Research On Film Data Analysis And Prediction Based On Ensemble Learning And Transfer Learning

In recent years,with the development of machine learning research and the emergence of new achievements,more and more machine learning algorithms are widely used in daily life.At present,the research on machine learning algorithms is mostly based on some datasets,and the information contained in data is mined by using reasonable data analysis methods.It has certain reference value for the development of some industries.Film data has the characteristics of large amount,wide range of sources and easy to obtain.Therefore,this thesis studies the ensemble learning method based on Stacking,the prediction method of film ratings based on multisource weighted Tr Ada Boost and the prediction model of film ratings based on domain adaptive algorithms.Firstly,in the part of film data preprocessing and feature engineering,this thesis takes IMDB film dataset as the research object,and carries out the research work of feature coding,constructing film composite type influence index,director and actor levels,feature selection,removing singular film samples.Then,this thesis studies the prediction model of film ratings based on Stacking ensemble learning method.The basic learner integrates a variety of single machine learning models.Through the experimental comparison,it is found that the Stacking ensemble learning model can effectively reduce the prediction error compared with these single models,which has certain advantages.Next,in order to solve the problem that a large number of film data samples lack rating labels in real scenes,which leads to the failure to train a machine learning model with good prediction effect,this thesis uses transfer learning theory to predict ratings.In this thesis,an instance-based multisource weighted Tr Ada Boost transfer learning method is used for research,and the Stacking ensemble learning model is used as base regression to predict ratings based on the research content of the ensemble learning section.At the same time,for verifying the effectiveness of multisource domain transfer learning,this thesis further conducts a comparative study of multisource domain transfer,single-source domain transfer and non-transfer.In order to further improve the prediction performance of the model,based on the study of single source domain transfer,this thesis proposes a multisource weighted Tr Ada Boost model with different regressions,and verifies its effectiveness by comparative study.Finally,this thesis adopts the feature-based domain adaptive transfer learning method for research.In the study,it is found that because of the inconsistency of feature dimensions between the target domain and the source domain,this thesis adopts the domain adaptive transfer learning model combined with dimension reduction algorithm.At the same time,in order to reduce the prediction error of film ratings,this thesis uses the Stacking ensemble learning model as a regression based on the research of the ensemble learning part.At last,through comparative experiments,the effectiveness of the domain adaptive transfer learning model is verified,which can better realize the rating prediction of target domain film samples.