MFSRank: An unsupervised method to extract keyphrases using semantic information Roque E. L´opez 1 , Dennis Barreda 2 , Javier Tejada 2 , and Ernesto Cuadros 2 1 School of System Engineering, San Agustin National University, Per´ u rlopezc27@gmail.com 2 School of Computer Science, San Pablo Catholic University, Per´ u {dennis.barreda,jtejadac,ecuadros}@ucsp.edu.pe Abstract. This paper presents an unsupervised graph-based method to extract keyphrases using semantic information. The proposed method has two stages. In the first one, we have extracted MFS (Maximal Fre- quent Sequences) and built the nodes of a graph with them. The weight of the connection between two nodes has been established according to common statistical information and semantic relatedness. In the sec- ond stage, we have ranked MFS with traditionally PageRank algorithm; but we have included ConceptNet. This external resource adds an extra weight value between two MFS. The experimental results are competitive with traditional approaches developed in this area. MFSRank overcomes the baseline for top 5 keyphrases in precision, recall and F-score mea- sures. Key words: Keyphrase Extraction, Maximal frequent sequences, Se- mantic Graphs 1 Introduction Currently, keyphrases extraction task has taken big importance. This is due to exponential growth of information and the need to understand it quickly. Keyphrases of a document are the words and phrases that can precisely and compactly represent the content of the document [1]. As they represent the main topics of a document, keyphrases can be used in many Natural Language Pro- cessing (NLP) tasks, such as summarization, information retrieval, classification and clustering of documents. Usually, keyphrases are assigned manually. In scientific articles, keyphrases help readers to have a global idea of the article and in web pages they serve like metadata which describe its content. Unfortunately many others documents do not have keyphrases assigned, wasting their benefits. The main reason for the absence of keyphrases in documents is that the manual assignment is a laborious task. As shown in [2], in recent years has reemerged interest in automatic keyphrases extraction. Different approaches have been developed to give solution to this task, many of them have obtained very good results. However, most of them do