تحلیل توزیع و تمرکز کلیدواژه‌های پارساها: میزان تطابق با توصیفگرها، عنوان، و چکیده

Index terms provided by authors and professional indexers are used in traditional information retrieval schemes. However, abstracts ideally contain the core message of a document. This can potentially give us the opportunities to use the abstracts to automatically extract index terms. The purpose of this work is to be used as a base or as the first stage in the automatic keyword extraction as well as the high-level perception of where the ongoing research is headed Iranian Theses and Dissertations (TDs). To achieve the aforementioned objectives, we studied on more than 500 samples in different engineering research area from 50 different universities: 1) the correlation between the authors and professional indexers keywords. We observed only 8% similarity between these two indices. 2) We studied the correlation between the index terms and words in abstract and title. We found that that 40% of author keywords are extracted from first 20% of the abstract (This figure changes to 45% for professional indexer) and 24% from the second 20% (19% from the next 20%) This finding can be further used to narrow down the input dimensions for the various machine learning schemes for automatic keyword extraction. 3) Using some classification schemes it can be perceived that the most of the ongoing research in Iran is headed toward neural network and optimization.