A novel approach to extract important keywords from documents applying latent semantic analysis

Abstract

Keywords extraction is one of the significant fields in natural language processing. The main aspect of this process is to retrieve a set of important terms that represent the core information of a document in details, which is directly related to the document context. In this research, an original approach is presented for keywords extraction technique, which is based on semantic relation. A semantic relation is a model to show the similarity between two or more terms in semantically, which is indicating the relation between words to words in a sentence as well as with the other sentences. This system extracts a specific number of key terms from documents to identify the core content of a text. The datasets are collected from different sources such as books, journals, newspapers etc. Support vector machine, Logistic regression, PAT-tree, word co-occurrences and some other well-known techniques from machine learning and statistical approach has been used to extract keywords. This system shows modified semantic relation approach gives an accuracy of 77.6% precision and 84.3% recall for selected datasets. However, we have found out that this model shows a better result than some conventional approach that has been applied in this field such as position weight and term frequency. Additionally, in this system term frequency, stemming, POS tagging, stop words, and n-grams play an important role to extract keywords from documents.