There are many application scenarios for text classification technology in the company. Whether it’s in customer service, risk management, or investment domains, building text classification models can help companies automate processes and thus save a significant amount of labor costs. When we want to build a text classification model, we must first convert unstructured text into text features that the model can use. The key to the model’s effectiveness lies in the way text features are processed. In this event, we will take as an example the 30,000 comments data from 19 banking applications on Google Play. We will delve into different methods of processing text features, including term frequency matrices, TF-IDF, word2vec, large language models, and fine-tuning large language models, to compare their respective advantages and disadvantages as well as the differences in subsequent model effectiveness.

The data for this competition consists of approximately 30,000 review entries from 19 banks in Taiwan on their Google Play apps. You can try as many approaches as possible and upload your predictions through machine learning/deep learning models on the day of the event, which is on Saturday, 6/22.