How Far Association Rules and Statistical Indices help Structure Terminology?
Abstract
Automatic or semi-automatic structuring of terminology extracted from large corpora still remain a bottleneck issue for managing the fast growing textual sources. This paper aims at defining a methodology to tackle this point using a text mining process for association rules extraction. We show the ability of the rules to enhance the quality of the terminology by filtering the ambiguous, noisy terms of a domain of speciality. However, the mining process often generates a huge number of rules. This issue leads us to raise the question of how can we find a subset of rules that constitutes a valid relational structure according to the knowledge domain. We use statistical indices to rank the rules that are more capable of reflecting the complex semantic relations between terms. We also study how far some rules can help the expert with identifying synonymy/hypernymy relations or with filtering terms.