Semi-automatic Detection of Persian Stopwords using FastText Library

A stopword is a word that does not add much semantic information to the text that despite of its very high frequency. Stopwords include prepositions, conjunctions, and pronouns. One of the steps in natural language processing is to remove stopwords to reduce dataset size and process faster. In this study, a semi-automatic method for collecting the Persian language’s stopwords is proposed. The proposed method lists the stopwords of each text depending on its subject. For this purpose, based on a corpus of news texts, the Inverse Document Frequency (IDF) weight in the text is calculated for each word and the stopwords candidates are determined. Then, using the fastText library, the vector of each word is obtained. In the next step, five neighbors are found for each vector. Next, by removing duplicate words, the final list of stopwords (1014 stopwords) is collected. The result of simulations show the accuracy of detecting stopwords by the k-nearest neighbor method is 94.6%.