Semantic similarity matching method and matching device based on cross-attention mechanism

Abstract

The invention provides a semantic similarity matching method, device, computer equipment and storage medium based on a cross-attention mechanism, which are suitable for the technical field of voice interaction and can realize cross-representation of two sentences at the semantic level. The present invention firstly uses word2vec to characterize each word segment in two sentences, obtains two splicing matrices through bidirectional LSTM respectively, and then cross-represents the two splicing matrices to each other, and obtains each splicing matrix in any sentence The importance of a participle relative to another sentence. On this basis, the maximization process is carried out and input to the fully connected layer, and finally the matching score between the two sentences is obtained. The above solution proposed by the present invention overcomes the limitations existing in the prior art when using LSTM alone or interactive matching, so that the calculation of the matching degree between two sentences is more accurate and complete, and approaches the real situation.

Description

technical field [0001] The invention relates to the technical field of voice interaction, in particular to a semantic similarity matching method, device, computer equipment and storage medium based on a cross-attention mechanism. Background technique [0002] The currently recognized semantic similarity matching methods based on deep learning include: 1) Siamese structure, that is, first two sentences or texts are represented by convolutional neural network (CNN), LSTM and other neural networks to obtain two sentence vectors, and then Carry out similarity calculation; 2) The method of interactive matching, that is, first perform an inner product operation between the word vectors of two sentences to obtain a three-dimensional matrix, and then input it into neural networks such as CNN and LSTM. However, the Siamese structure only independently represents the two sentences, and fails to accurately represent the interaction between the two sentences. On the other hand, the int...

AI Technical Summary

Problems solved by technology

However, the Siamese structure only independently represents the two sentences, and fails to accurately represent the interaction between the two sentences.

On the other hand, the interactive matching method only considers the point-to-point inner product operation, which can only express the local correlation between two sentences, and cannot effectively represent the semantic relationship

Images