Patent duplicate checking method and system based on complex network text language intention coding mode

Abstract

The invention discloses a patent duplicate checking method and a patent duplicate checking system based on a complex network text language intention coding mode. The method comprises the following steps of preprocessing a public data set obtained from the Internet into language material data; inputting the corpus data into a Word2Vector model for training and testing to generate a word vector model; respectively inputting the to-be-queried duplicate long text and the public long text into the word vector model to obtain a to-be-queried duplicate long text word vector and a public long text word vector; constructing a multi-dimensional complex directed graph of the long text to be subjected to duplicate checking and a multi-dimensional complex directed graph of the public long text; obtaining the tensor of the long text to be subjected to duplicate checking and the tensor of the public long text; and judging the text similarity degree by calculating the similarity between the to-be-checked duplicate long text tensor and the public long text tensor. The complex web text language intention coding mode provided by the invention can fully represent the relationship between characters and words in the document and the weight of the relationship, and is clear in expressive meaning, definite in physical significance and easy to calculate.

Description

technical field [0001] The present invention relates to the technical field of text similarity matching, in particular to a method and system for patent plagiarism checking based on complex network text language intent coding. Background technique [0002] Text matching mainly studies the similarity between two pieces of text. The problem of similarity includes two layers: one is how to represent two pieces of text so that the computer can easily process it; the other is how to define similarity as an optimization goal, such as semantic matching similarity, click relationship similarity, user behavior similarity, etc. Business scenarios are closely related. [0003] For the representation of text, the current mainstream method mainly stays at the word level, and the relationship between words in the entire text is less considered. Around 1970, the mainstream representation methods were vector space algorithms such as TF-IDF and BM25. This type of method is to express the w...

AI Technical Summary

Problems solved by technology

However, the way long texts are encoded has so far lacked a pleasing way of encoding

One is to consider the co-occurrence frequency of long texts and words, completely ignoring the relationship between words, but this is obviously unreasonable; the other is to consider the probability model of N-grams, but once too many contexts are considered, the calculation is particularly difficult It is complex, and it is difficult to associate full-text information, with obvious shortcomings; the third is the encoding method of Sequence to Sequence training based on deep learning, which requires a large amount of training data, not to mention its physical meaning is not clear, and its accuracy cannot be trusted. The scene is difficult to apply; the fourth is some coding methods, because the length of the document is different, and the text concept is written differently, which leads to the problem that the coding needs to be aligned, which is also difficult to solve

Images