Fuzzy Ontology Mining and Semantic Information Granulation for Effective Information Retrieval Decision Making

Abstract

The notion of semantic information granulation is explored to estimate the information specificity or generality of documents. Basically, a document is considered more specific than another document if it contains more cohesive domain-specific terminologies than that of the other one. We believe that the dimension of semantic granularity is an important supplement to the existing similarity-based and popularity-based measures for building effective document ranking functions. The main contributions of this paper is the illustration of the design and development of a fuzzy ontology based granular information retrieval (IR) system to improve the effectiveness of IR decision making for various domains. Based on the notion of semantic information granulation, a novel computational model is developed to estimate the semantic granularity of documents; these documents can then be ranked according to the information seekers' specific semantic granularity requirements. One main component of the proposed computational model is the fuzzy ontology mining mechanism which can automatically build domain-specific ontology for the estimation of semantic granularity of documents. Our TREC-based experiment reveals that the proposed fuzzy ontology based granular IR system outperforms a classical vector space based IR system in domain specific IR. Our research work opens the door to the applications of granular computing and fuzzy ontology mining methods to enhance domain specific IR decision making.

Mining Contextual Knowledge for Context-Aware Recommender Systems

“…Essentially, the windowing process is a way of incorporating the proximity factor into the text mining process. According to previous studies, a text window of 5 to 10 terms is effective [11], [7]. The Mutual Information metric has been applied to collocational analysis [12].…”

“…The Balanced Mutual Information (BMI) [11], [7] is examined in this paper to compute the association weights among terms. Different weight factors are applied to the term MI(t i , t j ) and the term MI(t i , ¬t j ) respectively so that the negative association between the presence of a term t i and the absence of another term ¬t j will not over dominate our term association weight calculation.…”