Automatically detecting the quality of the query and its implications in IR-based concept location

Concept location is an essential task during software maintenance and in particular program comprehension activities. One of the approaches to this task is based on leveraging the lexical information found in the source code by means of Information Retrieval techniques. All IR-based approaches to concept location are highly dependent on the queries written by the users. An IR approach, even though good on average, might fail when the input query is poor. Currently there is no way to tell when a query leads to poor results for IR-based concept location, unless a considerable effort is put into analyzing the results after the fact. We propose an approach based on recent advances in the field of IR research, which aims at automatically determining the difficulty a query poses to an IR-based concept location technique. We plan to evaluate several models and relate them to IR performance metrics.