Method and device for determining semantic similarity

Abstract

The invention provides a semantic similarity determination method. The method comprises the steps of obtaining a vector of a first target text and a vector of a second target text; respectively inputting the obtained vector of the first target text and the vector of the second target text into a recurrent neural network for encoding to obtain representation features of the first target text and representation features of the second target text; obtaining a similarity weight matrix of the first target text and a similarity weight matrix of the second target text according to the obtained representation features of the first target text and the representation features of the second target text; and determining the semantic similarity of the first target text and the second target text according to the obtained similarity weight matrix of the first target text and the similarity weight matrix of the second target text. Multi-dimensional semantic features can be obtained, and the accuracyof semantic analysis is improved.

Description

technical field [0001] This article relates to computer technology, especially a method and device for determining semantic similarity. Background technique [0002] With the rapid development of the Internet and artificial intelligence, various interactive human-computer dialogue systems have emerged, and at the same time, the number of searches by users has also increased significantly. For the above applications, how to accurately retrieve sentences that meet the user's intentions from the knowledge base or corpus is very critical, which directly affects the user experience. In response to this problem, this patent proposes a semantic similarity calculation method based on neural network and attention mechanism to obtain the semantic similarity between sentences with the same meaning but different expressions, which can be used for human-computer dialogue and search engine correlation. Application scenarios provide algorithmic support. [0003] The existing main methods...

AI Technical Summary

Problems solved by technology

[0003] The existing main methods include other neural network models and models based on pre-training. Other neural network models such as matchPyramid, ESIM, etc. are also neural network-based models, but there are still deficiencies, such as matchPyramid, which compares text data to images. From the point of view, operations such as convolution are used for processing, but after all, there are differences between the two, and sufficient features cannot be proposed; at the same time, although the ESIM model also uses a bidirectional LSTM network for encoding, it only calculates simple attention weight, and cannot represent comprehensive semantic features

Moreover, the above models all consider the position information of the sentence, and the feature extraction is insufficient.

[0004] Based on the pre-trained model, generally speaking, the matching accuracy is relatively high, but due to too many model parameters, a forward calculation takes a long time, which does not meet the needs of practical applications

Images