Design And Implementation Of Personalized Information Recommendation System

With the prosperity of mobile Internet and micro-computing terminals,the advent of the information age has greatly enhanced the number of people receiving information.Nowadays,information overload has developed into one of the most serious problems on the Internet.Most of the information posted on the Internet can not escape the drowning fate,and users are difficult to locate the information they want in the face of the huge amount of information.The birth of the recommendation system partially solved this dilemma.This system(personalized information flow recommendation system)provides a solution for the two.Through a robust and flexible recommendation system architecture design,it provides developers and operators with a recommendation system background that can quickly configure recommendation strategies.At the same time,the system essentially serves information consumers and provides users with personalized information flow solutions.After analyzing the system,the system is designed into four modules: user portrait,online recommendation engine,offline calculation,and traffic distribution.(1)User portrait module: extract user portraits from user behavior logs,generate user demographic portraits and user interest portraits in different time windows.(2)Online recommendation engine module: recommend personalized information flow lists to readers.The recommendation process is designed for candidate generation,selection recall,rule filtering,precise scoring,and fusion ranking.The selection recall uses the DSSM-based estimation model,and the accurate scoring uses the LR-based staylength estimation model while using the FTRL method for training.(3)Offline calculation module: offline calculation of the models used in the online recommendation engine,providing model parameters for the online recommendation function,and designing the following four functions: log processing,feature generation,sample generation,and model training.(4)Traffic distribution module: manages the traffic distribution of the entire system.This module is based on Google overlapping experiment framework,using a multi-layer nested traffic hierarchy of domains,layers,and experiments,using user IDs,device cookies,access dates,and random Four kinds of modular distribution and conditional distribution based on experimental attributes are used to distribute the flow.This system adopts a distributed architecture to ensure the availability,consistency,robustness and scalability of the system,and uses big data technology to process massive data to ensure the system's low latency and high response.After functional testing and non-functional testing,it is shown that the system meets the design requirements,the system response time is less than 1 second,and the click rate of the information flow recommendation result is above 15%.