The Automatic option of inference rules for the fuzzy TF-IDF

For several NLP tasks, we use the Fuzzy Inference System (FIS) to develop a set of decisional systems. Still, the selection rules FIS process needs some modification to be more useful. In this paper, we use FIS to produce terms weight for electronic documents basing on the popular TF-IDF (term frequency-inverse term frequency) components. To insert rules for FIS manually remains a real issue in the text processing domains, where the size of the processed data is enormous. In our work, we suggest including the Association Models as automatic technics to produce the inference rules instead of the traditional manual way. Precisely, our approach process in five steps, i.e., preprocessing, limiting membership function for every input and output, nominal representation of the database, call for the Apriori and Filter association as association models to select the most important rules, and finally, post-processing of the received rules. Applying the new technique to produce fuzzy weight for a benchmarked database achieves a significant gain in time, precision rules, and weighting quality. Thus, running the new descriptor matrix with statistical classifiers, e.g., Bayesian Networks, improves accuracy compared with the traditional way and existing methods.