Resumen
Background: The Rapid growth of technological developments in Indonesia had resulted in a growing amount of information. Therefore, a new information retrieval environment is necessary for finding documents that are in accordance with the user?s information needs.Objective: The purpose of this study is to uncover the differences between using Relevance Feedback (RF) with genetic algorithm and standard information retrieval systems without relevance feedback for the Indonesian language documents.Methods: The standard Information Retrieval (IR) System uses Sastrawi stemmer and Vector Space Model, while Genetic Algorithm-based (GA-based) relevance feedback uses Roulette-wheel selection and crossover recombination. The evaluation metrics are Mean Average Precision (MAP) and average recall based on user judgments.Results: By using two Indonesian language document datasets, namely abstract thesis and news dataset, the results show 15.2% and 28.6% increase in the corresponding MAP values for both datasets as opposed to the standard Information Retrieval System. A respective 7.1% and 10.5% improvement on the recall value at 10th position was also observed for both datasets. The best obtained genetic algorithm parameters for abstract thesis datasets were a population size of 20 with 0.7 crossover probability and 0.2 mutation probability, while for news dataset, the best obtained genetic algorithm parameters were a population size of 10 with 0.5 crossover probability and 0.2 mutation probability.Conclusion: Genetic Algorithm-based relevance feedback increases both values of MAP and average recall at 10th position of retrieved document. Generally, the best genetic algorithm parameters are as follows, mutation probability is 0.2, whereas the size of population size and crossover probability depends on the size of dataset and length of the query.Keywords: Genetic Algorithm, Information Retrieval, Indonesian language document, Mean Average Precision, Relevance Feedback