Media has long influenced our view of the world. One of the things that sets us humans apart is our ability to comunicate with words information which otherwise we would not be able to access. Media is one such way. And this source of information can change drastically in wake of major world events.
Using sentiment analyis I've uncovered some interesting and unexpected changes in the tone of the news we consume on a day to day basis
In order to examine the potential impact of major world events on the tone of day-to-day news, we will employ the following methodology:
- Selection of Major Events: We will carefully choose three significant global events that have had a substantial impact on public discourse and media coverage.
- Time Span and Data Collection: We will focus on a one-week period surrounding each selected major event. During this period, we will gather and analyse the sentiment expressed in the headlines of news articles published across various media outlets. By examining a specific time frame, we aim to capture the immediate effects of the events on news reporting.
- Control Group: To establish a baseline for comparison, we will create a control group consisting of 25 randomly selected days spanning from the beginning of 2020 to the end of 2022. On each of these days, we will include all headlines published, regardless of the topic covered.
The following major events have been selected as pivotal data points:
- Beginning of Lockdown — 15th of May, 2020: The implementation of lockdown measures in response to the global COVID-19 pandemic had a profound impact on societies worldwide.
- US Elections — 3rd of November, 2020: The United States presidential elections are a cornerstone of global political discourse.
- Russian Invasion of Ukraine — 24th of February, 2022: The Russian invasion of Ukraine sent shockwaves throughout the international community.
These events have been chosen for their significant impact, broad media coverage, and potential to elicit distinct reactions and sentiments from news sources. By studying these events, we aim to gain insights into the influence they exert on the tone of day-to-day news.
To ensure accurate sentiment analysis, we employed a model selection process using a small set of headlines. The purpose was to determine which model would provide the most accurate results for our study. For those more technically inclined, here is the code snippet used:
sentiment_analysis = pipeline(model="<model_name>")
sampleHeadlines = [
"No Corrections: April 24, 2023",
"Word of the Day: smug",
"Ron DeSantis Praises U.S.-Japan Ties on Visit to Tokyo",
"Is it Time for Me to Get Over My Dislike of High-Waisted Pants?",
"Book Review: 'Mott Street,' by Ava Chin",
"To Display 500,000 Ants, No Simple Ant Farm Will Do",
After testing several models, the best performing one for this task was j-hartmann/emotion-english-roberta-large · Hugging Face. This model has the capability to identify seven different emotions (anger, disgust, fear, joy, neutral, sadness, surprise), providing sufficient labels to accurately classify most headlines. For the sample headlines provided, the model produced the following results:
- “No Correction: April 24, 2023” — neutral
- “Word of the Day: smug” — joy
- “Ron DeSantis Praises U.S.-Japan Ties on Visit to Tokyo” — joy
- “Is it Time for Me to Get Over My Dislike of High-Waisted Pants?” — disgust
- “Book Review: ‘Mott Street,’ by Ava Chin” — neutral
- “To Display 500,000 Ants, No Simple Ant Farm Will Do” — neutral
Now for the fun part
Let’s examine the sentiment distribution within the control group. Below you can see the distribution of sentiments in the control group:
I will ask you to stop for a moment to appreciate this — despite encompassing all articles written over a one-week period, there are significant changes in sentiment compared to the control group.
Let’s take each event individually and see how much each of the sentiment changed relative to the control group. This way we can get a better understanding of the impact each of these events had.
The most obvious change was a substantial decrease in the emotion of disgust, coupled with a significant increase in joy. The drop in disgust by 2.3% (from 4.1% to 1.8%) and the rise in joy by 3.3% (from 5.5% to 8.6%) indicate a positive shift in sentiment.
This was coupled with a small decrease in fear and sadness, while anger experienced a negligible increase of 4%. There was a considerable increase in surprise as well, which aligns with the closely contested nature of the elections.
Overall, these sentiment changes suggest a more positive tone in the media during the US Elections.
This stands in stark contrast with the US Elections. Fear exhibited a significant increase, accompanied by smaller increases in sadness and neutral sentiments. In fact, this period marked the highest level of fear, coinciding with the lowest level of joy.
Surprisingly, anger experienced a massive 39% decrease, which is unexpected given the circumstances surrounding the lockdown. The decrease in surprise can be attributed to the media’s prior coverage and the gradual build-up leading to the implementation of lockdown measures.
Overall, this period was characterized by elevated levels of fear, sadness, and neutral sentiments, reflecting the challenging and uncertain nature of the lockdown period.
Russian Invasion of Ukraine
This time period can be described by a single word — anger. The substantial increase in anger aligns with the gravity and implications of the invasion. It is interesting to note that the other emotions, such as surprise, sadness, and disgust all suffered minor drops, likely contributing to the increase in anger.
This finding aligns well with the events surrounding the Russian invasion of Ukraine.
The analysis of sentiment changes during the selected events has revealed both expected and surprising findings. It is fascinating to observe that, at a collective level, the media’s tone does undergo significant shifts in response to major events.
As individuals, we have personal emotions and attachments associated with these events, and it is nice to see how those emotions are reflected in the media we consume. The experiment demonstrates the power of sentiment analysis and highlights the potential impact of major events on shaping news narratives and public discourse.
Stay tuned for Part 2, where we delve deeper into the analysis of less significant world events within the same time period.