A transformer-based multi-feature fusion method for detecting traffic events using Twitter data
Bibliographic record
Abstract
Early traffic event detection is essential for transportation networks’ quick response and accurate performance recovery. Social media data (e.g. Twitter data) can be a valuable source for detecting and describing traffic events, such as accidents, congestion, and road closures. Integrating spatial and temporal features inherent in Twitter can enhance models by revealing non-trivial information, patterns, and knowledge, resulting in improved performance. Recent research in Deep Learning (DL) has revealed the strength of learning features directly from data to extract potential hidden features that efficiently infer human activities and interactions and detect the relationships to generate fine-grained information. This research explores the efficiency of integrating Twitter data’s spatial, temporal, and semantic features for traffic event detection using DL. A novel framework employing a transformer-based multi-feature fusion approach is proposed, designed to detect traffic incidents comprehensively via Twitter data. The framework classifies tweets based on multiple dimensions: (1) semantic content is numerically represented and categorized traffic incidents, non-traffic, or traffic conditions and information; (2) spatial characteristics are analyzed through hot-spot analysis techniques, classifying locations into hot or cold spots; and (3) temporal attributes (date and time) are visualized and analyzed through heat maps reflecting incident densities. The performance of the models was then evaluated based on various fusion scenarios combining spatial, temporal, and semantic data using performance metrics such as F-score and accuracy. The results showed that the scenario of transformer-based multi-feature fusion of spatial, temporal, and semantic data for traffic event detection yielded better results. The model achieved a 7.93% accuracy improvement when distinguishing between the two classes, “Traffic Incident” and “Non-Traffic”, and a 6.09% increase in accuracy when classifying across three categories: “Traffic Incident,” “Non-Traffic,” and “Traffic Conditions and information.” These findings highlight the effectiveness of using a multifeatured Twitter dataset for improved detection accuracy.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".