Research map / Large language models / Language, safety and interpretability
Back to the Future – Temporal Adaptation of Text Representations
Johannes Bjerva, Wouter Kouw, Isabelle Augenstein
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add semi-supervised subspace alignment (10 labelled target points per class) to previous-year training | BERT + SVM · PeerRead acceptance | accuracy (avg): 53.75 -> 68.09 | Improved |
| semi-supervised SSA from past data vs training on all data | BERT + SVM · PeerRead acceptance | accuracy (avg): 58.82 -> 68.09 | Improved |
| use unsupervised instead of semi-supervised alignment (ablation) | SSA · PeerRead acceptance | accuracy (avg): 68.09 -> 52.05 | Hurt |
| add unbounded multi-step extension to semi-supervised SSA (ablation) | SSA · PeerRead acceptance | accuracy (avg): 68.09 -> 69.44 | Improved |
| add k-means cluster IDs (instance similarity) to unbounded SSA (ablation) | SSA unbounded · PeerRead acceptance | accuracy (avg): 69.44 -> 69.56 | Improved |
| semi-supervised SSA vs previous-year training | BERT-cased + MLP · Broad Twitter Corpus NER | F1 (avg): 56.85 -> 68.75 | Improved |
| unsupervised alignment on NER (ablation) | BERT-cased + MLP · Broad Twitter Corpus NER | F1 (avg): 56.85 -> 46.34 | Hurt |
| add cluster IDs to unbounded SSA (ablation) | SSA · Broad Twitter Corpus NER | F1 (avg): 68.75 -> 71.35 | Improved |
| training on same-year in-domain data vs semi-supervised SSA | BERT-cased + MLP · Broad Twitter Corpus NER | F1 (avg): 60.00 vs 68.75 | Mixed |
| semi-supervised SSA vs previous rumour training | BERT + SVM · RumourEval-2019 SDQC | F1 (avg): 37.00 -> 38.65 | Improved |
| unsupervised alignment vs previous rumour training (ablation) | BERT + SVM · RumourEval-2019 SDQC | F1 (avg): 37.00 -> 37.00 | No clear effect |
| add unbounded extension (ablation) | semi-supervised SSA · RumourEval-2019 SDQC | F1 (avg): 38.65 -> 37.90 | No clear effect |
Closest papers by meaning
- Arrows of Time for Large Language Models (ICML 2024 · similarity 0.86)
- Future Language Modeling from Temporal Document History (ICLR 2024 · similarity 0.86)
- Language Model Behavioral Phases are Consistent Across Architecture, Training Data, and Scale (NeurIPS 2025 · similarity 0.85)
- Don’t Let It Fade: Preserving Edits in Diffusion Language Models via Token Timestep Allocation (NeurIPS 2025 · similarity 0.85)
- From Bytes to Ideas: Language Modeling with Autoregressive U-Nets (NeurIPS 2025 · similarity 0.84)
- Incremental Sequence Classification with Temporal Consistency (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
