关键词: 数据仓库, 文本相似度, Spark , 论文推荐

Abstract: In order to promote the development of journal convergence media, improve the knowledge service ability of magazine website and provide readers with online paper recommendation service, a paper recommendation method based on data warehouse is proposed. First, set up a data warehouse, and establish papers recommended theme library, data extraction paper title, abstract, keywords and other data to estab⁃ lish a feature data set; then, the semi-structured word segmentation feature data set is obtained by prepro⁃ cessing these feature data sets. The above data is stored in the ODS layer of the data warehouse, the datasin original data layer are formatted and ETL，the dimension missing items are cleaned，then the datas are stored in the DWD layer. The dimension-paper weight matrix is stored in the created DWS layer, and the summary recommendation data is stored in the ADS layer of the top application layer. Finally, the similarity of the paper to be recommended is calculated from the word segmentation feature data set in the ADS layer topic table, and the similar literature is recommended for the target paper according to the similarity value. The results show that the proposed method improves the real-time and accuracy of paper recommendation, and the application effect is good.

Key words: data warehouse, text similarity, Spark, paper recommendation