An Evaluation of the Formal Concept Analysis-Based Document Vector on Document Clustering

In conventional approaches, documents are represented by the vector whose dimensionalities are equivalent to the terms extracted from a document set. These approaches, called bag-of-term approaches, ignore the conceptual relationships between terms such as synonyms, hypernyms and hyponyms. In the past, researches have applied thesauri such as Word Net to solve this problem. However, thesauri such as Word Net are developed more for general purposes and are limited in specific domain. Therefore, an automatically built ontology for terms is desired. In our previous study, we proposed a method which applies formal concept analysis (FCA), an automatic ontology building method, to extract the term relationships from a document set, and then apply the extracted information as the ontology of terms to represent the documents as concept vectors. In order to evaluate the usability and effectiveness of the proposed method for information retrieval related applications, we employed the concept vectors generated for the documents to the document clustering. In this study, we apply bisecting k-means clustering and hierarchical agglomerative clustering as the platforms with which to evaluate our method.