This repository aims to depositary the projects for text classification tasks in diverse contexts of the course designs. In this repository, there are models developed for the two tasks temporarily.
The project files lie in the folder Lyrics_News. Aiming to recognize the artist of one specific song through the lyric, the text classification models are developed.
However, a lot of feature enigneering methods and deep learning models show a poor performance on the lyric data, which can only be used for music recommendation from the lyrics. After that, the above models and methods are used for recognize the news genre from one specific news report, and show a remarkable performance.
About the data, the crawler inside the folder Lyrics-Cralwer is used for getting the lyric data from Netease Music. In this project, 41 most famous artists in recent three decades are selected with the song lyrics in all their formal albums.
All news data is from the public dataset which collected HuffPost news reports with categories from 2012 to 2018.
The project files lie in the folder Job_Postings. Basically, the feature engineering methods and the models used in this project are similar to the former, while an oversampling method is used in this project for settling the extremely unbalanced dataset.
The dataset is available on Kaggle. However, the dataset is extremely unbalanced. There are 17014 real job postings, and just 866 fake postings.