Method and system for analyzing events on basis of non-negative matrix factorization

Abstract

The invention provides a method and a system for analyzing events on the basis of non-negative matrix factorization. The method includes acquiring to-be-processed data of at least one data text; respectively segmenting words of each data text to obtain text space matrixes corresponding to the to-be-processed data; performing non-negative matrix factorization ion the text space matrixes, determining various events and keywords in the to-be-processed data according to basis matrixes, and determining data texts according to coefficient matrixes. The basis matrixes are obtained by means of factorization. The keywords are respectively used for describing the various events. The coefficient matrixes are obtained by means of factorization. The determined data texts respectively correspond to the various events. The method and the system have the advantages that the text space matrixes of the to-be-processed data are constructed and are subjected to non-negative matrix factorization, so that each matrix on a huge scale can be factorized to obtain two corresponding matrixes on small scales, the non-negativity of matrix elements can be guaranteed before and after the matrixes are factorized, the accuracy of event mining results can be guaranteed, the events in the to-be-processed data can be found by means of reducing dimensions, computation is simple and convenient, and the method and the system are good in expansibility.

Description

technical field [0001] The invention belongs to the technical field of data mining, and in particular relates to an event analysis method and system based on non-negative matrix decomposition. Background technique [0002] With the vigorous development of Internet technology, more and more users publish various news or express personal opinions on some social phenomena through social network platforms such as forums and Weibo, which leads to the emergence of various data information on the Internet. With explosive growth, how to effectively mine massive amounts of data information is a major research issue for various search engines. [0003] An existing method of data mining is to use hierarchical clustering to decompose a given set of data objects hierarchically until a certain cut-off condition is met. Specifically, it can be divided into: agglomerative hierarchical clustering: a bottom-up strategy that first treats each data object as an atomic cluster, and then merges ...

AI Technical Summary

Problems solved by technology

[0005] In view of the above-mentioned problems, the present invention provides an event analysis method and system based on non-negative matrix factorization to overcome the problem of high computational complexity and poor scalability caused by the hierarchical clustering method in the prior art defect

Images