| The recommendation system is a system that gives consumers recommendations for consumption.Given the explosive growth of information available on the web,consumers may face countless popular products,movies or meals in their daily lives.Therefore,personalized recommendation is the basic strategy to give consumers a better user experience.Today's recommendation systems play a vital role in various information access systems to facilitate the user's decision-making process.As such,recommendation systems are widespread in many areas such as e-commerce or media websites.In recent years,deep learning has achieved remarkable results in many aspects such as natural language processing,image recognition,and scholars have continued to try other areas to solve complex problems that are difficult to solve with traditional methods.This is because deep learning can accurately extract the implicit non-linear relationships between items from a large amount of data,and can convert various types of abstract behavioral features into more specific data for representation.In the field of recommendation,deep learning has led to dramatic changes in the recommended architecture and provided better methods for improving the performance of the recommendation system.The recommendation system based on deep learning overcomes various obstacles of the traditional recommendation mode and achieves higher recommendation quality,so it has received widespread attention.The main work of this thesis for deep learning-based personalized recommendation models is as follows:1)Perform data analysis and processing on the data set.The data processing mainly includes the preprocessing of the data set and the enhanced operation on the sequence data.In this paper,the preprocessing adopts methods such as defect value processing,data formatting,sample equalization and feature coding,while sequence enhancement uses truncation and random sampling.2)Extract the hidden features of the data after the data processing.The core idea of this step is to use a reasonable model to extract the hidden features of the user,so that the model can accurately obtain the user's different attention points and trends of interest changes.3)After extracting the invisible features,a feature weight extraction model is used to obtain the weights of different feature sequences,and multiple features such as explicit features,invisible features,user information,and item types are fused to generate a new Model,the recommendation effect of the new model has been greatly improved.4)Evaluate the new model structure,using Amazon dataset and MovieLens dataset,respectively,and comparing with other recommendation algorithms,confirming that the newly proposed model can improve the recommendation quality and has good universality. |
