Long-range tasks require reasoning over long inputs. Existing solutions either need large compute budgets, training data, access to model weights, or use complex, task-specific approaches. We present PRISM, which alleviates these concerns by processing information as a stream of chunks, maintaining a structured in-context memory specified by a typed hierarchy schema. This approach demonstrates superior performance to baselines on diverse tasks while using at least 4x smaller contexts than long-context models. Moreover, PRISM is token-efficient. By producing short outputs and efficiently leveraging key-value (KV) caches, it achieves up to 54% cost reduction when compared to alternative short-context approaches. The method also scales down to tiny information chunks (e.g., 500 tokens) without increasing the number of tokens encoded or sacrificing quality. Furthermore, we show that it is possible to generate schemas to generalize our approach to new tasks with minimal effort.

No existing dataset adequately tests how well language models can incrementally update entity summaries - a crucial ability as these models rapidly advance. The Incremental Entity Summarization (IES) task is vital for maintaining accurate, up-to-date knowledge. To address this, we introduce SUMIE, a fully synthetic dataset designed to expose real-world IES challenges. This dataset effectively highlights problems like incorrect entity association and incomplete information presentation. Unlike common synthetic datasets, ours captures the complexity and nuances found in real-world data. We generate informative and diverse attributes, summaries, and unstructured paragraphs in sequence, ensuring high quality. The alignment between generated summaries and paragraphs exceeds 96%, confirming the dataset's quality. Extensive experiments demonstrate the dataset's difficulty - state-of-the-art LLMs struggle to update summaries with an F1 higher than 80.4%. We will open source the benchmark and the evaluation metrics to help the community make progress on IES tasks.

Summarizing comparative opinions about entities (e.g., hotels, phones) from a set of source reviews, often referred to as contrastive summarization, can considerably aid users in decision making. However, reliably measuring the contrastiveness of the output summaries without relying on human evaluations remains an open problem. Prior work has proposed token-overlap based metrics, Distinctiveness Score, to measure contrast which does not take into account the sensitivity to meaning-preserving lexical variations. In this work, we propose an automated evaluation metric CASPR to better measure contrast between a pair of summaries. Our metric is based on a simple and light-weight method that leverages natural language inference (NLI) task to measure contrast by segmenting reviews into single-claim sentences and carefully aggregating NLI scores between them to come up with a summary-level score. We compare CASPR with Distinctiveness Score and a simple yet powerful baseline based on BERTScore. Our results on a prior dataset CoCoTRIP demonstrate that CASPR can more reliably capture the contrastiveness of the summary pairs compared to the baselines.

Users often struggle with decision-making between two options (A vs B), as it usually requires time-consuming research across multiple web pages. We propose STRUM-LLM that addresses this challenge by generating attributed, structured, and helpful contrastive summaries that highlight key differences between the two options. STRUM-LLM identifies helpful contrast: the specific attributes along which the two options differ significantly and which are most likely to influence the user's decision. Our technique is domain-agnostic, and does not require any human-labeled data or fixed attribute list as supervision. STRUM-LLM attributes all extractions back to the input sources along with textual evidence, and it does not have a limit on the length of input sources that it can process. STRUM-LLM Distilled has 100x more throughput than the models with comparable performance while being 10x smaller. In this paper, we provide extensive evaluations for our method and lay out future directions for our currently deployed system.

Understanding visually-rich business documents to extract structured data and automate business workflows has been receiving attention both in academia and industry. Although recent multi-modal language models have achieved impressive results, we find that existing benchmarks do not reflect the complexity of real documents seen in industry. In this work, we identify the desiderata for a more comprehensive benchmark and propose one we call Visually Rich Document Understanding (VRDU). VRDU contains two datasets that represent several challenges: rich schema including diverse data types as well as hierarchical entities, complex templates including tables and multi-column layouts, and diversity of different layouts (templates) within a single document type. We design few-shot and conventional experiment settings along with a carefully designed matching algorithm to evaluate extraction results. We report the performance of strong baselines and offer three observations: (1) generalizing to new document templates is still very challenging, (2) few-shot performance has a lot of headroom, and (3) models struggle with hierarchical fields such as line-items in an invoice. We plan to open source the benchmark and the evaluation toolkit. We hope this helps the community make progress on these challenging tasks in extracting structured data from visually rich documents.

Many business workflows require extracting important fields from form-like documents (e.g. bank statements, bills of lading, purchase orders, etc.). Recent techniques for automating this task work well only when trained with large datasets. In this work we propose a novel data augmentation technique to improve performance when training data is scarce, e.g. 10-250 documents. Our technique, which we call FieldSwap, works by swapping out the key phrases of a source field with the key phrases of a target field to generate new synthetic examples of the target field for use in training. We demonstrate that this approach can yield 1-7 F1 point improvements in extraction performance.

Consider a sequential active learning problem where, at each round, an agent selects a batch of unlabeled data points, queries their labels and updates a binary classifier. While there exists a rich body of work on active learning in this general form, in this paper, we focus on problems with two distinguishing characteristics: severe class imbalance (skew) and small amounts of initial training data. Both of these problems occur with surprising frequency in many web applications. For instance, detecting offensive or sensitive content in online communities (pornography, violence, and hate-speech) is receiving enormous attention from industry as well as research communities. Such problems have both the characteristics we describe -- a vast majority of content is not offensive, so the number of positive examples for such content is orders of magnitude smaller than the negative examples. Furthermore, there is usually only a small amount of initial training data available when building machine-learned models to solve such problems. To address both these issues, we propose a hybrid active learning algorithm (HAL) that balances exploiting the knowledge available through the currently labeled training examples with exploring the large amount of unlabeled data available. Through simulation results, we show that HAL makes significantly better choices for what points to label when compared to strong baselines like margin-sampling. Classifiers trained on the examples selected for labeling by HAL easily out-perform the baselines on target metrics (like area under the precision-recall curve) given the same budget for labeling examples. We believe HAL offers a simple, intuitive, and computationally tractable way to structure active learning for a wide range of machine learning applications.