Research On Popularity Prediction Based On Social Network Hot Event Database

The popularity of social network information is a quantitative index of the attention degree of hot events.The majority of network users release and forward information through online social networks,express their positions and attitudes towards social hot events,and promote the spread and popularity of information.Popularity can reflect the occurrence and development of public opinion on the Internet.Online public opinions can influence the thoughts and behaviors of the majority of network users,and may trigger public security incidents.Today,with the rapid development of social network analysis demand,research on popularity prediction method is helpful to improve the prediction ability of online public opinion,which is of great practical significance for achieving a good public opinion atmosphere and guaranteeing social public security.However,the application of the existing popularity prediction methods in public opinion governance still faces some challenges,including the data representation and organization problems brought by the huge and complicated big data of social networks for the application of upper popularity analysis,and the limitations caused by the insufficient consideration of the unique evolution rules and influencing factors of social networks in the existing popularity evolution prediction research.This article revolves "popularity of evolution prediction oriented social network data representation and organization problem","peak time of popularity evolution prediction problem" and "popularity of multi-factor index prediction problem" three key problems,from popularity analysis and prediction oriented social network hot event database mode,prediction for peak time popularity based on social network hashtag,prediction for popularity based on social network event database multi-factor coupling and implementation and application of analysis and prediction platform of popularity of four aspects to carry out the study,the main work and contributions are as follows:Firstly,a social network hot event database model is built to solve the data expression and organization problems in the popularity evolution trend of analyzing and predicting information with massive and complicated big data of online social networks.The multidimensionality of big data in social network is analyzed,and a data model is established,including defined entities and attributes,data constraints,constraint checking and query.On this basis,a time series extraction method based on social network hot event database is designed.Experiments show that the data constraint check method in the social network hot event database model has good performance,and the proposed time series extraction method based on the hot event database has better performance in terms of accuracy rate and recall rate compared with the manual extraction method,and can further improve performance by enabling the constraint check method.Secondly,aiming at the evolution analysis and prediction of information popularity based on historical popularity index,the evolution law of popularity was analyzed based on real online social network data,and a peak prediction method of popularity evolution was proposed.An empirical study of the Twitter data set shows that popularity has generally peaked at an early stage of evolution since the beginning of evolution.On this basis,the popularity peak prediction method is proposed,three types of data resources are comprehensively utilized,multi-dimensional matrix transformation is carried out by LSTM,DeepWalk and other embedded algorithms,the average pool layer is used for feature representation,internal attention and mutual attention are learned,and the final output is formed by input to the neural network nonlinear layer,and the peak time of popularity evolution is predicted.Experiments show that the median absolute error of the method designed in this paper is lower than that of the baseline methods such as NAM,SVR,SpikeM and BLR.This not only proves the effectiveness of multi-mode deep learning to learn advanced features,but also proves that the prediction method designed in this paper has better prediction effect.Thirdly,aiming at the problem of popularity evolution analysis and prediction based on multi-factor index,a method of popularity prediction based on multi-factor coupling of event database is proposed by using event database to obtain multi-factor index.By using the unified storage of event database for social network data,the indexes of each factor are extracted from the heterogeneous data of multiple sources.On this basis,a grouping embedding method is proposed.The embedding method based on deep learning provides the possibility for dimension reduction and fusion of time series data.According to their physical meaning and characteristics,the factors are grouped into cumulative factors and intrinsic factors.Then,different neural networks are embedded to obtain the status representation of these factors,and on this basis,a prediction method is proposed.The method proposed in this study achieves innovative design in the aspects of factor selection,definition of factor,grouping of factor,acquisition of factor index data and comprehensive utilization of index data.The experimental results show that the proposed prediction method is superior to the existing deep neural network model,support vector regression model and SH popularity prediction model.Finally,a platform for popularity analysis and prediction is designed and implemented.Using the model and method proposed in this paper,the function of event prevalence analysis and prediction is realized.The case verifies the research results of this paper and achieves good application effect.