Paper Review 8: Lexicon-Based Methods for Sentiment Analysis

In this post, the paper “Lexicon-Based Methods for Sentiment Analysis” is summarized.

Link to paper: https://www.mitpressjournals.org/doi/abs/10.1162/COLI_a_00049

Maite Taboada, Julian Brooke, Milan Tofiloski, Kimberly Voll, Manfred Stede, 2011, “Lexicon-Based Methods for Sentiment Analysis,” in Computational Linguistics, Volume 37, Issue 2, p.267–307

In this article, Sentiment Analysis and Semantic Orientation Calculation is done with lexicon-based methods. Using a hand-crafted dictionary from a specific corpus and assigning semantic orientation scores to words to calculate semantic orientation values of sentence inputs, via analyzing the sentiments, is conducted in this research. Interestingly this research appeared to be successful than the others in the field, thanks to its effectiveness in other datasets from different topics, caused by the careful analysis and alignment of words by the classifier in the input sentences. Applying tricky methods like using shift instead of negation of semantic scores and irrealis blocking (excluding irrelevant words), the classifier managed to carry its success to the other corpora mostly.

In addition to that, the authors not only verified their models’ success with other test datasets, but also changed their training dataset as well. Still, the success of the classifier was mostly preserved in this new configuration. After that, they decided to verify their training data -the dictionary- and assigned semantic orientation scores so that authors will be sure of the success at both sides of the work. To do this, they employed the Amazon Mechanical Turk service to manually verify their own semantic orientation scores, considering both polarity and strength values. In this way, the used dictionary is also verified by human beings which had shown the quality of the research in this article.

To conclude, as an academic article published in 2011 by a MIT Press Journal -Computational Lingustics- with more than thousand citations, this research is one of the important papers which provided a successful alternative to machine learning by lexicon-based methods in the field of natural language processing, specifically semantic orientation calculation and sentiment analysis as a classifier application.