Research On Intelligent Recommendation System Of Characteristic Agricultural Products Based On LFM-MBGD

With the economic development and the improvement of people’s living standards,people’s demand for specialty agricultural products is also constantly changing.For the sale of specialty agricultural products,due to the influence of regional,se asonal,climatic and other factors,the prices of specialty agricultural products have fluctuated greatly.In the sales process,how to achieve accurate recommendations to increase the sales of characteristic agricultural products has become a hot issue in current agricultural research.By analyzing the applicability of Spark distributed computing engine and LFM hidden semantic model to the recommendation system,this thesis constructs an intelligent recommendation system for characteristic agricultural products.The specific work is as follows:(1)By analyzing the operation mechanism and characteristics of Content-Based recommendation(CB)and Collaborative Filtering(CF)commonly used in recommendation systems,we explore the applicability of two traditional mainstream algorithms in intelligent recommendation of characteristic agricultural products,and on this basis,we study the advantages of Model-based Collaborative Filtering(MCF)in alleviating data sparsity,and propose an intelligent recommendation system of characteristic colored agricultural products based on LFM-MBGD.(2)Through data acquisition and analysis processing,it is found that user behavior habits are non-linearly distributed in time.Therefore,Kafka’s high throughput is utilized to cache and sharpen the data during user data acquisition.Subsequently,offline recommendations and real-time recommendations are computed with the powerful data computing power of Spark.The processed data is fed into the LFM model for dimensionality reduction decomposition.Mini-Batch Gradient Descent(MBGD)is used to iteratively optimize the loss function to achieve offline recommendations.For real-time recommendations,real-time incremental user data is fused with offline recommendation results to achieve real-time results.Finally,three comparison experiments were designed based on the three models of LFM,LFM based on bias,and LFM-MBGD to derive the RMSE and MAE index errors when the F hidden eigenvalue was 10,and the results were RMSE(0.906,0.828,0.0.784),MAE(0.22,0.642,0.582)when the number of iterations was 20,respectively,0.582),and it was also verified that the LFM-MBGD model recommendation was most effective when the sub-data sample was 512.(3)In this thesis,the constructed LFM-MBGD featured agricultural products intelligent recommendation system model is combined with Mongo DB,Java and other technologies to develop the featured agricultural products intelligent recommendation system.It uses Angular JS2,Spring and other components to realize user visualization interface,designs three major pages of user registration and login,product detail page and recommendation homepage,and provides functions of offline recommendation,real-time data collection,real-time recommendation,statistical recommendation,product search and product rating,which provides a reference basis for the sales of featured agricultural products online platform.