Source-linked AI summary
Divergent discourse between protests and counter-protests: #BlackLivesMatter and #AllLivesMatter
Ryan J. Gallagher, Andrew J. Reagan, Christopher M. Danforth, Peter Sheridan Dodds
TL;DR
The paper asks how #BlackLivesMatter and #AllLivesMatter discourses diverge and whether counter-protest activity hijacked Black Lives Matter conversations. Using multi-level analysis of their tweets, it finds pro-law-enforcement opposition in #AllLivesMatter and substantial hijacking by #BlackLivesMatter advocates, while counter-protest discussion remains concentrated in #AllLivesMatter.
Problem
The paper examines the under-studied, data-driven question of how #AllLivesMatter discourse diverges from #BlackLivesMatter and whether it hijacked #BlackLivesMatter conversations.
Method
The study applies word-level and topic-level analysis to the full text of tweets containing #BlackLivesMatter and #AllLivesMatter.
Results
#AllLivesMatter supports pro-law-enforcement sentiments and facilitates opposition with #PoliceLivesMatter and #BlueLivesMatter, while a significant portion of its use reflects hijacking by #BlackLivesMatter advocates.
Takeaways & Limitations
The findings suggest that Black Lives Matter avoided derailment by concentrating discussion of counter-protest opinions in #AllLivesMatter rather than in #BlackLivesMatter.
Abstract
from arXiv · showhide
Since the shooting of Black teenager Michael Brown by White police officer Darren Wilson in Ferguson, Missouri, the protest hashtag #BlackLivesMatter has amplified critiques of extrajudicial killings of Black Americans. In response to #BlackLivesMatter, other Twitter users have adopted #AllLivesMatter, a counter-protest hashtag whose content argues that equal attention should be given to all lives regardless of race. Through a multi-level analysis of over 860,000 tweets, we study how these protests and counter-protests diverge by quantifying aspects of their discourse. We find that #AllLivesMatter facilitates opposition between #BlackLivesMatter and hashtags such as #PoliceLivesMatter and #BlueLivesMatter in such a way that historically echoes the tension between Black protesters and law enforcement. In addition, we show that a significant portion of #AllLivesMatter use stems from hijacking by #BlackLivesMatter advocates. Beyond simply injecting #AllLivesMatter with #BlackLivesMatter content, these hijackers use the hashtag to directly confront the counter-protest notion of "All lives matter." Our findings suggest that Black Lives Matter movement was able to grow, exhibit diverse conversations, and avoid derailment on social media by making discussion of counter-protest opinions a central topic of #AllLivesMatter, rather than the movement itself.
I. INTRODUCTION
The paper uses multi-level analysis of over 860,000 tweets to compare #BlackLivesMatter and #AllLivesMatter discourse. It finds pro-law-enforcement opposition and substantial hijacking within #AllLivesMatter, while discussion of counter-protest opinions remains concentrated there.
- I. INTRODUCTION: #AllLivesMatter advocates promote equal attention to all lives, while #BlackLivesMatter supporters argue that this race-neutral framing can derail the movement.The paper presents #AllLivesMatter as a relatively understudied counter-protest hashtag and addresses whether it hijacked #BlackLivesMatter conversations.
- I. INTRODUCTION: The study analyzes over 860,000 tweets to quantify how #BlackLivesMatter and #AllLivesMatter discourses diverge.Its methods use the entirety of textual data rather than only hashtag trends or curated term lists.
- I. INTRODUCTION: #AllLivesMatter diverges through pro-law-enforcement sentiments, placing #BlackLivesMatter in opposition to #PoliceLivesMatter and #BlueLivesMatter.The authors characterize this opposition as echoing historical depictions of tensions between Black protesters and law enforcement.
- I. INTRODUCTION: A significant portion of #AllLivesMatter use reflects hijacking by #BlackLivesMatter supporters, who directly interrogate the phrase’s counter-protest stance.By contrast, #BlackLivesMatter contains richer and more diverse conversations, with hijacking representing a much smaller portion.
A. Data Collection
The dataset contains tweets using both hashtags from August 2014 through August 2015, sampled through Twitter’s 10% Gardenhose feed. Analysis is restricted to eight one-week periods when both hashtags simultaneously peaked.
- A. Data Collection: The collection covers August 8, 2014 through August 31, 2015 and includes tweets containing either hashtag, case-insensitively.Tweets containing both hashtags appear in each corpus during analysis.
- A. Data Collection: The sample includes 767,139 #BlackLivesMatter tweets and 101,498 #AllLivesMatter tweets from Twitter’s 10% Gardenhose feed.The tweets came from 375,620 and 79,753 unique users, respectively; 23,633 tweets contained both hashtags.
- A. Data Collection: The study restricts analysis to eight one-week periods with simultaneous spikes in #BlackLivesMatter and #AllLivesMatter use.The selected periods correspond to major events including the non-indictments of Darren Wilson and Daniel Pantaleo, police deaths, shootings, and protests over Freddie Gray and Sandra Bland.
B. Entropy and Diversity
The paper uses Shannon entropy to characterize textual unpredictability and diversity. It converts entropy into effective diversity so diversity ratios can be compared linearly across texts.
- B. Entropy and Diversity: Shannon entropy H measures the unpredictability of a text from the probabilities of its n unique words.Higher entropy indicates a less predictable text.
- B. Entropy and Diversity: The paper interprets higher Shannon entropy as higher linguistic diversity.This raw diversity measure is used to examine the language surrounding individual words.
- B. Entropy and Diversity: The Shannon index gives equal weight to common and rare words, but diversity comparisons require care even when the same index is used.The authors therefore convert it to effective diversity for linear comparisons.
- B. Entropy and Diversity: Effective diversity D is the perplexity of a text and supports ratio-level comparisons between texts.For equally likely vocabularies of sizes n and 2n, effective diversity doubles, unlike raw Shannon entropy.
C. Jensen-Shannon Divergence
The paper uses Jensen-Shannon divergence to compare word distributions between hashtag corpora while avoiding the infinite values that can arise with Kullback-Leibler divergence. It decomposes overall divergence into word-level contributions.
- C. Jensen-Shannon Divergence: Kullback-Leibler divergence measures distributional differences between two texts but can become infinite when a word appears in only one text.Because this situation is plausible in Twitter data, the study uses Jensen-Shannon divergence instead.
- C. Jensen-Shannon Divergence: Jensen-Shannon divergence smooths Kullback-Leibler divergence using a mixed distribution M weighted by the sizes of texts P and Q.The weights π1 and π2 sum to 1.
- C. Jensen-Shannon Divergence: Jensen-Shannon divergence is bounded from 0 to 1, equaling 0 for identical word distributions and 1 when the texts share no words.Its linearity permits extracting each word’s contribution to total divergence.
- C. Jensen-Shannon Divergence: A word contributes zero divergence exactly when its probabilities are equal in the two texts.When its contribution is nonzero, the larger probability identifies which text contributes more to the divergence.
A. Word-Level Divergence
The authors compare #BlackLivesMatter and #AllLivesMatter language using Jensen-Shannon divergence and word-level diversity across selected periods. The divergence reflects distinct event focuses, law-enforcement alignment in #AllLivesMatter, and hijacking by #BlackLivesMatter supporters.
- Divergence method: Jensen-Shannon divergence ranks words by their percentage contribution to differences between the two hashtag corpora.Tweets are represented as bags of words after removing handles, links, punctuation, stop words, retweet markers, and both hashtags.
- Divergence method: Word-shift bars identify which hashtag uses each word more, while shading indicates the diversity of language surrounding it.Lighter shading reflects popular retweets; darker shading reflects use across many different tweets.
- Event-focused divergence: #BlackLivesMatter diverges through proportionally greater discussion of deaths including Ferguson, Freddie Gray, Eric Garner, Walter Scott, and Sandra Bland.These event-related terms contribute to divergence across the examined periods.
- Law-enforcement alignment: During important protest periods, #AllLivesMatter diversifies around law-enforcement lives, including #policelivesmatter, #bluelivesmatter, and the deaths of NYPD officers.This produces a law-enforcement-aligned response amid Black Lives Matter protests.
- Hijacking: Words suggesting engagement with structural racism and police brutality entered #AllLivesMatter through hijacking by #BlackLivesMatter supporters criticizing that absence.The same pattern included critical material confronting the notion of “All Lives Matter.”
B. Topic Networks
The topic-network analysis finds that #BlackLivesMatter supports more numerous and diverse conversations, while #AllLivesMatter is more tightly clustered around law-enforcement themes and is more heavily shaped by hijacking.
- Network construction: The topic networks visualize hashtags as nodes connected by weighted co-occurrence edges, with the network core extracted using the disparity filter.Node sizes represent hashtag-use frequency, and colors provide a visual guide to detected communities.
- Topic-network structure: #BlackLivesMatter contains more topics and generally lower clustering than #AllLivesMatter, indicating more diverse connections.#AllLivesMatter’s stronger clustering indicates conversations that are more tightly connected and revolve around similar themes.
- Topic centrality: The most central topics track period-specific events, but major racial-justice hashtags consistently rank higher in #BlackLivesMatter than in #AllLivesMatter.Examples include #mikebrown, #ericgarner, #icantbreathe, #freddiegray, #baltimore, and #sandrabland.
- Law-enforcement alignment: #AllLivesMatter ranks #nypd, #policelivesmatter, and #bluelivesmatter higher than #BlackLivesMatter during several periods.By contrast, anti-police hashtags such as #killercops, #policestate, and #fuckthepolice appear almost exclusively in #BlackLivesMatter.
- Hijacking and network asymmetry: #AllLivesMatter networks show more limited scope, while #BlackLivesMatter is more central within #AllLivesMatter than the reverse.This asymmetry indicates that hijacking is more prevalent within #AllLivesMatter, while #BlackLivesMatter users delegate hijacking to only a portion of their discourse.
C. Conversational Diversity
The diversity analysis shows that #BlackLivesMatter sustains substantially broader lexical and hashtag variety than #AllLivesMatter, even after accounting for differing tweet volume.
- Lexical diversity: #BlackLivesMatter has larger lexical diversity in eight of ten months, averaging 5% more than #AllLivesMatter.The two periods favoring #AllLivesMatter coincide with large non-police-involved shootings of people of color, when the hashtag was used in solidarity.
- Hashtag diversity: #BlackLivesMatter’s average hashtag diversity is six times that of #AllLivesMatter.This result aligns with the expansive #BlackLivesMatter topic networks and tightly clustered #AllLivesMatter networks.
- Temporal robustness: #AllLivesMatter’s low hashtag diversity remains relatively constant, despite changing conversations across different time periods.Both hashtags overlap on major topics, but #BlackLivesMatter contains far more diverse topics even after accounting for volume.
IV. DISCUSSION.
The discussion finds that #AllLivesMatter centers opposition involving law enforcement while also serving as a venue for #BlackLivesMatter advocates to contest the counter-protest stance. This containment helped #BlackLivesMatter sustain diverse conversations without derailment.
- #AllLivesMatter includes some discussion of Black deaths, but these signs of solidarity are often low-diversity and other periods largely lack discussion of non-Black deaths.The Chapel Hill shooting period is identified as the exception showing discussion of non-Black deaths.
- #AllLivesMatter significantly discusses law-enforcement lives, especially during heavy protesting, reinforcing historically contentious framings of Black protesters versus police.The authors connect this opposition to portrayals of police and protesters as “enemy combatants” and claims that movements jeopardize law-enforcement lives.
- A significant portion of #AllLivesMatter use reflects hijacking by #BlackLivesMatter supporters who directly interrogate the phrase’s underlying worldview.The authors distinguish this from simple content injection: hijackers use #AllLivesMatter to confront the counter-protest notion itself.
- #BlackLivesMatter supporters’ discussions of #AllLivesMatter were largely relegated to the counter-hashtag, allowing #BlackLivesMatter to retain diverse conversations.The discussion presents this pattern as evidence that #BlackLivesMatter countered content injection rather than being derailed by it.
- Analyzing complete text structure at word and topic levels, rather than only hashtag trends or curated term lists, helps unpack divergent protest narratives.The authors present this approach as generalizable to comparing discourse across protest movements and political polarization.
Appendix B: Hashtag Topic Networks
Appendix B documents the construction and contents of hashtag topic networks for #BlackLivesMatter and #AllLivesMatter across the paper’s periods of interest. It also reports filtering and network-statistic summaries used to inspect these networks.
- Network filtering: The appendix shows the percentage of each original hashtag network retained under varying disparity-filter significance levels.It notes a sharp size reduction by approximately one factor when moving from α = 0.03 to α = 0.02 in two specified networks.
- Network statistics: Summary-statistics tables describe topic networks created from full hashtag networks using disparity-filter significance levels α = 0.04 and α = 0.05.The appendix separately labels the two significance-level settings in Tables B1 and B2.
- Hashtag rankings: Appendix tables rank topic-network hashtags by betweenness centrality and PageRank, with some #AllLivesMatter networks containing fewer than ten nodes.The smaller networks are attributed to their relatively small size; one example has #blacklivesmatter as the only node visited for betweenness paths.
- Topic-network periods: Topic-network figures pair #BlackLivesMatter and #AllLivesMatter for periods spanning non-indictments, shootings, deaths, and the Baltimore protests.The listed networks cover Darren Wilson, Daniel Pantaleo, Chapel Hill, Walter Scott, Baltimore, Charleston, and Sandra Bland periods.