The data story : https://claraledraoulec.github.io/BLM-Data-Story/
In 2013, the Black Lives Matter (BLM) movement emerged in response to the police brutality experienced by Black people in the United States. The movement has become increasingly popular, presumably due to the myriad protests that have been organized since its inception. However, its popularity has sparked mixed sentiment within America’s socio-political landscape. Thus, we analyze how select mediatized protests have impacted discourse on the movement. We focus on the fact that although a civilan is killed by the police EVERY SINGLE DAY, the media only sensationalizes certain cases, and then moves on and stops discuissing the problem. Additionally, we show how too much attention is given to "counter-movements" and we see the efforts to diminsh the impact of Black Lives Matter.
To do so, we do an analysis of occurrences of BLM movement and police brutality in the dataset at specific periods of time.
Analyze the mediatization of BLM after a police officer killed a black person:
- How does each kill affect and how often BLM is talked about?
- Do these events cause a positive or negative shift in how BLM is viewed?
- How do demographics affect how each speaker reacts to an event?
- What are the most impactful events/crime and why?
We finally used only 1 extra dataset to introduce and motivate the project. The dataset was found on the Washington Post website: https://www.washingtonpost.com/graphics/investigations/police-shootings-database/ . It contains all the persons shoot by US police between 2015 and today.
- Establish a list of interesting keywords and interesting people to search in the quotations
- Quotebank dataset: explore data, extract all related quotes, visualization of amount of relevant quotes over time, are there any peaks?
- Speaker dataset: explore dataset, find relevant demographic information of all speakers in our related quotes dataset (race, ethnicity, gender identity, age, political party affiliation, socioeconomic status?, occupation - politician, activist, other).
- Pick the most cited black persons who were killed by US police to perform some correlations with the found peaks of relevant quotes.
- NLP
- Sentiment Analysis of quotes
- Sementic analysis based on lexical categories and topic detection
- Find relevant topics, keywords and relevant people to analyze and correlate with the quotes dataset
- Clustering
- Cluster people into categories based on demographics and if their sentiment changed in positive or negative way (we will come up with a quantitative way to determine this) for each event
- Display demographics of our speakers to show who is active in the conversation surrounding the movement.
- Visualization to introduce the datastory (map and some facts)
- Make conclusions about our findings, figure out a way to present results
By Friday November 19: Complete (a) in Methods. Complete (b) in Methods for the Quotebank dataset with final cleaned version.
By Friday November 26: Complete (b) in Methods for all external datasets we will use, decide more formally the role of each one in the project.
By Friday December 3: Have started (c) (i) and (ii) in methods. More specifically, find a model for sentiment analysis NLP (so we don’t have to train it all ourselves, not realistic) and begin setting up any functions needed for clustering and try some examples.
By Friday December 10: Finish modeling NLP and clustering (c)
By Friday December 17: Complete data visualization of our model and our findings on the datasets (d)
By Friday December 24: Discuss analysis of results and make final conclusions (e)
Over Break: Write Data Story (f)
- Together: Find preliminary set of significant events/people
- Clara & Danae: Establish list of keywords, first by ourselves, an then by topic detection but it didn't get us anywhere...
- Paula & Danae: Speaker dataset: get it cleaned and have all important visualizations done, get it prepped for combining with Quotebank
- Clara & Hugo: Quotebank Dataset: get it cleaned and have all important visualizations done, get it prepped for modeling
- Hugo & Paula: Analyze if the extra dataset can be useful and make some visualizations with to introduce the data story
- Paula & Danae: NLP, obtain a usable sentiment analysis. After many iterations to improve the model, it wasn't accurate enough and we were not able to make conclusion
- Danae: Sementic analysis based on lexical categories in order to improve the selected keywords
- Clara & Danae & Hugo : Name entity recognition to select by hand most relevant cited names associated with BLM in order to perform correlations between the death of some of these people and the occurence of our quotations
- Hugo & Clara: Perform clustering and regressions on speakers attributes but it did not yield anything interpretable...
- Paula & Clara & Danae: time-series, visualize change in sentiment as a function of time
- Hugo & Clara: Make visualizations with the speakers attributes and the total generated quote dataset
- All together
- All together
There are 4 main notebooks, .
introduction_visualization.ipynb: This is the code to generate the few visualizations and facts in the introduction.quotes_analysis.ipynb: This notebook is the main code used to deal with our selected quotes.data_wrangling.ipynb: Notebook used to extract quotations from the quotebank.aspect_based_sentiment.ipynb: Notebook containing our attemps for the sentiment analysis.
Unfortunately, as explained in the aspect_based_sentiment.ipynb notebook, we were not able to carry out our project as initially hoped. Our analysis had to be based on sentiment analysis and we tried different methods to make it work but even after several trials, we could not find a solution. Indeed, we realized that our keywords like "BLM", "police brutality" had very often a negative connotation. Some quotations that had a positive meaning for us, also gave a negative sentiment score, depending on how the sentence was turned. This completely distorted our results and we were not able to perform certain methods and analysis correctly. We tried to find some alternative solutions, this is why we changed our initial plan slightly.