A Continuous Dynamic Prediction Method for Microblog Event Information Dissemination

Abstract

The invention discloses a continuous dynamic prediction method for microblog event information dissemination, which belongs to the field of data mining. For Sina Weibo, on the basis of the current dissemination information, try to predict the total number of Weibo in the next stage; divide the event propagation by hour, and use the propagation characteristics of the event from occurrence to the current time period, such as the number of Weibo, participation The number of people, Weibo sentiment, etc., based on the GBDT model to predict the total number of event Weibo transmissions in the next hour. In the prediction model of the present invention, the optimal time segment length and microblog feature combination are screened out on the basis of comprehensively measuring the contribution and correlation of each feature, which can not only effectively improve the prediction accuracy of the model, but the average model accuracy exceeds 70%, It can also reduce computational complexity, avoid useless calculations, and effectively support early warning and intervention measures for events.

Description

technical field [0001] The invention belongs to the field of data mining, and relates to a continuous dynamic prediction method for microblog event information dissemination. Background technique [0002] In recent years, with the extensive penetration and innovative development of Internet technology, social media represented by Weibo has been widely and deeply integrated into every aspect of people's lives. Social media has become an important platform for people to find information, express opinions and communicate. [0003] Research based on Twitter shows that the attributes of social media are closer to event networks than to social attributes. The diversity of information sources in social networks, the suddenness of events, and the extensiveness of dissemination make event analysis and dissemination prediction widely used, such as politically sensitive event monitoring, news hotspot discovery, business public opinion analysis, stock market public opinion fluctuations...

AI Technical Summary

Problems solved by technology

However, these predictive models are all evaluated for a single microblog, and the correlation and collinearity of features are not fully considered when the model considers the dependent variable.

[0006] Although there are also a lot of research work that transforms the problem of social network communication prediction into a binary or multi-classification problem, by extracting the context information, content features, communication network and other features in the process of microblog communication, and analyzing them according to the amount of microblog communication Popularity classification, such as 0 is a class, 1-100 is a class, 100-10000 is a class, and more than 10000 is a class, and a classifier based on a logistic regression model or a classification tree is constructed for training and learning; however, this classification or The regression method can only give a very rough prediction of the spread scale for a single microblog, and it cannot be applied in practice.

[0007] Studies have shown that it is still quite difficult to predict the final scale and development trend based on given certain dissemination information, or a large amount of background information and perfect dissemination information are required, which is difficult to achieve in practical applications

Images