Design And Implementation Of A Short Video And Live Broadcast Platform Based On DeepFM Algorithm

With the advent of the 5G era,an online,networked and intelligent information revolution has been launched.In this new information age,with the continuous growth of mobile Internet users,and the emergence of new information transmission methods and marketing models,short video live broadcast platforms have emerged in this revolution.In order to keep up with the pace of the times,promote enterprise culture,enterprise services,enterprise products through multiple channels,and seize their own traffic to realize traffic monetization,the company’s own short video live broadcast platform came into being.According to the needs of the enterprise,the system can accurately and efficiently push short videos to users based on the intelligent recommendation algorithm,and guide users to understand the enterprise and its related products.Based on the software development process in software engineering,this thesis analyzes the project background and current situation at home and abroad,and understands and determines the feasibility and development direction of the project.According to this direction,technology selection,system requirement analysis,system outline design,system detailed design and implementation,system testing,project summary and outlook are carried out in turn.On the whole,the project is divided into two parts: the short video live broadcast system server and the short video live broadcast system client.The system server is mainly divided into short video management,live broadcast management,enterprise employee management and system management,which realizes the overall data management of the enterprise to the platform.The system client pushes short videos to users in a sub-channel mode.In order to achieve the push effect of thousands of people and thousands of faces,the Deep FM recommendation algorithm is used to achieve intelligent recommendation.The model training data comes from the Douyin dataset,it is recommended to use the Lambda architecture for data processing to integrate offline computing and real-time computing.Through front-end embedding,Flume is used to collect log data,and user behavior data is stored in HDFS.The offline part uses the collected user behavior for user portrait construction,video portrait construction,offline video recall and sorting model training.The real-time online part uses Kafka to connect with Flume to collect user behavior message queues,and uses Spark Streaming to process click logs sent by Kafka in real time to update features and recall sets in real time.The short video live broadcast system involved in the thesis has been put into use online,runs stably,has the expected usability,and has good security and maintainability.The project broadens the business channels for the enterprise,realizes the commercial realization of its own traffic,and at the same time provides a good corporate culture publicity platform for the enterprise,contributing to the development of the enterprise.