A Method for Assessing the Coherence Quality of Multiple Rounds of Dialogue

Abstract

The invention discloses a method for evaluating the coherence quality of multi-round dialogues, which is characterized in that the text of the multi-round dialogues is taken as an input, the hierarchical attention mechanism is adopted, and the semantic information and the intention information of the dialogues are fused at the single discourse level and the whole level of the multi-round dialoguesrespectively, so as to realize the automatic evaluation of the coherence quality of the multi-round dialogues. compared with that prior art, the accuracy of analysis is high, Entities do not need tobe extracted from text, It avoids the propagation of entity extraction error and is especially suitable for large-scale and real-time corpus. Combined with the semantic information and intention information contained in dialogue, it can effectively and automatically evaluate the coherence quality of multi-round dialogue, guide the multi-round dialogue generation system to generate higher quality dialogue text, and promote the better and faster development of dialogue generation system.

Description

technical field [0001] The invention relates to the technical field of Internet deep learning models, in particular to a method for evaluating the coherence quality of multiple rounds of dialogues by using deep learning. Background technique [0002] In recent years, human-computer dialogue systems, such as Apple Siri, Microsoft Xiaoice and other chat / customer service robots (Chatbot), have increasingly appeared in daily life. Dialogue is the basic information interaction method in human social activities, including TV interviews, question-and-answer dialogues, WeChat chats, etc. A key core technology in human-machine dialogue is Multi-turn Dialogue Generation, which is The symbol of the development level of artificial intelligence is also a research hotspot in the field of natural language processing, and has attracted the attention of more and more researchers. The quality evaluation of multi-round dialogue generation usually adopts the following two methods: 1) Borrowing...

AI Technical Summary

Problems solved by technology

[0008] Existing dialogue coherence quality assessment techniques have the following deficiencies: (1) ignoring the natural language diversity and multi-round interaction of dialogue, so the accuracy of dialogue quality assessment is low; (2) lacking the overall Semantic information; (3) lack of intent information contained in multiple rounds of dialogue; (4) lack of deep integration of semantic information and intent information in multiple rounds of dialogue; (5) high labor and time costs, unable to apply to large-scale and real-time dialogues Quality assessment; (6) The method based on entity grid or sequence depends on the performance of entity extraction, and the extraction error will directly affect the performance of subsequent steps

Images