Pasokh: A standard corpus for the evaluation of Persian text summarizers

Abstract

The increasingly vast amount of information, particularly on the Web, has resulted in a profound need for automatic summarization systems. The systems, in turn, need to be evaluated in terms of how desirably they can retrieve information. The evaluation is done by comparing the machine summaries against a standard reference corpus containing a reasonably large number of text sources and the summaries that human beings have made out of them. Due to the lack of such a standard corpus for Persian, the summarizers that were developed used to be evaluated against the small corpora constructed by the developers of the proposed systems. This made the systems non-comparable. Thus, Pasokh was constructed as a standard large enough reference corpus. It took over 2000 man-hours of work.