Deep Dive into Climate Change Attitudes: Classification of Tweets by NLP

Abstract

Introduction

In recent years, social media platforms have become a rich source of information for exploring public opinions and attitudes on a variety of topics. Climate change is one such topic that has recently gained considerable attention in relation to NLP, as highlighted by Stede and Patz [1]. In this project, we propose the use of neural models to classify tweets related to climate change. Our goal is to develop robust and accurate models capable of analyzing attitudes toward climate change. To achieve this goal, we use the available dataset from Kaggle [2]. The dataset contains 60000 tweets about climate changes collected in a span of 10 years. In this dataset, tweets are labeled as believers or non-believers in climate change, but it also includes a sentiment tag to indicate whether the tweet is positive or negative. Thanks to this possibility, we can research the difference between attitude and sentiment.

Targets and Research Questions

In this project, the goal is to use neural models to classify tweets related to climate change. Our secondary goal is to compare transfer learning models to identify the most appropriate model for this specific classification task. Our final goal is to compare the sentiment and attitudes of a given tweet. From the paper by Wilson [3], we can infer that there are differences between sentiment and attitudes toward a phenomenon.

Which transfer learning will help to achieve the best evaluation?

Which neural networks will help qualify tweets into categories?

Is there a relationship between attitude and sentiment?

Methods

Using natural text processing techniques, we will classify the tweets into three groups, namely those that do and do not believe that climate change is a real problem. We use lemmatization instead of stemming in text processing, a study by Balakrishnan and Lloyd-Yemoh [4] showed that lemmatization is a better way of processing text, although the difference was not completely significant. We use base deep neural networks (CNN and RNN) for classification with a combination of transfer learning to achieve higher classification accuracy. From the review paper by Weiss, Khoshgoftaar, and Wang [5]… [ed.: abstract cut due to exceeding the length requirement]

References

[1] M. Stede and R. Patz, "The Climate Change Debate and Natural Language Processing," in Proceedings of the 1st Workshop on NLP for Positive Impact, pp. 8-18, 2021. [Online]. Available: https://doi.org/10.18653/v1/2021.nlp4posimpact-1.2. (accessed May 09, 2023)

[2] “10 Years of Climate Science Denial on RCGroups,” 10 Years of Climate Science Denial on RCGroups | Kaggle. /datasets/rickt15/10-years-of-climate-science-denial-on-rcgroups (accessed May 09, 2023)

[3] T. A. Wilson, "Fine-grained Subjectivity and Sentiment Analysis: Recognizing the Intensity, Polarity, and Attitudes of Private States," in Proceedings of the 1st Workshop on NLP for Positive Impact, pp. 8-18, 2021. [Online]. Available: http://d-scholarship.pitt.edu/7563/1/TAWilsonDissertationApr08.pdf  (accessed May 09, 2023)