Method and device for obtaining text semantic similarity value, storage medium and electronic equipment

Abstract

The embodiment of the invention discloses a method for obtaining a text semantic similarity value. The method comprises the steps of inputting at least two to-be-detected texts and similarity parameters between the at least two to-be-detected texts into a bidirectional encoder representation BERT model from a transformer; generating a query vector and a key value vector pair based on the at least two to-be-detected texts; generating dot product data based on the query vector, the key value vector pair and the similar parameters; and performing similarity processing on the dot product data to obtain a similarity value between the at least two to-be-detected texts. According to the method, when the similarity value between the at least two to-be-detected texts is obtained, the model effect can be improved on the premise of not increasing the training duration and memory loss.

Description

technical field [0001] The present application relates to the technical field of natural language learning, and in particular to a method, device, storage medium and electronic equipment for acquiring text semantic similarity values. Background technique [0002] Transformer's Bidirectional Encoder Representation from Transformers (Bidirectional Encoder Representation from Transformers, BERT) model, as the best deep language model at present, is one of the most mainstream language models for future Natural Language Process (NLP) research and industrial applications. In order to enable the BERT model to complete its own masked word (Masked Language Mode, MLM) task and predict the next sentence (NextSentence Prediction, NSP) task, it also has other functions such as text semantic similarity tasks. fine-tuning. In the existing technology, the BERT model is usually fine-tuned by adding other unsupervised pre-training models to the BERT model, such as semantic role prediction mo...

AI Technical Summary

Problems solved by technology

In the existing technology, the BERT model is usually fine-tuned by adding other unsupervised pre-training models to the BERT model, such as semantic role prediction models, entity recognition models, etc., but these pre-training models will increase the training time and time of the BERT model. training cost

Images