Text Preprocessing in Neural Text Classification

Jose Camacho Collados and Mohammad Taher Pilehvar

The following repository includes the pre-trained word embeddings and preprocessed text classification datasets for the paper On the Role of Text Preprocessing in Neural Network Architectures: An Evaluation Study on Text Categorization and Sentiment Analysis .

Pre-trained word embeddings

We release the 300-dimension word embeddings used in our experiments as binary bin files. The embeddings were trained on the UMBC corpus with the following preprocessing techniques:

• Vanilla (simple tokenization): Download here [~1.8 GB]
• Lowercased: Download here [~1.6 GB]
• Lemmatized: Download here [~1.7 GB]
• Multiword-grouped: Download here [~2.1 GB]

Preprocessed datasets

We also release the text categorization and sentiment analysis datasets already preprocessed:

• Text categorization: Available here
• Sentiment analysis: Available here

Note 1: If you use any of these datasets, please acknowledge the original sources (you can find them in the reference paper).
Note 2: For each class file in the dataset directories, each line corresponds to an instance in the corpus, be it a phrase, sentence or document (depending on the dataset).

Code

The code to run our experiments is available in the following complementary repository: https://github.com/pilehvar/sensecnn

Reference paper

If you use any of these resources, please cite the following paper:

@InProceedings{camacho:preprocessing2018,
  author = 	"Camacho-Collados, Jose and Pilehvar, Mohammad Taher",
  title = 	"On the Role of Text Preprocessing in Neural Network Architectures: An Evaluation Study on Text Categorization and Sentiment Analysis",
  booktitle = 	"Proceedings of the EMNLP Workshop on Analyzing and interpreting neural networks for NLP",
  year = 	"2018",
  publisher = 	"Association for Computational Linguistics",
  location = 	"Brussels, Belgium"
}