An Incremental Social Event Detection Method Based on Graph Neural Network

Abstract

The invention discloses an incremental social event detection method of a graph neural network, which extracts information in a text to model a heterogeneous information network in the face of streaming incoming social network data; acquires a homogeneous network; obtains homogeneous social information Graph; use the graph attention model to learn the isomorphic message graph to obtain message encoding based on knowledge retention increment; at the same time, sample the message encoding for comparative learning to calculate the loss, and adjust the graph attention model according to the returned loss Parameters to train the graph attention model; the coding obtained by detecting the graph attention model is clustered to obtain social events. The present invention fully integrates rich semantic and structural information into social messages, retains the knowledge acquired from social messages through the graph neural network, adopts comparative learning technology, and regularly maintains message graphs, which can be achieved without consuming too many resources. Improve the accuracy of event detection.

Description

technical field [0001] The invention belongs to the technical field of social network event detection, and in particular relates to an incremental social event detection method of a graph neural network. Background technique [0002] With the development of Internet technology, global informatization data presents the characteristics of explosive growth, massive accumulation, and rapid dissemination. Human society has entered the "big data era", which has had a major impact on cultural communication and social governance. From massive data The technology of detecting social events in China has attracted more and more attention, and has become a hot spot at present. Event detection refers to the technology of mining events in real society by analyzing social network data. Compared with traditional text, social messages are generated by time and streamed; the content is short and often contains acronyms that are not in the dictionary; it contains various types of elements, in...

AI Technical Summary

Problems solved by technology

[0003] The complexity and streaming characteristics of social messages have brought great challenges to traditional event detection technology, mainly facing the following problems: real-time data transmission, difficulty in short text representation, and huge amount of text

Many Document-Pivot-based methods are difficult to effectively use the information in social messages when calculating text similarity, ignoring its hidden structural information, and this shortcoming is more obvious in the short text environment

Many Feature-Pivot-based methods cannot effectively detect "fermentation events", and can only capture the event outbreak stage, that is, as time goes by, the corresponding social messages gradually increase, which weakens the event detection effect

Images