A generation method of enterprise appropriate policy tags based on LDA Model
Konferenz: EEI 2022 - 4th International Conference on Electronic Engineering and Informatics
24.06.2022 - 26.06.2022 in Guiyang, China
Tagungsband: EEI 2022
Seiten: 8Sprache: EnglischTyp: PDF
Autoren:
Tan, Cuiping (National Science Library, Chinese Academy of Sciences, Beijing, China & Department of Library, Information and Archives Management, School of Economics and Management, University of Chinese Academy of Sciences, Beijing, China & Information Centre of Beijing Municipal Bureau of Economy and Information Technology, Beijing, China)
Yu, Feng (Institute of Data Science and Agricultural Economy, Beijing Academy of Agricultural and Forestry Sciences, Beijing, China)
Gao, Yuan (Central University of Finance and Economics, School of International Economics and Trade, Beijing, China)
Hu, Bin (Beijing Century Durango software technology development Limited Company, Beijing, China)
Inhalt:
This paper summarizes the common methods of constructing enterprise portrait tags, and puts forward a generation method of enterprise appropriate policy tags, in order to solve the difficulties of enterprises in finding policies. The policy corpus about Beijing are built, and 362 policy data is collected in it from 2012 to 2021. Based on the correlation research between qualification tags and policy texts in enterprise portraits, a method of generating enterprise appropriate policy tags based on LDA is proposed, and a comparative experiment is carried out with LSA model. The F1 value of enterprise appropriate policy matching based on LDA is 86.08, which is higher than that of LSA model, which greatly reduces the manual labeling cost and provides a reference method for peers. The models are based on the actual data. There is no standard data set for comparative experiments in fact, which has some limitations. This paper provides a new solution for the construction of enterprise label model, qualitative tags topic mining and tags utilization of the enterprise portrait.