Research map / Large language models / Language, safety and interpretability
Explainable and Discourse Topic-aware Neural Language Understanding
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 explainable topic representation (ETR) composed with LSTM-LM output, versus latent topic proportion only (ablation) | LSTM-LM · APNEWS, IMDB, BNC (small-NLM) | perplexity: APNEWS 55.48→49.34; IMDB 68.21→59.20; BNC 98.31→95.62 | Helped |
| add sentence-level topical discourse (SDT) to latent+explainable topic composition (ablation) | LETA-NLM · APNEWS, IMDB, BNC (small-NLM) | perplexity: APNEWS 48.33→42.98; IMDB 58.10→56.65; BNC 94.78→88.30 | Helped |
| add sentence-level topical discourse (SDT) to ETA-NLM (ablation) | ETA-NLM · IMDB and APNEWS (small-NLM) | perplexity: APNEWS 49.34→48.75 (better); IMDB 59.20→57.83 (better); BNC 95.62→95.64 (slightly worse) | Mixed |
| add explainable topic representation (ETR) to CNN text classifier features | CNN-ETA · 20NS and R21578 accuracy | accuracy: 20NS .745→.775; R21578 .750→.763; IMDB .899→.903 | Helped |
| add latent+explainable topic features (LETA) to CNN classifier | CNN-LETA · IMDB accuracy | accuracy: .899→.908 | Helped |
| add explainable topic representation (ETR) to CNN classifier features on 20NS | CNN-ETA · 20NS accuracy | accuracy: +.030 vs CNN-LSTM | Helped |
| add ETR (explainable topics) to LTA-based composite for retrieval | ETA-NLM · 20NS retrieval P@5 | P@5: .264→.287 | Helped |
| add topic composition (latent or explainable) to BiLSTM-LM for WSD | BiLSTM-LM · WSD all-words ALL test F1 | micro F1: 59.8→60.7 | Helped |
| add explainable topic composition vs latent topic composition in WSD | BiLSTM-LM with topics · WSD ALL test F1 | micro F1: LTA-NLM 60.7, ETA-NLM 60.7, LETA-NLM 60.7 | No clear effect |
| add topic composition with LSTM-LM for topic-aware language modeling (LTA baseline) and then explainable topics | LSTM-LM · topic coherence (NPMI) | average coherence | No clear effect |
Closest papers by meaning
- textTOvec: DEEP CONTEXTUALIZED NEURAL AUTOREGRESSIVE TOPIC MODELS OF LANGUAGE WITH DISTRIBUTED COMPOSITIONAL PRIOR (ICLR 2019 · similarity 0.88)
- Improving Topic Modeling by Distilling Soft Labels from Language Models (ICML 2026 · similarity 0.88)
- Knowledge-Aware Bayesian Deep Topic Model (NeurIPS 2022 · similarity 0.87)
- Meta-Complementing the Semantics of Short Texts in Neural Topic Models (NeurIPS 2022 · similarity 0.87)
- Neural Dynamic Focused Topic Model (AAAI 2023 · similarity 0.86)
- In-Context Pretraining: Language Modeling Beyond Document Boundaries (ICLR 2024 · similarity 0.85)
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.