The following is an abridged version of a final report I and two other Brandeis students (Samiya Islam and Yuming Ruan) worked on for our late 2025 deep learning project.
Abstract
This analysis examines how Security and Humanitarian framing strategies have evolved in Reddit immigration discourse from 2015 to 2025. Through mapping frame prevalence against a timeline of major policy events such as including the Trump administration travel bans, the ongoing family separation crisis, and the Biden administration’s policy reversals; we provide empirical evidence of how political discourse responds to government actions and breaking news in real time.
1. Dataset Composition
Our analysis draws from the 5,703 labeled Reddit posts described in Milestone 2, focusing specifically on the prevalence of Humanitarian (11.3% of dataset) and Security frames (11.0%): The former centers on refugee protection, human rights, and compassionate responses to migration crises, while the latter emphasizes terrorism, public safety, and crime concerns.
2. Methodology
2.1 Temporal Aggregation
The model aggregated posts by month to identify broader trends within the data and minimize noise from day-to-day fluctuations. Each month tracked statistics for total number of posts containing each frame, percentage of posts in that month containing each frame, and overall posting activity related to immigration.
2.2 Smoothing Technique
Within the data, a 3-month rolling average balanced noise reduction with temporal sensitivity, preserved short-term changes, filtered out isolated spikes, and maintained sufficient granularity to detect policy-related shifts. Three month windows effectively smooth random fluctuations without obscuring the rapid discourse shifts that can follow major political events; shorter windows retain excessive noise and longer windows mask important short-term dynamics.
3. 10 Key Policy Event Markers
| Date | Event | Type |
| November 13, 2015 | Paris Attacks | Security incident |
| January 27, 2017 | Travel Ban 1.0 | Restrictive policy |
| March 6, 2017 | Travel Ban 2.0 | Restrictive policy |
| September 24, 2017 | Travel Ban 3.0 | Restrictive policy |
| April 6, 2018 | Zero Tolerance/Family Separation | Restrictive policy |
| June 20, 2018 | Family Separation Ends (Executive Order) | Policy reversal |
| July 15, 2019 | Asylum Ban (Third Country Rule) | Restrictive policy |
| March 20, 2020 | Title 42 (COVID-19 border restrictions) | COVID-related policy |
| February 2, 2021 | Biden Reverses Travel Ban | Biden-era reversal |
| May 11, 2023 | Title 42 Ends | Biden-era reversal |
These events span restrictive policies (Travel Bans, Family Separation, Asylum Ban), policy reversals (end of Family Separation, Biden’s Travel Ban reversal), COVID-related measures (Title 42), and major security incidents (Paris Attacks).
4. Key Findings
The Security frame (16.52% average prevalence) appears approximately 75% more frequently than the Humanitarian frame (9.46% average prevalence) within these 10 events. As a result, security-oriented rhetoric has somewhat greater salience in Reddit immigration discourse overall.
4.1 Temporal Peaks
The Security frame reached its maximum prevalence of 100.00% in June 2021. During this month, every single immigration-related post in our curated dataset invoked security concerns. This extraordinary spike likely reflects concentrated discussion around border security during the early Biden administration and reactions to changes in immigration enforcement priorities.
Conversely, the Humanitarian frame peaked at 93.03% in June 2019, when nearly all immigration discourse in scraped posts emphasized refugee protection, human rights, and compassionate responses. This peak coincides with widespread media coverage of detention conditions at the United States-Mexico border, as well as congressional investigations into family separation policies and public outrage over treatment of asylum seekers.
4.2 Temporal Patterns by Period
The first period from 2015 to 2017 established important baseline patterns in immigration discourse prior to significant policy shifts. Following the November 2015 Paris Attacks, both Security and Humanitarian framing increased simultaneously, with Security spiking to 6.36% and Humanitarian rising to 9.78%. By January 2017, coinciding with the first iteration of the Travel Ban, Security prevalence stood at 3.29% and Humanitarian at 4.02%. While early Trump administration policies had begun reshaping discourse, it had not yet triggered the intense polarization that would characterize later periods.
2018 marked a dramatic shift in frame dynamics, driven primarily by the family separation crisis. In January 2018, Security framing registered at 3.28% while Humanitarian stood at 6.67%, already showing humanitarian concerns outpacing security rhetoric. The most significant transformation occurred during and after the family separation policy implementation from April through June 2018. By September, Security framing had collapsed to just 0.88% while Humanitarian maintained a 4.38% share. The humanitarian surge reached its apex in 2019 as border detention conditions gained sustained national attention. In June 2019, Humanitarian framing exploded to an unprecedented 93.03%, representing the most dramatic single-month concentration of any frame in our dataset. This extraordinary spike coincided with congressional visits to detention facilities and widely circulated reports documenting overcrowding, inadequate sanitation, and allegations of mistreatment. The Asylum Ban (Third Country Rule) implemented in July 2019 followed this period of intense humanitarian concern. The pattern suggests that concrete, visceral evidence of impact primarily drives humanitarian frame activation.
Conversely, the COVID-19 pandemic fundamentally disrupted immigration discourse patterns. February 2020 showed relatively balanced framing with Security at 2.43% and Humanitarian at 1.77%. However, by April 2020, our dataset recorded zero immigration-related posts. The implementation of Title 42 in March 2020, which invoked public health authority to expel migrants at the border, coincided with this dramatic reduction in immigration discourse. As discourse gradually resumed during the Biden administration transition, the January and February 2021 period showed moderate attention to both frames (Security at 5.15% then 2.18%; Humanitarian at 1.34% then 1.97%) following Biden’s reversal of the Travel Ban.
The expiration of Title 42 in May 2023 marked another potential inflection point in immigration discourse, though our dataset’s temporal coverage limits our ability to fully capture subsequent shifts. This event, widely anticipated and extensively covered in media reports, represented the formal end of pandemic-era border restrictions and raised expectations of increased migration flows. Future research extending beyond our study period could examine whether the post-Title 42 environment generated frame dynamics similar to earlier crisis periods or whether the discourse had fundamentally changed. Taken together, these temporal patterns reveal that immigration framing on Reddit responds dynamically to external events, with humanitarian concerns showing particular sensitivity to visible human suffering while security framing maintains a more stable but lower baseline presence throughout the decade.
The time-series data reveal several patterns in how policy events and human crises shape framing. Crises drive humanitarian framing more powerfully than formal policy announcements. This suggests that visible suffering, media coverage of harm, and emotionally resonant imagery have greater discursive impact than abstract policy changes. Security events, by contrast, produce more complex and multidimensional effects rather than drive unidirectional securitization, and they are more policy-reactive. Both frames demonstrate delayed response patterns: discourse reacts to policy consequences rather than announcements alone.
4.3 Implications for Model Performance and Generalization
These temporal patterns have important implications for the BERT-based frame detection model developed in Milestone 2. First, although the model was trained on data spanning 2015-2025 and thus captures periods of both high and low framing intensity, the extreme concentration of humanitarian framing in mid-2019 may disproportionately influence learned patterns, potentially biasing the model toward associating certain linguistic features with humanitarian frames based on this anomalous period. Second, the simultaneous elevation of Security and Humanitarian frames following the Paris Attacks (2015) validates our Hypothesis 3 finding that the model successfully captures frame co-occurrence patterns; because real political events often trigger multi-dimensional framing, our multi-label architecture appropriately represents the complexity of actual discourse rather than forcing artificial single-label classifications.
If framing strategies shift substantially over time; for example, if security rhetoric evolves from terrorism-focused language (2015) to border-focused rhetoric (2021) and then models trained on historical data may struggle with contemporary discourse, exhibiting temporal drift in performance. This suggests the need for periodic retraining or temporal adaptation techniques to maintain model accuracy as political language and framing strategies evolve in response to new events, policy changes, and shifting public concerns.
5. Limitations and Considerations
Several limitations constrain the interpretation of these findings. Months with very few posts (e.g., April 2020) can produce extreme percentage values that may not reflect meaningful discourse patterns; for instance, the Security frame’s 100% prevalence in June 2021 likely results from small sample size rather than genuine discursive dominance. Furthermore, these patterns reflect Reddit’s particular user demographics and community norms, meaning that discourse on Twitter, Facebook, or mainstream news media may follow different temporal trajectories and exhibit distinct framing dynamics. Third, our dataset may not extend sufficiently beyond 2023 to capture the full aftermath of Title 42’s expiration or recent policy developments under the Biden administration, limiting our ability to assess the long-term impact of these events.
Finally, while we observe temporal associations between policy events and framing patterns, establishing causal relationships would require more rigorous quasi-experimental designs or natural language processing of explicit causal claims within posts. The current analysis identifies correlations but cannot definitively determine whether policy events directly cause shifts in framing or whether both are driven by underlying political and social factors.
6. Conclusion
This time-series analysis demonstrates that Security and Humanitarian framing in Reddit immigration discourse exhibit substantial temporal variation, with clear peaks associated with specific political events and crises. Humanitarian framing surged dramatically during visible detention crises (2019), while Security framing showed more modest and dispersed elevation across multiple periods. The COVID-19 pandemic temporarily displaced immigration from public discourse, while major policy announcements (Travel Bans, Title 42) produced varied and sometimes delayed framing responses.
These findings validate the multi-label, temporally-sensitive approach taken in Milestone 2, demonstrating that immigration discourse is inherently dynamic and multidimensional. Future work should explore these temporal patterns in greater depth, examining whether specific policy events trigger predictable framing shifts and whether the BERT model’s performance varies across different time periods. Understanding these dynamics is essential for both academic research on political communication and practical applications in real-time discourse monitoring.
7. References
Bail, C. A. (2016). Combining natural language processing and network analysis to examine how advocacy organizations stimulate conversation on social media. Proceedings of the National Academy of Sciences, 113(42), 11823-11828.
Chong, D., & Druckman, J. N. (2007). Framing theory. Annual Review of Political Science, 10, 103-126.
Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. NAACL-HLT, 4171-4186.
Entman, R. M. (1993). Framing: Toward clarification of a fractured paradigm. Journal of Communication, 43(4), 51-58.
Huang, J. T., Choi, J., & Wan, Y. (2024). Political bias in Reddit moderation and the formation of echo chambers.
King, G., Lam, P., & Roberts, M. E. (2017). Computer-assisted keyword and document set discovery from unstructured text. American Journal of Political Science, 61(4), 971-988.
Küçük, D., & Can, F. (2020). Stance detection: A survey. ACM Computing Surveys, 53(1), 1-37.
Lakoff, G. (2004). Don’t think of an elephant! Know your values and frame the debate. Chelsea Green Publishing.
Wehrmann, J., Cerri, R., & Barros, R. (2018). Hierarchical multi-label classification networks. ICML, 5075-5084.
Yang, P., Sun, X., Li, W., Ma, S., Wu, W., & Wang, H. (2018). SGM: Sequence generation model for multi-label classification. COLING, 3915-3926.
8. Acknowledgments
We thank the course staff of BUS244A for guidance throughout this project. Computational resources were provided by Brandeis University. We acknowledge HuggingFace for providing pre-trained BERT models and the scikit-learn developers for machine learning infrastructure.
- Resources Used: Claude (for coding)
- Website: https://jojo142.github.io/US-Immigration-Sentiment-Trend-using-Reddit/
- Code Repository: https://github.com/jojo142/US-Immigration-Sentiment-Trend-using-Reddit