An Improved of Stemming Algorithm for Mining Indonesian Text with Slang on Social Media

Stemming is one of the important processes in text mining. The result of stemming process gives an impact on the next text mining process. Therefore, the technique of processing text data such as text mining, natural language processing, and information retrieval for every language have different treatment, especially in the stemming process. Lately, stemming algorithm for Indonesian text has been growing rapidly. However, there is no stemming algorithms that specifically accommodate Indonesian text with slang, whereas text data from social media contains many natural languages and slangs. Therefore, the aim of this research is to improve Indonesian stemming algorithm that suitable for Indonesian text data with slang from social media. We analyzed and elaborated Porter stemmer algorithm to create a new stemmer algorithm that appropriate with Indonesian slang characteristics. This study proposed two factors as the state of the art, (1) a utilization of both Indonesian dictionary and Indonesian Slang dictionary to anticipate re-stemming an infinitive, and (2) accommodating natural languages and slangs to strengthen stemming process on word particles. We examined the algorithm with two scenarios, the first scenario used 379 of words and 20 text data for the second. Furthermore, we compared the algorithm with Porter, Nazief&Adriani, and Lucene stemmer algorithm. The result showed that the proposed algorithm could do stemming process with Indonesian slang well, with the percentage of accuracy about 88.65%. The accuracy level was increase, although the memory usage of the algorithm was no better than the others algorithm, but the duration of process was not different significantly.