Experiments in Farsi Text Retrieval

-A series of experiments is being conducted on the Farsi language in the domain of laws in the university of Tehran. One of the goals of these experiments is to establish the performance of different weighting schemes and retrieval models. For the lack of a Farsi stemmer and some characterisitics of the language, it was decided to experiment with N-grams. With un-stemmed words and 2-grams, 3-grams and 4-grams and three weighting schemes (dnb.dtn, lnc.ltc and tfc.nfx), 12 experiments were conducted and the first top 20 documents were judged. It seems 3grams perform as well as unstemmed systems for lnc.ltc and tfc.nfx weights. However dnb.dtn outperforms any Ngram weighting schemes. Two linear combinations of the 3-grams and unstemmed words and a linear combination of the 3-grams, 4-grams and unstemmed words were also tried with the hope that each method could capture some aspects of the language. However the results did not show any improvement.