Text matching method and text matching system

Abstract

The embodiment of the invention provides a text matching method and a text matching system. The text matching method comprises the following steps: respectively acquiring a first text vector corresponding to a target text and a second text vector corresponding to a source text; calculating cosine similarity of the first text vector and the second text vector; screening the source texts according to the cosine similarity, generating a text set by using the screened source texts, and representing each candidate text in the text set by using a third text vector; and respectively calculating various similarities between each third text vector and the first text vector and the fusion similarity of the various similarities, and selecting the candidate text corresponding to the third text vectorwith the maximum fusion similarity as the prediction text most matched with the target text. According to the method, the cosine similarity of the text vectors of the target text and the source text is calculated, the text set is generated according to the cosine similarity, and the prediction text is selected by fusing the similarity, so that the matching steps and the calculation amount are reduced, and the matching efficiency and precision are improved.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a text matching method and a text matching system. Background technique [0002] Text matching is an important natural language technology. Many tasks can be abstracted as text matching problems. For example, in question answering systems and search engines, there are a large number of scenarios that need to match similar or even identical texts in multiple source texts according to the target text. With the rapid growth of text data sets, the computational complexity of text matching algorithms is also increasing. However, in practical application scenarios, computing resources are limited, and it is often difficult to balance the accuracy and real-time performance of text matching algorithms. [0003] Existing text matching algorithms such as deep text matching algorithms can be divided into four types: single semantic model, multi-semantic model, matching matrix model and d...

AI Technical Summary

Problems solved by technology

[0004] The single semantic model uses multi-class perceptron (MLP), convolutional neural network (CNN) or recurrent neural network (RNN) to encode sentences into feature vectors, and then calculates the matching degree between two sentences based on feature vectors, without considering The local structure of similar phrases in the sentence; the multi-semantic model interprets the sentence from the perspective of multi-granularity (sub, word, phrase and long sentence), but the calculation is complicated; the matching matrix model takes more into account the interaction of different words between the sentences to be matched , to calculate the matching degree between two pairs, and then use the deep network to extract features. The connection between words, the calculation steps are more, the efficiency is low

[0005] Therefore, the inventor believes that the above-mentioned text matching method has great limitations. The model based on deep learning has complex models, difficult parameter adjustment, is not portable enough for deployment, and has generalization ability when the amount of labeled data is limited. , the problem of low matching efficiency

Images