A Study Of Prediction Model System Of Electronic Sports Based On Word Embedding And Machine Learning

With the rapid development of E-sports in the world,the Chinese gaming industry has also been injected fresh blood,capital and talent.Since there are more people and money into it,it is naturally inevitable to move closer to the traditional direction of competitive sports.In the traditional competitive sports,the prediction and analysis of the result of the competition is a very important part of the game:NBA,and French Open and many other games have already used the professional data analyst to establish some related models,which are used to deal with the big game data.But in the e-sports industry,this analysis was still remain at the primary,simple and rough level,more often depend on past experience simple statistics and their own processing.This is obviously unscientific,because the characteristics of the e-sports itself,all of the data are easy to obtain,we obviously can use a number of better methods for the modeling of related prediction analysis.Now let’s begin our analysis from one of the hottest E-sports in the world,DOTA2(Chinese Name:DaoTa2).Firstly this paper introduces some related traditional machine learning models:K nearest neighbor model(KNN),logistic regression model(Ir),support vector machine model(SVM)and decision tree model(DT).Xgboost is introduced as a representative of a class of integrated meta learning algorithm model and also presents neural network model(ANN)algorithm with making a analysis of the advantages and disadvantages of each model.Secondly,in view of the above models in dealing with our related data of E-sports,we can not get a good classification performance index.This paper then innovatively introduces the latest research results,word vector(word embedding)in text vectorization,as a solution to this difficult problem.This paper also introduces various models of the generation of word vectors,and makes a brief summary of the relation and comparison between each model and clearing up the obstacles for the subsequent construction of the related models.Then,this paper also designs and builds a competition prediction system,and has done the related work in modularization so as to improve the independent module in the future.Finally,based on the already modularized system,the related data experiments are carried out on the individual PC side,and the correlation performance of the new model is analyzed and compared with the different classification models and the combination of word vectors.Fortunately,based on the experimental results,it is shown that the classification accuracy of the classifiers is improved significantly when the relevant word vectors are introduced as the input of the model.