Effects of Preprocessing in Word Embedding

"A Comprehensive Analysis of Preprocessing for Word Representation Learning in Affective Tasks"

"The Role of Preprocessing for Word Representation Learning in Affective Tasks"

Contributions

1. We conduct a comprehensive analysis of the role of preprocessing techniques in affective tasks (including sentiment analysis, emotion classification and sarcasm detection), employing different models, over nine datasets.
2. We perform a comparative analysis of the accuracy performance of word embedding models when preprocessing is applied at the training phase (training corpora) and/or at the downstream task phase (classification dataset).
3. We evaluate the performance of our best preprocessed word vector model against state-ofthe-art pretrained word embedding models.

List of Preprocessing Factors:

1. Punctuation (Punc)
2. Spelling correction (Spell)
3. Stemming (Stem)
4. Stopwords removal (Stop)
5. Negation (Neg)
6. Pos-Tagging (POS)

Training Models:

1. Word2Vec (Skip-gram)
2. Word2Vec (CBOW)
3. BERT (Feature-based Approach)

Datasets used for Training Word Embeddings:

1. News : https://www.kaggle.com/snapcrack/all-the-news
2. Wikipedia : https://www.kaggle.com/jkkphys/english-wikipedia-articles-20170820-sqlite

Datasets used at downstream Tasks for Classification:

a) Sentiment Analysis:

1. IMDB : http://ai.stanford.edu/ amaas/data/sentiment/
2. Semeval 2016 : http://alt.qcri.org/semeval2016/task4/index.php
3. Airlines : https://www.kaggle.com/crowdflower/twitter-airlinesentiment
4. SST-5: https://nlp.stanford.edu/sentiment/index.html

b) Emotion Detection:

1. SSEC: http://www.romanklinger.de/ssec/
2. ISEAR : https://github.com/PoorvaRane/Emotion-Detector/blob/master/ISEAR.csv
3. Alm : http://people.rc.rit.edu/~coagla/affectdata/index.html

c) Sarcasm Detection:

1. Onion : https://github.com/rishabhmisra/News-Headlines-Dataset-For-Sarcasm-Detection
2. IAC : https://nlds.soe.ucsc.edu/sarcasm2
3. Reddit : https://nlp.cs.princeton.edu/SARC/0.0/

Reference: Nastaran Babanejad, Ameeta Agrawal, Aijun An and Manos Papagelis, A Comprehensive Analysis of Preprocessing for Word Representation Learning in Affective Tasks, Proceedings of The 58th Annual Meeting of the Association for Computational Linguistics (ACL 2020).