Design And Implementation Of Recommendation System Based On Commodity Trust And Social Network

With the rapid development of e-commerce,social and mobile technologies,a space based on mobile Internet has been created.With social software as a tool,people-centered,social as a link,a new business different from traditional e-commerce-social E-commerce.The social e-commerce platform is a platform for publishing and regulating social e-commerce products.Since social e-commerce products come from thousands of mobile-registered personal businesses and merchants have lower entry thresholds,and the quality of goods is uneven,it is necessary to ensure that the platform prioritizes the promotion of genuine goods,and the social e-commerce platform introduces the trust product recommendation system.Guide users to buy or share high-trust items,improve the shopping experience,enhance word of mouth to retain more customers,and gain more profit.This project is mainly based on the actual project recommendation of the company's social e-commerce platform,researching the trust degree of commodities in millions of commodities,and data collection,offline calculation and real-time recommendation architecture design based on part of the open source framework.According to the user behavior,the product is recommended to the user in real time,the difficulty of the user to select the product is reduced,and the user's shopping experience and the sales volume of the social e-commerce platform are improved.The project uses a big data framework to implement real-time recommendations,using a commodity sorting algorithm and a commodity similarity algorithm to train the recommended models and generate recommendations.The research contents are as follows:(1)Global Commodity Trust Algorithm: Calculate commodity ranking based on Markov chain Monte Carlo model and sequential user behavior data.Firstly,based on the user's behavior,the matrix of the product correlation coefficient is calculated.Then,according to the correlation coefficient matrix,the state transition probability matrix is calculated.Finally,according to the state transition probability matrix,after a sufficiently long time evolution,the probability of each commodity being purchased is calculated.,that is,the user's trust in the product.The focus is on establishing a commodity connection matrix and a state transition probability matrix.(2)Real-time recommended architecture design: first design the buried point log,collect the user behavior record;then use Flume to collect the user behavior record into Kafka;analyze the user behavior through Spark Streaming and generate real-time recommendation results.(3)Off-line computing architecture design: offline data acquisition,offline calculation.Off-line computing uses Spark to generate recommendation models and generate initial recommendation results based on models.The business process layer is mainly used for business intervention,filtering,and sorting to generate final recommendation results.In the algorithm proposed in this paper,the calculation of commodity trust,and the combination of commodity trust and social network for recommendation are the main innovations;while high-performance offline computing and high real-time online recommendation are the main characteristics of the system.At the end of the paper,a number of comparative experiments are set up.The performance and real-time performance of the proposed algorithm and system are evaluated.The main evaluation indicators are average absolute error,root mean square error and coverage.The performance evaluation index is calculated.The time consumed by different data volumes,the real-time evaluation index is the response time of the system under different concurrency numbers.The experiment proves that the recommendation of combining product trust and social network is about 0.01 lower than the average of MAE based on social network recommendation,the average of MASE is about0.05,and the coverage is about 10% higher on average.The average MAE of recommendation is about 0.02 lower than the original recommendation,the average MASE is about 0.1,and the average coverage is increased by about 30%.The system consumes no more than 20 minutes in a single iteration of the offline calculation and the response time of the online recommendation is also no more than 3 seconds.