From Terminology Extraction to Terminology Validation:An Approach Adapted to Log Files
Hassan Saneifar (LIRMM - University Montpellier 2 - CNRS, France)
Stéphane Bonniol (Satin Technologies, France)
Pascal Poncelet (LIRMM - University Montpellier 2 - CNRS, France)
Mathieu Roche (TETIS - Cirad - Irstea - AgroParisTech, France)
Abstract: Log files generated by computational systems contain relevant and essential information. In some application areas like the design of integrated circuits, log files generated by design tools contain information which can be used in management information systems to evaluate the final products. However, the complexity of such textual data raises some challenges concerning the extraction of information from log files. Log files are usually multi-source, multi-format, and have a heterogeneous and evolving structure. Moreover, they usually do not respect natural language grammar and structures even though they are written in English. Classical methods of information extraction such as terminology extraction methods are particularly irrelevant to this context. In this paper, we introduce our approach EXTERLOG to extract terminology from log files. We detail how it deals with the specific features of such textual data. The performance is emphasized by favoring the most relevant terms of the domain based on a scoring function which uses a Web and context based measure. The experiments show that EXTERLOG is a well-adapted approach for terminology extraction from log files.
Keywords: information extraction, log files, natural language processing, terminology extraction, terminology ranking, text mining
Categories: H, I.2.7, I.7