A Personalized Recommendation Algorithm Based On The Feature Dataset Of User Profiles

Internet technology has developed rapidly in recent years,and along with the development of the Internet,data is being generated and accumulated at an unprecedented rate.In order to effectively dig out the effective information contained in the massive data and to provide recommendations and references for the client users,the recommendation system emerges as the times require,which has huge development potential with research value and attracts much public attention.The general recommendation system has limitations on the processing of user features.It depends too much on human experience and understanding of the business and therefore its ability of data mining and utilization is limited,and it is difficult to obtain higherdimensional features that are more conducive to the recommendation.In addition,the traditional random forest algorithm directly integrates all the base decision trees without filtering,and the recommended dataset is usually a high-dimensional data set containing noise and redundant features.Therefore,there will be disturbances which lead to a decrease in performance when the random forest algorithm is applied to the recommendation.In order to solve the above problems,this thesis studies and implements a personalized recommendation algorithm based on a dataset of features from user profiles.Firstly,this dissertation constructs a dataset of features from user profiles on methods of artificial feature engineering and deep learning algorithms,which mainly includes four parts: 1)Data analysis and preprocessing performed on a real data set from Alibaba's mobile ecommerce platform,which makes the dataset more suitable for data mining;2)Processing based on artificial experience,business understanding,and statistical analysis,which starts from general analysis with prior knowledge,used to construct features;3)A process of feature construction based on an improved feed-forward neural network.The network can handle high-dimensional and sparse features and has the ability to automatically learn features.It can combine and transform low-level features to obtain high-level features.4)The dataset of user profile features constructed from the integration of the first two parts as fine-grained user profiles and user labels serves as coarse-grained user's profiles.The experiment shows that a better result can be obtained by using the dataset,compared with the result when using a single type of feature.And the recommendation task for users can be accurately and efficiently completed.Next,this paper proposes an optimized random forest algorithm.The optimized algorithm comes with an instructive fusion method,which makes the probability of the base decision tree with good performance higher,and the impact,to the fusion result,of the those with lower performance is reduced,while the algorithm's generalization ability is ensured.Experimental results show that the improved random forest algorithm has higher classification ability and performance than the traditional random forest algorithm on real recommended data sets with noise or redundant features,and furthermore,it can meet actual needs better.Finally,combined with the dataset of features from user profiles proposed in this paper and the optimized random forest algorithm,we design and implement an personalized recommendation system based on user behavior.The system is divided into two modules: recommendation of package plans and recommendation of broadband products,then analysis and modeling work is done separately for both two modules.The final experimental results show that the system has good performance and can be well adapted to real-world recommendation scenarios.