arXiv:2602. 17907v2 Announce Type: replace-cross Abstract: Traditional neural topic models are typically optimized by reconstructing the document's Bag-of-Words (BoW) representations, overlooking contextual information and struggling with data sparsity.
arXiv:2602. 17907v3 Announce Type: replace-cross Abstract: Traditional neural topic models are typically optimized by reconstructing the document's Bag-of-Words (BoW) representations, overlooking contextual information and struggling with data sparsity.
The paper introduces Label Semantic Expansion (LSE), a method that enriches sparse label representations by adding descriptive topic words grounded in a corpus. It proposes a Label-Guided Neural Topic Model (LGNTM) that learns label-aligned topics, integrates lexical and document semantics, and maintains consistency between topic and label structures. Experiments show that LSE and LGNTM improve label-topic alignment, label expansion, topic quality, and downstream classification performance.
arXiv:2608. 16269v1 Announce Type: cross Abstract: Recent advances in neural topic models with pre-trained language models (PLMs) have achieved strong performance by leveraging general-domain pre-training, yet their topic interpretability often degrades on specialized corpora.
arXiv:2507. 23220v2 Announce Type: replace-cross Abstract: Traditional topic models are effective at uncovering latent themes in large text collections.
MonoTM is an interpretable topic modeling framework that separates the estimation of document–topic mixtures from the generation of topic descriptors. It uses sparse autoencoders to extract dense, interpretable features for mixture estimation, then learns topic descriptors from a distinct set of corpus‑grounded semantic features. This approach preserves global topic structure while providing more meaningful, semantic‑unit descriptors than traditional top‑word lists.
Dynamic Embedded Topic Models (D-ETM) are extended to enable stable cross‑corpus temporal analysis by first learning a shared dynamic topic space—called the shared backbone—over a merged multi‑corpus collection. Corpus‑specific residual adaptation is then applied around this frozen backbone, allowing each corpus to specialize lexically without creating separate latent topic spaces. Experiments on three corpora spanning 97 years show that this approach yields much stronger alignment of topic trajectories (97.5 ± 0.7 % Retrieval@1) compared to full fine‑tuning or independent training with post‑hoc matching.
The paper introduces SeLATM, a framework that improves topic modeling by generating topics at the segment level and refining them through agentic feedback loops. This approach addresses limitations of LLM-based topic assignment methods, such as the inability to produce topic distributions, overly broad or narrow topics, and high resource consumption. Experiments on multiple datasets show that SeLATM reduces LLM resource usage while maintaining superior performance.
The paper examines how new vocabulary tokens are added to language models for generative recommendation tasks. It shows that the common practice of initializing these tokens as the mean of existing embeddings collapses them into a degenerate subspace, hindering fine‑tuning. The authors propose Grounded Token Initialization (GTI), which places new tokens at semantically meaningful positions in the pretrained embedding space using linguistic supervision, and demonstrate that GTI outperforms mean initialization and other adaptation methods across several benchmarks.
The paper introduces LLM-QL, a dense retrieval model that harnesses large language models (LLMs) by maximizing query likelihood (QL) as an auxiliary task. It incorporates an Attention Block to limit predictive token attention to document tokens before the ending token and a Document Corruption component that masks parts of the document during prediction. Experiments on MS MARCO and BEIR datasets show that LLM-QL outperforms other LLM-based retrievers, and detailed analyses confirm the effectiveness of its components.
arXiv:2510. 16152v2 Announce Type: replace-cross Abstract: Scientific literature is increasingly fragmented by disciplinary boundaries, specialized terminology, and potentially sparse keyword systems, making it difficult to capture the evolving structure of modern science.
The paper investigates why token prediction, a common pre‑training objective for language models, yields useful representations. It introduces a statistical framework linking token prediction accuracy to the geometry of token embeddings, showing that accurate predictions organize embeddings according to Hellinger distances between context distributions. The authors also propose a self‑consistency principle that refines contextual representations through repeated application of a shared block, and provide downstream guarantees for token generation, community recovery, and linear classification.
The paper introduces Retrieval-Augmented Decoding (RAD), a decoding-time method that improves the truthfulness of large language models without retraining. RAD uses a small reference set of up to ten annotated examples to build a grounding space of context embeddings and next-token logits, which it retrieves and aggregates during inference to shape the model’s output. Experiments on four open-ended generation benchmarks and four different LLMs show that RAD consistently outperforms strong baselines and generalizes well across tasks.