关键词: 潜在语义索引, 主题挖掘, 科技文献

Abstract: Based on a method improved by Latent Semantic Indexing, a topic mining for scientific papers is proposed. This paper describes a process which is used to mine the topics of the scientific papers. It performs conversion, removes non-alphabetic and stop word before further processing. It constructs the term-document matrix based on all words’ weight. It uses modified LSI algorithm to cut the dimension of the matrix and gets a new topic-document matrix. It takes the highest weight of the top five themes as the papers’ topic. This method utilizes the Frobenius norm to regulate matrix, reducing the dimension of the matrix. So the theme of the scientific papers can be mined quickly and accurately.

Key words: latent semantic indexing, topic modeling, scientific documents