With the ever-rising volumes of available data, there is a pressing need to organize it. This leads to defining several modern classification problems that often involve the prediction of multiple labels simultaneously associated with a single instance which is known as Multi-Label Classification.
We investigated multi label document classification with the help of pre-trained embeddings. For the dataset, we used the Reuters news wire dataset which contains news articles distributed over 90 classes. Each news article belongs to either one or multiple classes. We leveraged the help of pre-trained word embeddings to finetune our neural network.
The results of our proposed model didn’t perform as expected but we were still able to get some valuable insights which has significant contribute to multi label document classification’s research. The Python notebook demonstrated how Multi-Label Document Classification was performed.