Automatic thesaurus construction by machine learning from retrieval sessions

摘要

Users of information retrieval systems (IRS) know and use many relationships between concepts a long time before these find their way into textbooks, printed thesauri, or classification schemes. We present here an IRS component called TEGEN, which taps this expertise by automatically drawing conclusions from actual search behavior about possible thesaurus entries. This is done during an iterative knowledge acquisition process: only after explicit or implicit confirmation by other users of the IRS during the knowledge verification process, the results are incorporated into a thesaurus. TEGEN is written in PASCAL using a knowledge-based programming method. It uses the relational database system IMF2 and is implemented at the Technical University of Munich and at the Leibniz Computer Center of the Bavarian Academy of Sciences.