogue systems, also known as intelligent dialogue bodies or chat robots, have been widely used in various life scenarios, including customer service, entertainment chat, emotional escort, knowledge sharing, etc. [0003] In terms of usage, dialogue systems can be roughly divided into chatting dialogue systems and task-oriented dialogue systems. Technically, dialogue systems can be divided into rule-based, retrieval-based, and generation-based dialogue systems. Among them, the rule-based dialogue system requires a lot of human resources to design rules, and the retrieval-based dialogue system relies heavily on the quality of the corpus and size, and rule-based or retrieval-based dialogue systems severely limit the diversity of generated responses. The generation-based dialogue system can generate more fl...

AI Technical Summary

Problems solved by technology

One problem faced by single-turn dialogue systems is that the model ignores the influence of dialogue history on the current reply

[0005] Although a lot of research work has improved the performance of the multi-turn dialogue chat system, it still faces some problems. One of the more prominent problems is that the multi-turn dialogue chat system often generates meaningless responses.

Images