Patent text retrieval method and device based on multi-modal matrix vector representation

Abstract

The invention discloses a patent text retrieval method and device based on multi-modal matrix vector representation, a word vector set of all words is obtained through training according to an existing patent data set, so that word vectors can contain information of all keywords, an image vectorization representation model is obtained through training of the existing patent data set, and the retrieval efficiency is improved. According to the method, the attached drawings in the patent are extracted, the graph vectors corresponding to the attached drawings are obtained, the graph vectors and the word vectors are combined, when the patent is retrieved, a large amount of useful information contained in the attached drawings in the patent is fully utilized, and meanwhile, some retrieval requirements of searching for a text through a graph, searching for a graph through a text and searching for a graph through a graph in the current market are met; and the patent retrieval result is more accurate.

Description

technical field [0001] The invention relates to the technical field of text retrieval, in particular to a patent text retrieval method and device based on multimodal matrix vector representation. Background technique [0002] Traditional text retrieval is completed through regularization matching. When the user uses synonyms or words with similar meanings to the key words in the document to search, the records will not be retrieved. Moreover, the algorithm based on the LDA topic model trains a large-scale document corpus in an unsupervised manner, so that the topic model of each document can be obtained, so that the retrieval based on the document topic can be completed. [0003] In recent years, semantic retrieval technology based on word vectors has emerged, which can be obtained through unsupervised training of massive texts. word2vec is powerful in capturing lexical relationships between words, but the resulting vectors are largely uninterpretable and difficult to chara...

AI Technical Summary

Problems solved by technology

This kind of vector representation is difficult to accurately represent the scattered information in the document and involves more technologies, and the combination of different word vectors on the same dimension may cause mutual cancellation, and the final vector representation cannot contain all keywords. word information

Images