SentimentAnalysisForClimateChange

Syntactic and sentiment analysis on climate change-related news and tweets.

Code authors: Gabirel Ichcanziho Pérez Landa and Jorge Francisco Ciprián Sánchez.
Paper authors: Gabriel Ichcanziho Pérez Landa and Jorge Francisco Ciprián Sánchez.
Date of creation: May 10th, 2021.
Code Author Email: ichcanziho@outlook.com

Overview

We employ natural language processing techniques such as sentiment and syntactic analysis to analyze selected news and tweets datasets, obtaining the prevailing sentiment and structure of climate change-related news and tweets. We then extract and compare the most relevant words associated with positive, negative, and neutral feelings in both datasets. Finally, we visualize their change over time to extract relevant patterns that can provide insight into the evolution of the climate change-related narrative. For more details on the implementation and results, see the paper here.

System specifications

This work was developed and tested on Ubuntu 16.04.6 LTS with Python 3.6.9, and Windows 10 with Python 3.8.1.

We employ VADER and TextBlob for sentiment analysis and spaCy for syntactic analysis. We include a requirements.txt file for ease of use; to install the required dependencies run the following command:

$ pip install -r requirements.txt

Data

We analyze the following publicly available datasets:

A New Dataset For Exploring Climate Change Narratives On Television News 2009-2020: This dataset contains 95,000 entries for news that contain the keywords climate change or global warming or climate crisis or greenhouse gas or greenhouse gases or carbon tax across four stations: BBC News, CNN, MSNBC, and Fox News, from 2009 to 2020. Each entry includes the UTC time, the station, the show, a 15 second clip of the captioning containing the mention, and a URL to the website to view the specific clip containing it.
Twitter Climate Change Sentiment Dataset: This dataset, generated by the University of Waterloo, contains tweet IDs on the subject of climate change collected between April 27, 2015 and February 21, 2018. The dataset contains a total of 43,943 tweets. The authors labeled the tweets according to four categories related to climate change belief or denial. For the purposes and scope of the present project, we only consider the tweet text and date, as we perform our own syntactic and sentiment analysis.

Data pre-processing

We filter the entries by the subject of climate change and adjust the news dataset so that it contains entries for the same period of the tweets one, also removing the entries for the BBC news chain, as these entries start in the year 2017. After performing the said filters, we obtain 16,124 news entries and 28,116 tweet entries.

We provide the raw datasets in the inputs directory and the pre-processed datasets in the outputs folder. The datasets containing the results for the syntactic and sentiment analysis are the outputs/clean datasets/News_syntactic_sentiment.csv and outputs/clean datasets/Twitter_syntactic_sentiment.csv files.

Running the program

We provide 5 options: dataset cleaning, sentiment analysis, syntactic analysis, plotting the results' timelines, and all operations. You can select the operation you with to carry on the config.ini file, changing the following line:

; Mode options: CLEAN, SENT, SYNT, PLOTS, CC, ALL.
[MODE]
mode = SENT ; change this line.

Once you have set the option you wish to carry out, you simply need to run the following command:

$ python main.py