Research On Movie Recommendation Algorithm Based On Image Information Mining

With the rapid development of Internet and information technology,people need to spend more and more time every day to screen valuable content from the mass of information,and it is increasingly difficult for movie information service providers to provide personalized quality services for each individual.Traditional information data processing methods such as search engine and other technologies have been unable to meet the requirements of today's information quality,so how to break the barrier between network information and users has become an issue that is hard to ignore.In such a social context,recommendation algorithm as an effective means to solve the problem of information overload has attracted more and more attention from researchers and information service providers.Among them,movie recommendation has become the key research direction in this field due to its high commercial value,frequent use and diverse data types.Traditional movie recommendation algorithms often use existing text labels and user information to recommend,such as the name of the movie,release time,leading actor,etc.,but they ignore the multimedia information of the movie itself.As a highly condensed film story and art,movie posters not only affect the audience's first impression of the film but also reflect the abstract category to which the film belongs.In recent years,with the rapid progress of electronic hardware technology,the processing and mining of unstructured data information such as images is becoming more and more convenient,and related derived technologies are becoming more and more abundant.Therefore,it is of great significance to apply the method of image information mining to the traditional film recommendation system.Based on the public Internet movie poster image as the research object,the fusion using multiple tags disequilibrium of method,data processing method and artificial convolution neural network of information mining,in order to achieve the purpose of expansion of the original film type collection,using the new film type set again to optimize the traditional movie recommendation algorithm,in order to gain more accurate list of recommended.The main contents of this article are as follows:(1)According to the data characteristics of movie posters recommended scenario image,on statistical analysis of the distribution type of film samples by fusion of tabbed learning framework(Calibrated Label Ranking),unbalanced data processing method(Easy Ensemble)and convolutional neural network solves the problems of movie posters type is difficult to extract.(2)The mathematical principles,advantages and disadvantages of a single shallow learning method and a deep learning model are studied,and the performance differences between the two methods in film poster classification are analyzed.By setting up a variety of experimental schemes for performance comparison,a base classifier suitable for classification between the two film types is constructed.(3)To solve the problem of sparse distribution caused by the large difference between movie ID,name and other data contents,the dimension explosion caused by traditional one-hot coding was avoided by introducing the embedded matrix.From the perspective of feature fusion and model training,the text convolution network is chosen to avoid the difficulty of feature extraction of movie names.In this paper,the performance of recommendation algorithm was analyzed by using the movie data of 1M from the public data set MovieLens and the movie poster image crawled from the IMDB website.With the test set accuracy and Top_N accuracy as the evaluation indexes,the combination with better performance was selected by comparing a variety of model structures.The experimental results prove that the fusion of movie poster information into the movie recommendation algorithm is more advantageous than the traditional method,and also prove that the movie poster can reflect the general type of a movie from another perspective.