an experiment to compare the performance of four different machine learning models on a corpus that was processed by four natural language processing models. The machine learning models we used were random forest, XGBoost, logistic regression, and support vector machine (SVM), while the NLP models we employed were word2vec, GloVe, tf–idf, and text bleaching.
Our results showed that the random forest model delivered the highest accuracy at 75%, with XGBoost coming in second with an accuracy of 73%. In terms of the NLP models, word2vec had the best result with an accuracy of 72%, followed by GloVe with 68%.
However, we also found that when we changed the context of the corpus, the accuracy of the models changed. Specifically, when we used text bleaching method to remove context-specific information from the corpus, the accuracy of the models dropped less significantly than when we used word2vec. In fact, the accuracy drop was only 5% with text bleaching compared to 10% with word2vec.