Abstract: 【Purpose/significance】This paper discusses the research of scientific literature clustering based on keywords, including,
the use of keywords with different characteristics to achieve document clustering, what is difference in the clustering effect; how to se⁃lect keywords according to the characteristics to achieve better clustering results.【Method/process】According to the frequency and se⁃mantic type of keyword, the control group was set for empirical study of literature clustering. To observe the impact of keyword charac⁃teristics on literature clustering density and semantic representation effect.【Result/conclusion】According to the research results, the proper clustering density can be obtained by using the keywords with the characteristics of ultra-high, sub-high, research topic or lim⁃ited range alone for literature clustering. Ultra-high frequency characteristics are usually reflected in other frequencies. Sub-high fre⁃quency words can simultaneously reflect the characteristics of keywords of different frequencies , but its ability to express the charac⁃teristics of the mid –frequency words is not comprehensive enough; separating keywords with different semantic types and implement⁃ing document clustering is better than using words with multiple semantic types, and there is mutual exclusion between the keywords
with different semantic types.【Innovation/limitation】This paper finds that when document clustering is based on the co-occurrence re⁃lationship between keywords, it is effective to select sub-high frequency keywords or keywords of a certain semantic category to achieve document clustering. However, further research on the semantic structure relationship between keywords is lacking.