CS Paper Radar

Interpretable Feature Discovery from Unstructured Data Using LLMs and Statistical Choice Models

Baptiste Arnaudo, Amine Bennouna, Sébastien Martin

cs.AI stat.MLcs.LG ·submitted 2026-10-27 23:54 UTC · v1
code/data

Abstract

Feature extraction from unstructured data typically relies on opaque embeddings unsuited to interpretable decision-making. Interpretable features, on the other hand, typically require costly, and often infeasible, hand-engineering. Large language models (LLMs) offer a natural bridge, as they can read raw data and describe it in interpretable terms. One could simply ask an LLM to extract features from a large sample of the dataset, but context limitations and diluted attention can lead to unsatisfactory results. We introduce a method that automatically discovers interpretable, decision-relevant features from unstructured data. Our key idea is to perform pairwise comparisons between items, asking an LLM to articulate, in interpretable terms, the features distinguishing each pair. We treat these explanations, or rationales, as data: we develop a statistical choice model that combines these rationales across the comparisons to identify relevant features and estimate per-item, per-feature scores. We first test our method on a synthetic hiring dataset of cover letters with known ground-truth features and show that it significantly outperforms direct LLM feature extraction in feature recovery. We then apply it to the ChangeMyView (Reddit) debate dataset, where it discovers several interpretable features associated with persuasion while reaching predictive accuracy comparable to established baselines. We also show how the same rationales can be used to audit LLM decisions, yielding a holistic score whose decomposition over interpretable features reveals what the LLM rewards.

Comment
Preprint. 88 pages, 21 figures, 12 tables. Code: https://github.com/ArnaudoB/Interpretable-Feature-Discovery-LLM

Closest papers

Nearest neighbours by meaning across the whole index.

LLMs as Feature Engineers for Text-and-Tabular Prediction

cs.LG ·2026-09-18 ·Merwan Barlier, Blaz Skrlj
similarity 0.72
Abstract

We introduce an iterative framework that automates the extraction of interpretable, schema-bound categorical features from unstructured text for tabular prediction models. To navigate the feature space, a generator LLM proposes semantic definitions, a separate extractor LLM materializes the features, and a downstream tabular model evaluates their predictive performance. We optimize this search by translating explicit model errors, such as AUC ranking inversions, into natural-language feedback, steering the LLM to resolve specific predictive failures. Evaluated across three public datasets, this error-driven loop accelerates feature discovery by up to $3\times$ compared to unguided search. Empirically, the generated features demonstrate strong multi-view complementarity, strictly outperforming any subset when combined with TF-IDF and dense embeddings. Finally, the framework guarantees instance-level interpretability: the discovered features dominate SHAP importance rankings and provide a fully transparent, semantic audit trail for every prediction.

0.39

Converting Expert Deliberation into Financial Signals Through A Context-Aware NLP Pipeline

cs.LG cs.CE ·2026-08-19 ·Vivek Batra, Kristin Chen, Sanjiv Das, Samuel Judge, Harshad Khadilkar +4 more
similarity 0.65
Abstract

We introduce the CDSP (context-conditional deliberation signal pipeline), converting an investment committee's meeting transcripts into structured predictive features. CDSP segments the meeting transcripts into topical chunks, assigns asset-class context labels using a large language model (LLM), maps financial keywords to a pre-determined taxonomy of labels, and constructs complementary features: sentiment polarity and mention frequency. This feature engineering framework is applied to a dataset spanning 48 monthly committee meetings to predict if global equities will perform better or worse than global bonds in the following month. In experiments with engineered features, raw transcript text, sentence embeddings, and combined representations, the prediction accuracy ranges from 62% to 73%, compared to always choosing stocks, which outperforms bonds 60.4% of the time. The best (73% accurate) model combines sentence embeddings with engineered CDSP features, achieving a 0.73 F1 score (although this is not statistically significant compared to always choosing stocks). Sentiment carries a stronger signal than mention frequency for several taxonomy categories. These findings suggest that experts' deliberations may contain forward-looking information that context-aware NLP can extract.

0.20

SleuthBench: Benchmarking Statistical LLM Evaluation Using Tabular Hidden Signals

cs.LG ·2026-09-28 ·Jingyun Jia, Antoine Remond-Tiedrez, Aaron Alvarez, Joshua Shunk, Rich Caruana +1 more
benchmark similarity 0.65
Abstract

Evaluating statistical discovery by large language model (LLM) agents requires verifiable analytical ground truth. Establishing such ground truth for real-world datasets is costly, and prior knowledge of public datasets can influence agent responses. We introduce SLEUTHBENCH, a benchmark that addresses both problems by injecting controlled data-quality problems and feature effects into public tabular datasets: the injected pattern determines the answer, so reference answers are computed automatically and memorized knowledge of the original table is insufficient, while the table keeps its background structure. The injected patterns are modeled on phenomena reported in real data analyses. The benchmark defines 17 question templates in two families: data-quality questions and feature-contribution questions. We evaluate six state-of-the-art LLMs that analyze the data using a Python coding tool, on data-science and business phrasings of 70 validated dataset-template combinations, yielding 1680 graded responses in total. The models detect data-quality problems reliably (83.8% accuracy) but recover feature contributions poorly (41.9%). Finding how features shape the target requires searching over both candidate variables and analytical procedures. To address this issue, we propose the Empirical Layer, a set of precomputed statistical artifacts comprising summaries, fitted feature and interaction effects, and dataset descriptions, which exposes candidate patterns for direct inspection. Access to these artifacts raises feature-contribution accuracy from 41.9% to 68.0%.

0.19

Beyond Token Scale: Chunk-Level Sparse Autoencoders for Reliable Semantic Feature Discovery

cs.CL cs.AIcs.LG ·2026-09-28 ·Xu Wang, Yifan Yang, TingHao YU, Difan Zou
similarity 0.65
Abstract · 27 pages

Sparse autoencoders (SAEs) expose features that help us understand and steer language models, but faithful reconstruction does not guarantee informative concepts. Token-level objectives reward lexical and formatting details alongside semantic content, all competing for a limited sparse budget. We introduce a family of chunk-level SAEs that encode mean-pooled activations over chunks, each a contiguous span of tokens: Mean-Chunk reconstructs the observed chunk, Cross-Chunk predicts an independently processed neighbor, and Joint-Chunk combines both targets. These designs separate the effect of a larger observation unit from that of predicting information shared across passages. With matched training data, chunk-level SAEs remain powerful interpretability tools while learning reliable semantic features that capture high-level concepts and respond selectively to relevant content. Their strengths are complementary: Mean-Chunk improves high-level feature discovery, reasoning detection beyond surface cues, and steering; Cross-Chunk leads document retrieval and classification transfer while producing selective, persistent features. Changing what an SAE sees and predicts yields reliable semantic features for more meaningful tasks. We demonstrate their practical value through gains across downstream tasks such as retrieval, reasoning detection, and steering.

0.23

From the Zoo to the BULLPEN: Sample-Efficient LLM Profiling via Multi-Channel Model Embeddings

cs.AI cs.LG ·2026-10-30 ·Russell Ro, Qiming Cui, Reza Abbasi-Asl
benchmarkcode/data similarity 0.64
Abstract · 28 pages, 7 figures, 12 tables. Code and data: https://github.com/wcquesera/bullpen

Choosing among the hundreds of thousands of available large language models (LLMs) requires a thorough understanding of each candidate model's properties. In practice, each property is measured by its own benchmark suite, so profiling a model on many properties means paying for many separate evaluations. To address this, model embeddings compress each model into a fixed-length vector from which many properties can be predicted. Existing methods, such as EmbedLLM, build these vectors from correctness patterns and question features, and often need large probe batteries to do so. However, many behaviors that are important when choosing a model, such as its verbosity, safety and style, show up more in how it answers than in whether it is correct. We introduce BULLPEN (Behavioral Understanding of Large Language models via Profile ENcodings), a framework that extends EmbedLLM in three ways. First, BULLPEN organizes the downstream tasks that prior work uses and adds 20 task benchmarks we built on behavior, style and provenance. Second, it is the first framework to read a model's own answer text jointly with its correctness and the question text. Third, it encodes models from a few hundred questions instead of the full 24,118-question pool. Across 260 open-weight models, our answer-text encoder scores 0.48 +/- 0.10 on the new tasks against 0.35 +/- 0.10 for the best prior encoder. Our encoders that read answer text keep at least 91% of their full-pool skill from only 400 questions (1.3% of the inference compute of the full question pool), while a correctness-only baseline keeps 83% with the same 400 questions. Because these embeddings need only model outputs, they support behavioral auditing and low-cost profiling, and could extend to API-only models. Code and data: https://github.com/wcquesera/bullpen

0.29

Long Text to Predictive Features: LLM-Guided Blockwise Feature Engineering via Executable Program Search

cs.LG cs.CL ·2026-10-08 ·Ziming Dai, Dabiao Ma, Ziheng Guo, Jack Dong, Zimu Zhou
similarity 0.63
Abstract

Industrial risk-control systems typically rely on structured-data models for efficient prediction, yet substantial valuable information remains embedded in unstructured long text. Extracting this information through manual feature engineering is labor-intensive, while requiring a large language model (LLM) to process every real-time input may not meet practical deployment requirements. To address this challenge, we propose LLM-BlockFE, an LLM-guided offline feature construction framework that converts long text into executable feature programs, thereby avoiding LLM calls during online inference. LLM-BlockFE constructs feature programs by incrementally appending immutable code blocks and evaluates candidate features using a downstream model. To address the tendency of conventional greedy search to become trapped in suboptimal solutions, our method introduces a block-level rollback mechanism based on depth-calibrated credit allocation and advances multiple independent search trajectories in an interleaved manner, reducing redundant exploration by sharing fixed descriptions of each trajectory's exploration direction. After the search, the resulting programs are frozen and deployed to extract structured features for downstream prediction models. Across two public and two private datasets, LLM-BlockFE achieves absolute AUC improvements of 0.0069 to 0.0358 over the strongest baseline on each dataset in the full-dataset comparison. Post-launch monitoring across five deployed financial risk-control applications shows absolute KS improvements of 0.02 to 1.56 percentage points over the existing manually designed strategy.

0.15

ProToMEx: Rapid, Interpretable Explanations via Structured Representations

cs.LG cs.AI ·2026-09-02 ·Athina Georgara, Adarsh Valoor, Sarvapali D. Ramchurn
similarity 0.63
Abstract

Existing post-hoc explainers for machine learning classifiers primarily focus on feature attribution, assigning importance scores to individual features. While valuable, this approach struggles to articulate the complex, combinatorial patterns that often drive a model's decision-making process. To overcome this limitation, we introduce ProToMEx, a new paradigm for explainability that leverages Probabilistic Topic Models (PTMs). Our model-agnostic framework learns latent ''topics'' that represent distinct, high-level reasons for a classification, moving beyond simple feature importance to reveal underlying semantic structures. ProToMEx naturally provides both global explanations of a model's overall behaviour and local explanations that can disentangle multiple co-existing reasons for a specific prediction. We demonstrate empirically that ProToMEx not only produces explanations of comparable fidelity to popular methods like SHAP and LIME but also drastically reduces the amortised computational cost of generating local explanations, making it highly suitable for real-time applications. Specifically, we show that ProToMEx is ~30-40x faster than SHAP and LIME over standardised tabular datasets and synthetic datasets.

0.21

From Stories to Features: Scaling Concept Vectors for AI Alignment & Interpretability

cs.CL cs.AIcs.LG ·2026-10-17 ·Dmitry Scherbakov, Oleg Chernikov, Victoria Khamash, Kirill Afendulev, Alexey Dontsov +3 more
code/data similarity 0.63
Abstract · 51 pages, 41 figures, 26 tables. Code: https://github.com/josephofthebread/Concept-Vector-Scaling

We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception. Existing unsupervised methods like sparse autoencoders attempt to recover the whole feature space, often at the cost of feature quality and interpretability. We take a more pragmatic route: focusing on concepts relevant to AI safety and alignment, we generate synthetic concept-antagonist story pairs and use the difference of their mean model activations as a feature vector. We scale this approach to 1,036 alignment-relevant concepts using 2M synthetic stories across multiple languages and genres, which we publicly release. Across six models from three model families, our feature vectors reliably generalize to detecting their corresponding concepts in chat dialogues and web text. They form an interpretable geometry shared across models and can causally steer model behavior when injected into activations. We further use these features to analyze model representations during jailbreak attacks, mathematical and agentic reasoning. Our targeted feature construction thus offers a scalable route to studying concepts in language models, potentially extending to millions of features across arbitrary domains.

0.37