The Positive Effect of PMI on the Selection of Meaningful Words

The aim of this study is to enhance the quality of automatic created dictionary. As far as we know, PMI (Pointwise Mutual Information) is the first time applied in order to determine representative words in the literature. A previously developed system is revised to experiment PMI effect on the selection of representative words additionally. Firstly, the meaningful words of the seed document/s are determined by Helmholtz Principle. Then, PMI value is calculated for each of the meaningful words. Meaningful words are sorted by using the total PMI value in decreasing order. Finally, the top n meaningful words are added to the dictionary, where n is experimentally worked. Hash similarity is used to measure the performance of the dictionary. PMI enhances the dictionary quality for all n value. The dictionary hash similarity increases from 40.46% up to 68.75% for the best n value in the experiments.