NLP-based Message Prioritisation for Opportunistic Routing: A Case Study of Nepal Earthquake Emergency Communication

Yihang Cao; Milena Radenkovic

cs.NIOct 27, 2026

NLP-based Message Prioritisation for Opportunistic Routing: A Case Study of Nepal Earthquake Emergency Communication

Authors: Yihang Cao, Milena Radenkovic

Organizations: School of Computer Science, University of Nottingham, Nottingham, United Kingdom

Abstract

This dissertation investigates whether Natural Language Processing (NLP) based semantic message prioritisation can improve the treatment of urgent disaster messages in opportunistic emergency communication, using the 2015 Nepal earthquake as a case study. TF IDF with Logistic Regression and BERT are used to classify social media messages into nine humanitarian categories, which are mapped to four routing priorities from P1 to P4. These priorities are integrated into the ONE simulator within a controlled Kathmandu earthquake scenario using a modified Binary Spray and Wait router. BERT achieves higher priority accuracy and P1 recall than Logistic Regression, while reducing under prioritisation and severe priority errors. However, improved NLP performance does not produce uniform network gains. P1 delivery probability remains high and similar across routing conditions, and no consistent latency reduction is observed. The clearest difference appears under buffer pressure, where the BERT condition shows fewer observed ground truth P1 buffer losses than Logistic Regression. The results indicate that semantic prioritisation is most valuable when limited transmission and storage resources are contested, although it cannot overcome mobility and contact constraints in opportunistic networks.

Figures & tables

Fig. 1: Overall NLP–routing architecture for the Nepal earthquake case study.

Dataset Characteristic	Value
Cleaned messages	3,003
Training messages	2,402
Test messages	601
Training proportion	80%
Test proportion	20%
Humanitarian categories	9

TABLE I: Nepal Earthquake Dataset and Split Summary.

Fig. 2: Distribution of humanitarian message categories in the cleaned Nepal earthquake dataset.

Fig. 3: Text preprocessing workflow for the Nepal earthquake social media messages.

Setting	Logistic Regression	BERT
Input representation	TF–IDF vectors	Token sequences
N-gram range	(1, 2)	–
Minimum document frequency	2	–
Maximum document frequency	0.9	–
Sublinear TF scaling	Enabled	–
Maximum iterations	2,000	–

TABLE II: NLP Classifier Configuration

Humanitarian Category	Level	Priority
Injured or dead people	High	P1
Missing, trapped, or found people	High	P1
Displaced people and evacuations	Medium-high	P2
Infrastructure and utilities damage	Medium-high	P2
Caution and advice	Medium-high	P2
Donation needs, offers, or volunteering services	Medium	P3

TABLE III: Mapping from Humanitarian Categories to Routing Priorities

Notation	Meaning
$i$	Index of a disaster message.
$C$	Set of nine humanitarian message categories.
$y_{i}$	Ground-truth humanitarian category of message $i$ .
$\hat{y}_{i}^{(q)}$	Category predicted for message $i$ by classifier $q$ .
$q$	NLP classifier, where $q\in\{\mathrm{LR},\mathrm{BERT}\}$ .
$\pi(c)$	Fixed mapping from category $c$ to routing priority.

TABLE IV: Main Notation Used in the Paper

Fig. 4: Execution of the Kathmandu earthquake communication scenario in the ONE simulator.

Group	Nodes	Addresses	Speed (m/s)	Buffer
Residents	60	0–59	0.3–1.2	5 MB
Rescue workers	20	60–79	1.0–2.5	20 MB
Rescue vehicles	15	80–94	5.0–12.0	50 MB

TABLE V: Node Groups Used in the Kathmandu Earthquake Scenario

Parameter	Value
Simulation duration	43,200 s
Transmission range	50 m
Transmission speed	50 KB/s
Message size	100–500 KB
Message interval	30–60 s
Offered messages	959

TABLE VI: Kathmandu Earthquake Scenario Main Communication and Traffic Parameters

Condition	Priority Used	Purpose
Standard	None	Baseline
Oracle	Ground truth	Ideal reference
Logistic Regression	LR prediction	Predicted priority
BERT	BERT prediction	Predicted priority

TABLE VII: Routing Conditions Used in the Evaluation

Evaluation Level	Metric	Purpose
Humanitarian Classification	Accuracy	Overall proportion of correctly classified humanitarian messages.
	Macro-F1	Measures class-balanced classification performance by giving equal importance to all nine categories.
	Weighted-F1	Measures overall F1 while accounting for the number of samples in each category.
Routing Priority	Priority Accuracy	Proportion of messages assigned to the correct P1–P4 priority level.
	Priority Macro-F1	Measures balanced performance across the four routing priority levels.
	P1 Recall	Proportion of ground-truth P1 messages correctly predicted as P1.

TABLE VIII: Summary of Evaluation Metrics Used in the Study

Metric	Logistic Regression	BERT
Humanitarian category accuracy	0.6772	0.7188
Humanitarian category macro F1	0.5549	0.5838
Humanitarian category weighted F1	0.6748	0.7096
Priority accuracy	0.7804	0.8070
Priority macro F1	0.7256	0.7560
P1 recall	0.8302	0.8868

TABLE IX: Overall NLP Classification and Routing Priority Performance on the Fixed Test Set

Fig. 5: Priority confusion matrices for Logistic Regression and BERT.

Condition	P1 Delivery Probability	P1 Delivered Total	Average Latency (s)	Median Latency (s)
Standard	0.9586	810	1835.68	1392.6
Oracle	0.9586	810	1826.23	1378.0
Logistic Regression	0.9527	805	1839.80	1377.9
BERT	0.9574	809	1842.80	1379.6

TABLE X: P1 delivery and latency performance across the five mobility seeds.

Fig. 6: P1 delivery probability across the four routing conditions.

Fig. 7: P1 delivery latency across four routing conditions and five mobility seeds.

Condition	P1	P2	P3	P4	Total
Oracle	0	0	0	73	73
Logistic Regression	5	0	16	53	74
BERT	1	1	11	52	65

TABLE XI: Buffer losses by ground-truth priority across priority-aware routing conditions.

Fig. 8: Buffer losses by ground-truth priority across priority-aware routing conditions.

Metric	Logistic Regression	BERT
Priority Accuracy	0.7804	0.8070
P1 Recall	0.8302	0.8868
Under-prioritisation Rate	0.1048	0.0899
Severe Priority Error Rate	0.0632	0.0349
P1 Delivery Probability	0.9527	0.9574
Total P1 Delivered	805	809

TABLE XII: End-to-end comparison of Logistic Regression and BERT.

Fig. 9: Parallel coordinates comparison of Logistic Regression and BERT.

Disaster Response

Disaster response uses data, models, and planning tools to assess hazards and coordinate evacuation, relief, and recovery.

See all 60 papers

Lightweight Machine Learning for Classifying Disaster-Related Social Media Posts
Oct 17, 2026Chengjun Jiang, Milena RadenkovicText Classification Social Media Analysis
School of Computer Science, University of Nottingham
Coping in Crisis: Computational Modeling of Coping Styles in Digital Crisis Discourse During the 2023 Turkiye Earthquake
Jun 12, 2026Şevval ÇakıcıSocial Media Analysis Computational Social Science
Koç University
From Seasonality to Semantics: Benchmarking a Hybrid Probabilistic Forecasting System for Roadblocks in Bolivia
Jul 23, 2026Rodrigo Vargas Sainz, Christian Berón CurtiProbabilistic Time Series Forecasting
Universidad Privada de Santa Cruz de la Sierra

Oct 17, 2026J/K move · Enter open · S save

Lightweight Machine Learning for Classifying Disaster-Related Social Media Posts

Chengjun Jiang, Milena Radenkovic

School of Computer Science, University of Nottingham

Text Classification Social Media Analysis

Open paper AskPDF arXiv

Classifying disaster-related social-media posts into humanitarian information labels is challenging because individual messages are typically short and context-limited, humanitarian categories can overlap semantically, and the CrisisBench label distribution is strongly imbalanced. This study investigates whether lightweight lexical machine-learning methods can provide effective and computationally efficient classification under these conditions. Using the CrisisBench v1.0 English humanitarian task, comprising 87,455 records across predefined Train, Development and Test splits, we conduct a controlled comparison of TF-IDF representations and lightweight classifiers, with Macro-F1 as the primary model-selection metric. Beyond aggregate predictive performance, the study jointly examines per-label behaviour, directional error patterns, relative decision margins, computational requirements, and the effect of domain-specific training through a tropical-cyclone specialist model. The selected balanced LinearSVC with unigram-bigram TF-IDF and C=0.5 achieved an Accuracy of 0.8282, a Macro-F1 of 0.6578 and a Weighted-F1 of 0.8315 on the held-out Test split. Lower decision margins were substantially associated with prediction errors, while the tropical-cyclone specialist achieved modestly higher in-domain aggregate performance with substantially lower feature and fitting requirements. The study contributes a controlled, label-sensitive and resource-aware evaluation of lightweight humanitarian text classification that characterises not only aggregate performance, but also where these models fail and how their behaviour changes under domain-specific training.

Abstract

Disaster Response

Explore similar work

Lightweight Machine Learning for Classifying Disaster-Related Social Media Posts

Coping in Crisis: Computational Modeling of Coping Styles in Digital Crisis Discourse During the 2023 Turkiye Earthquake

From Seasonality to Semantics: Benchmarking a Hybrid Probabilistic Forecasting System for Roadblocks in Bolivia