Automatic Keyphrase Extraction from Persian Scientific Documents Using Semantic Relations

With the rapid growth of Persian scientific documents in online databases, the need for quick access to these documents has increased in recent years. As a common way to provide a succinct summarization of a document, Automatic Keyphrase Extraction (AKE) plays an important role in various applications, e.g. indexing, summarization and information retrieval. Although lots of methods have been proposed for AKE, they have not resulted in satisfactory performance for Persian texts. The main reasons include dependency on linguistic resources, training datasets, and special NLP tools that are not properly available for Persian language. Considering the special need for an effective and efficient method to carry out the task of AKE on Persian documents, we have proposed a novel Thesaurus based method. The method leverages semantic relations between words of a given document to construct graphs of related words and then extract candidate keyphrases which are semantically and grammatically correct using a scientific and technical thesaurus. Final keyphrases will be selected based on a weighting method in which the position of the phrase in the text, i.e. abstract, body, etc., would affect its weight. The experimental results show that our method outperforms the similar works on Persian texts under three evaluation metrics.