Abstract:This paper applies the personalized recommendation method to assist users to carry out file retrieval research, and based on user history search records and user website behavior log to analyze the search results that users want, and passive search is active recommendation. The article selects several parts from the construction of the recommendation system, the overall architecture design, the data collection source analysis, the data processing strategy, the recommendation engine model design, and the machine learning computing framework. The algorithm core of the file-based collaborative filtering algorithm superimposed graph-based recommendation model is emphasized. By calculating the similarity between the files, and generating a recommendation list according to the similarity of the files and the historical behavior of the user, the recommendation results are filtered and sorted according to the relationship diagrams established by the entity associations such as posts and knowledge points. Through the research on document-based personalized recommendation of machine learning, it has made useful exploration for the development and utilization of documents and information resources based on big data and artificial intelligence technology.