Research On Prediction Model Of APP Recommendation Content Based On Data Mining

With the widespread popularity and rapid development of the Internet and the continuous upgrading of mobile intelligent terminals,the word APP is no longer a new vocabulary.It has gradually penetrated into all aspects of life,and because of the convenience brought by the popularity of these apps.Using APP has become a new lifestyle.Therefore,in this environment,there are various APPs in the market that can meet different needs,but the specific user groups are limited.How to make your products always liked by users and have high loyalty in this market.In fact,it is a traffic campaign,which is the core of every enterprise.For the Internet companies in the market,users are life,just like the resources competition between countries,the user is the wealth,so improving the quality of content operations is an issue that enterprises must consider.The content is novel and can satisfy the user's psychological appeal.Such content is useful,and it is the valuable output that both enterprises and users need.Therefore,the operation of content should be better grasped and managed,and data mining The predictive model in the middle is well suited to the needs.The main purpose of this paper is to predict whether a content will be liked by users,that is,the prediction of APP recommendation content is regarded as a two-category problem,because the amount of data is not very large,and the logistic regression algorithm is chosen to establish the prediction model.On the selection of the feature system,the method of combining multiple regression and iterative decision tree is chosen.The data used in the training model comes from all the posts published by Facebook on a Facebook brand provided by Facebook,a world-famous social platform,with a total of 710 instances and 21 attribute values.Because the article is to be regarded as a two-category problem,the target column is divided,and the mark larger than the average favorite amount is 1,and the flag is 0.By preprocessing the data,a basic data that is more conducive to research is obtained,in order to play the most important role in modeling.Through the correlation analysis of the features and the importance analysis using the GBDT algorithm,the features that have no effect on the results,the correlation is not good,and the unimportant features are removed.Finally,14 features are formed to form the feature system of the APP recommendation content.This feature system was substituted into the logistic regression algorithm,and the final prediction accuracy was 84.8%,and the result was good.Through further analysis,suggestions can be given for the post-posting of the cosmetics brand.For example,the content of the post should be given priority to the needs of the fan users,the content of the article should not be too short,and the content of the video category is more popular.