Design And Implementation Of Box Office Forecast System Based On Social Network Analysis

In the 21st century,neural networks,emotional analysis,and other things that have been widely applied to the box office,have improved the accuracy of the film’s box-office forecasting.Due to the complexity and uncertainty of film production,the development trend of movie box office prediction in the future will absorb more data,adopt more fitting model tracking.The existing Product Neural Network(PNN)model has achieved good results in box office prediction,but there are some defects.Aiming at the defects of PNN model,this thesis proposes the following three improved networ-k model architectures.In the stage of feature engineering,PNN model divides all features into domains,and features between domains are mapped to the same vector space by pairwise cross mapping.This kind of method ignores the consistency of feature expression in each domain,in this thesis,the Group Product Neural Network(GPNN)model introduces the Group-wise Sum Embedding concept(GSE),when the feature group is divided,the consistency and difference of expression are considered for all the features in the group.In addition,on the basis of GPNN feature engineering,some cross-continuous features that have great influence on box office prediction are strengthened,regarded as Continuous Group Product Neural Network(CGPNN).In the stage of feature crossover,GPNN and CGPNN models adopt the cross product operator to realize feature crossover,map features to the higher-dimensional vector space,and learn the deep expression of features more comprehensively.However,it is difficult for the model to converge under the condition of insufficient data.Aiming at the characteristics of these two network structures,a Probe-based CGPNN model was proposed in this thesis.The intersecting of features is accomplished by inner product operator,and the time and space complexity is reduced reasonably on the premise that the prediction effect is not decreased obviously.Based on the three kinds of improved deep learning network architecture above,learnt from the structure of the existing box office prediction system,this thesis designed and implemented the box office forecast system based on social network analysis,the whole system is divided into six main functional modules,including web crawler submodule,data storage module,text emotion analysis submodule,feature combination submodule,movie box office prediction submodule,and data front-end display module,its internal implementation are elaborated in detail.