Method for realizing fast-speed short text bi-cluster

Abstract

The invention relates to a method for realizing fast-speed short text bi-cluster. The method comprises the following steps of: (1) preprocessing short text disturbance items, and carrying out fast-speed unrelated-language and word-class recognition and processing recognition on short texts with the support of an unrelated-language dictionary and a word-class dictionary; (2) calculating the similarity of two preprocessed short texts to form a short text similarity sparse matrix; (3) carrying out short text first-level clustering on the short text similarity sparse matrix, and dividing similar short texts into clusters one by one according to the calculation result of the short text similarity; and (4) carrying out second-level clustering on the basis of the result of the first-level clustering.

Description

technical field [0001] The invention relates to natural language processing in the field of artificial intelligence computers, in particular to a fast short text bi-clustering method and its realization by using natural language processing and data clustering. Background technique [0002] In a large number of natural language applications, there is a basic and common problem: for a corpus composed of short texts (hereinafter referred to as short text corpus or corpus), how to organize the short texts according to a certain similarity clustered into different classes. [0003] Generally speaking, the basic idea of ​​text clustering is to cluster "similar" texts into a class; in this class, the "differences" between texts are small. Texts that are not "similar" are clustered into other classes. The "gap" between different classes is large. Here, "similarity" / "gap" is a measure between some texts, which depends on different application requirements. There are many traditio...

AI Technical Summary

Problems solved by technology

[0006] (2) Accurate calculation of short text similarity

At present, although there are many similarity algorithms (such as Euclidean distance method, cos distance method, Pearson coefficient method, VDM method, etc.), according to our research, they all have defects, and the effect is not good in practical applications.

[0007] (3) Fast and accurate clustering of short texts

Traditional single clustering (such as K nearest neighbor method, hierarchical clustering method, etc.) is difficult to achieve accurate clustering. When facing open corpus, the clustering accuracy is generally very low, which cannot meet the needs of practical applications.

Moreover, when the length of the short text is slightly higher, the clustering accuracy is lower

Images