Improved multi-round dialogue model based on sequential matching network

Abstract

Multiple rounds of conversations involved in chat robots and intelligent customer services are hotspots of current researches. In multiple rounds of conversations based on a retrieval method, a sequence matching (SMN) model is representative, the model uses a single-layer GRU network in a conversation reply matching part, but the single-layer GRU network has limited capability of extracting deep features, and the obtained coded information contains some noises. A dialogue matching part of the model uses a CNN (Convolutional Neural Network), and the CNN mainly pays attention to local information, so that the capability of extracting overall semantic information of a natural language sequence is limited, and information obtained after the information passes through the CNN is incomplete. The invention relates to an optimal matching algorithm for a multi-round dialogue model sequential matching network. The method comprises the following steps: (1) changing a single-layer GRU network into a multi-layer deep network; and (2) advancing the aggregation operation of the feature matrixes M1 and M2. And (3) replacing the CNN convolutional network with a GRU network. And (4) the accuracy of the improved SMN network is improved by about 2%.

Description

technical field [0001] The invention relates to the field of natural language processing, that is, an algorithm for selecting an optimal answer for an answer selection model in a multi-round dialogue system. Background technique [0002] In recent years, with the popularity of artificial intelligence, chatbots and intelligent customer service have also been widely used. Among them, how to obtain accurate answers is a research hotspot. The conversation between chatbots and intelligent customer service is a multi-round dialogue process. A round of dialogue should not only consider the question information, but also need to pay attention to the context of the dialogue, because the context can provide a lot of useful information and play an important role in building a coherent dialogue. A more representative model for retrieval-based methods in multi-turn dialogue is the Sequential Matching (SMN) model. The model consists of three parts: UtteranceResponse Matching, MatchingAcc...

AI Technical Summary

Problems solved by technology

Due to the limited ability of the single-layer GRU network to extract deep-level features, the resulting encoded information will contain some noise (useless semantic information)

Moreover, the dialog matching part of the model uses a convolutional neural network to extract deeper matching information from the matching matrix at the word and sentence levels. However, the convolutional neural network mainly focuses on local information. For the overall semantic information of time series sequences such as natural language The ability to extract is limited, and some information will be lost

This will lead to incomplete matching information contained in the matching vector generated by the CNN network

Images