Otomatik metin sınıflandırma tekniklerinin çok sınıflı analizi

Text classification and clustering; are one of the most popular areas of research in natural language processing applications. These areas offer different possibilities to the researchers for determining the metrics that can measure corpora dynamics in the automatic text analysis applications. It is observed that English-based systems developed for the text analysis applications were not studied extensively for Spanish, which is the second most spoken language. In particular, it seems that the number of studies on multi-class text classification is very small compared to English language. The purpose of this work is to develop classifiers with machine learning methods on a corpus that can be used for Spanish text classification and to perform comparative analysis over different parameters. It is also aimed to calculate the optimum performance effects by measuring the critical parameter values in the methods by applying sensitivity analysis. Spanish corpus was created by preparing a set of 10 different topics from texts of electronic newspapers and magazines. The indexing was achieved according to the topics where the pre-processing steps were completed for the machine learning methods. Naive Bayes, Decision Trees, Maximum Entropy and Decision Support Vector Machines are used. The basic parameters affecting the performance of the classifiers were examined and analyzed. The results of more than 1800 tests indicate that the methods can successfully classify the topics. Sensitivity analysis improves the accuracy of the classifier from 2% to 16%. The methods that yield the best performance have reached an accuracy of 89%, 88% and 87%, respectively. In addition to the accuracy, precision and recall of the test results, the computation time has been integrated to the analysis where the classifier models have been computed.