Tag
Columbia University researchers introduce the Hierarchical Embedded Topic Model (hETM), which couples document- and section-level topic distributions via a hierarchical prior in embedding space, with amortized variational inference showing improved perplexity, topic quality, and clustering across legal, financial, and forum corpora.
This paper presents two randomized SVD approximations for spectral co-clustering of word-document matrices, showing that the random-projection method is more reliable across settings, while the sampling-based method is better for denser matrices.
This paper introduces an open modular transformer-based framework for detecting and structuring social tipping point evidence in climate documents, utilizing models like DistilBERT, RoBERTa, Mistral 7B, and LLaMA 3.2 3B, with evaluation on expert-labeled benchmarks.
This paper tests the assumption that sentiment tools validated against human labels also predict market signals in financial NLP, finding that benchmark agreement does not determine predictive effectiveness and spam influences volume-based analysis.
This paper constructs a multimodal dataset of 1000 academic papers with text, images, and audio to study keyword extraction, showing that fusing multiple modalities improves performance.
This paper introduces a text-based causal inference methodology using an enhanced CausalBERT to disentangle the effects of individual aspects (e.g., school administration, academic performance) on overall online review ratings, validated on 600K+ U.S. K-12 school reviews. Key improvements include temperature scaling, hyperparameter optimization, and interpretability methods to reduce confounding bias.
The paper presents Prometheus, a framework that uses large language models to extract local causal claims from text and organizes them into navigable causal atlases, enabling deep causal research across diverse domains.