Hardware Optimization Design And Implementation Of Advertising Recommendation System

By establishing the information relationship between internet users and advertisements,the new type of Internet advertising uses intelligent advertising recommendation technology to match the targeted advertisements that users are potentially interested in.The current intelligent advertising recommendation system is given priority to with traditional software CPU implementation.But with the increase of the amount of data,caused by a lack of CPU computing performance,software of CPU is implemented the long user response time and low throughput problem.For the advertisement recommendation system studied in this paper,the hardware acceleration scheme based on FPGA was comprehensively analyzed and selected to improve the computational performance of the advertisement recommendation system.This paper summarizes the model of advertising recommendation system,analyzes the advantages and disadvantages of FM algorithm and FFM algorithm,and designs a hybrid recommendation algorithm based on FFM and DNN.In order to solve the problem of insufficient computational performance of AD recommendation system,this paper has done four aspects: first,the hardware acceleration architecture of advertisement recommendation system based on FPGA is designed by making full use of the repeated computation of recommendation algorithm and the parallel computing ability of FPGA.Secondly,this paper designs a data sharding scheme based on URAM,and optimized the data transmission time consumption in FPGA calculation.Thirdly,this paper improves the parallel computing capability of FPGA through the optimal design scheme of reconfigurable computing,distributed parallel computing and DSP resource reuse.Fourth,this paper proposes the piecewise polynomial algorithm based on LSM and piecewise linear lookup table and ec-cordic algorithm to fit the activation function,which optimizes the accuracy of the traditional fitting method and reduces the use of FPGA logic resources.Finally,this paper builds the FPGA hardware acceleration experimental test platform for the advertising recommendation system.Then the experimental test proves that the fpga-based recommendation system acceleration scheme improves the computational performance by 115% compared with the traditional CPU implementation scheme,and the advertising request timeout rate drops from 7.37% to 0.2%.