The Design And Implementation Of Recommendation System Based On Douyin Point Of Interest

Douyin is a short video software developed by ByteDance.With its personalized recommendation technology,it has gained a huge number of users in just a few years.However,the current recommended content of Douyin is only limited to videos,the recommended forms are not diverse enough,and the content ecology is not perfect.In addition,the way that Douyin monetizes traffic currently mainly relies on advertising,and the commercialization methods are not rich enough and lack the ability to effectively convert online traffic offline.When using Douyin,in addition to watching videos,users can also take pictures of their own works and publish them.Douyin allows users to bind a geographic location to published videos.This information is called Douyin points of interest.Generally,points of interest are only displayed during video playback,which cannot satisfy the user's desire to explore the shooting location when watching the video.Offline merchants corresponding to the points of interest also find it difficult to obtain revenue from Douyin except for advertising.In order to enrich the recommended content ecology of Douyin,extend relevant information about points of interest,and at the same time explore scenarios for commercializing traffic by guiding users to purchase recommended products of points of interest and consume them offline,and ultimately enhance commercial value.The company considers making relevant recommendations regarding points of interest.Based on the analysis of the needs of Internet users and third-party suppliers,this thesis designs and implements a recommendation system based on Douyin points of interest around video information,point of interest information and user behavior information.The system is divided into recommendation management module,similar recommendation module,commodity shelf module and related video module according to function.Each module provides corresponding detailed design and implementation.The system development uses the Golang language that natively supports high concurrency,and the overall microservice architecture is used.The services interact through the RPC communication framework.The data storage uses MySQL as the basic data storage,the hot data uses Redis for caching.ElasticSearch is used for the online search of large data sets.Hive is used for offline calculation and building a data warehouse.Kafka is used for message components.Through the combined use of business components and middleware,high concurrency and high availability of the system are ensured.This system recommends relevant information about points of interest in the dimensions of commodities,videos,etc.It has been running stably since it went online,satisfying users'desire to explore points of interest,and greatly improving the user experience of Douyin.The bound merchants have obtained considerable traffic from Douyin,which has increased their revenue.The system has good scalability at the same time.Recommendation modules can be added freely with the iteration of the business without affecting overall performance.