Event discovery method and device

Abstract

The embodiment of the invention provides an event discovery method and device. The method comprises the steps of obtaining a representation vector of each news document in a current time period according to all word classes; clustering the representation vectors of the news documents in the current time period according to a clustering algorithm, and forming a news document set for reporting the same event by using the news documents corresponding to the representation vectors belonging to the same cluster; for each event, if it is known that the event is the same as any determined event according to the first similarity between the event and all the determined events, combining the news document set reporting the event with the news document set reporting the determined event. The devicecomprises a document representation module, an event acquisition module and an event merging module. According to the event discovery method and device provided by the embodiment of the invention, thedimensionality of document representation is effectively reduced, the semantic sparsity problem is relieved, and the event discovery efficiency and accuracy can be improved.

Description

technical field [0001] Embodiments of the present invention relate to the field of computer technology, and in particular, to an event discovery method and device. Background technique [0002] Event discovery aims to discover news reports belonging to the same event from the news data stream, and organize the news according to the reported events. With the explosive growth of network news, event discovery has become very important, which can improve the efficiency of users' news reading and facilitate users to read interesting event news. [0003] At present, there have been many works on event discovery, among which the common one is to build an online recognition system to identify new events in news reports, calculate the similarity between news documents in the news data stream and known events, and compare similarity degree and the set similarity threshold to judge whether the news is a new event. At present, the relevant researches on event discovery are basically b...

AI Technical Summary

Problems solved by technology

[0005] The traditional TF-IDF document representation and its improvement, the document representation vector is based on words, the dimension is high, the dimension of the vector can reach tens of thousands or even hundreds of thousands of dimensions, and the semantics are sparse, resulting in low efficiency and accuracy of event discovery. Applicable to large-scale online news event discovery

Images