Classification of Text on Social Media Data Using the TF-IDF Approach, Word2Vec and Transfer Learning

On social media, there is a lot of information that can be used for further analysis, such as text data, including posts, comments, and reviews. The data has unstructured properties. The problem in this study is the difficulty of analyzing text on social media data to determine standard and non-standard words, so this research utilizes an approach from word representation with TF-IDF, Word2Vec, and transfer learning and using the CNN algorithm. This study produced 3 models in classification performance. In the evaluation model 3 obtained better accuracy compared to model 1 and model 2 because the number of epochs and batch sizes were different. For model 3, the accuracy obtained based on batch values of 64 and epoch values of 20 is 82% for the TF-IDF-CNN model, while the second model uses Word2Vec-CNN with batch values of 128 and epoch values of 60 to obtain 86% accuracy for the Transfer Learning-CNN model with batch values of 256 and epoch values of 100, the accuracy obtained is 91%.