Organize the Web: Constructing Domains Enhances Pre-Training Data Curation

• URL: http://arxiv.org/abs/2502.10341v1
• Date: Fri, 14 Feb 2025 18:02:37 GMT
• Title: Organize the Web: Constructing Domains Enhances Pre-Training Data Curation
• Authors: Alexander Wettig, Kyle Lo, Sewon Min, Hannaneh Hajishirzi, Danqi Chen, Luca Soldaini,
• Abstract summary: We develop a framework for organizing web pages in terms of both their topic and format. We automatically annotate pre-training data by distilling annotations from a large language model into efficient curations. Our work demonstrates that constructing and mixing domains provides a valuable complement to quality-based data curation methods.
• Score: 129.27104172458363
• License:
• Abstract: Modern language models are trained on large, unstructured datasets consisting of trillions of tokens and obtained by crawling the web. The unstructured nature makes it difficult to reason about their contents and develop systematic approaches to data curation. In this paper, we unpack monolithic web corpora by developing taxonomies of their contents and organizing them into domains. We introduce WebOrganizer, a framework for organizing web pages in terms of both their topic and format. Using these two complementary notions of domains, we automatically annotate pre-training data by distilling annotations from a large language model into efficient classifiers. This allows us to study how data from different domains should be mixed to improve models on downstream tasks, and we show that we can combine insights about effective topics and formats to further boost performance. We demonstrate that our domain mixing also improves existing methods that select data based on quality. Furthermore, we study and compare how quality-based methods will implicitly change the domain mixture. Overall, our work demonstrates that constructing and mixing domains provides a valuable complement to quality-based data curation methods, opening new avenues for effective and insightful pre-training data curation.

Related papers

• Web-Scale Visual Entity Recognition: An LLM-Driven Data Approach [56.55633052479446]
  Web-scale visual entity recognition presents significant challenges due to the lack of clean, large-scale training data. We propose a novel methodology to curate such a dataset, leveraging a multimodal large language model (LLM) for label verification, metadata generation, and rationale explanation. Experiments demonstrate that models trained on this automatically curated data achieve state-of-the-art performance on web-scale visual entity recognition tasks.
  arXiv  Detail & Related papers  (2024-10-31T06:55:24Z)
• Dual Consolidation for Pre-Trained Model-Based Domain-Incremental Learning [64.1745161657794]
  Domain-Incremental Learning (DIL) involves the progressive adaptation of a model to new concepts across different domains. Recent advances in pre-trained models provide a solid foundation for DIL. However, learning new concepts often results in the catastrophic forgetting of pre-trained knowledge. We propose DUal ConsolidaTion (Duct) to unify and consolidate historical knowledge.
  arXiv  Detail & Related papers  (2024-10-01T17:58:06Z)
• Structure-aware Domain Knowledge Injection for Large Language Models [38.08691252042949]
  StructTuning is a methodology to transform Large Language Models (LLMs) into domain specialists. It significantly reduces the training corpus needs to a mere 5% while achieving an impressive 100% of traditional knowledge injection performance.
  arXiv  Detail & Related papers  (2024-07-23T12:38:48Z)
• Adapting to Distribution Shift by Visual Domain Prompt Generation [34.19066857066073]
  We adapt a model at test-time using a few unlabeled data to address distribution shifts. We build a knowledge bank to learn the transferable knowledge from source domains. The proposed method outperforms previous work on 5 large-scale benchmarks including WILDS and DomainNet.
  arXiv  Detail & Related papers  (2024-05-05T02:44:04Z)
• RanLayNet: A Dataset for Document Layout Detection used for Domain Adaptation and Generalization [36.973388673687815]
  RanLayNet is a synthetic document dataset enriched with automatically assigned labels. We show that a deep layout identification model trained on our dataset exhibits enhanced performance compared to a model trained solely on actual documents.
  arXiv  Detail & Related papers  (2024-04-15T07:50:15Z)
• Adapting Large Language Models to Domains via Reading Comprehension [86.24451681746676]
  We explore how continued pre-training on domain-specific corpora influences large language models. We show that training on the raw corpora endows the model with domain knowledge, but drastically hurts its ability for question answering. We propose a simple method for transforming raw corpora into reading comprehension texts.
  arXiv  Detail & Related papers  (2023-09-18T07:17:52Z)
• Learning to Cluster under Domain Shift [20.00056591000625]
  In this work we address the problem of transferring knowledge from a source to a target domain when both source and target data have no annotations. Inspired by recent works on deep clustering, our approach leverages information from data gathered from multiple source domains. We show that our method is able to automatically discover relevant semantic information even in presence of few target samples.
  arXiv  Detail & Related papers  (2020-08-11T12:03:01Z)
• Domain Adaptation for Semantic Parsing [68.81787666086554]
  We propose a novel semantic for domain adaptation, where we have much fewer annotated data in the target domain compared to the source domain. Our semantic benefits from a two-stage coarse-to-fine framework, thus can provide different and accurate treatments for the two stages. Experiments on a benchmark dataset show that our method consistently outperforms several popular domain adaptation strategies.
  arXiv  Detail & Related papers  (2020-06-23T14:47:41Z)
• Dynamic Fusion Network for Multi-Domain End-to-end Task-Oriented Dialog [70.79442700890843]
  We propose a novel Dynamic Fusion Network (DF-Net) which automatically exploit the relevance between the target domain and each domain. With little training data, we show its transferability by outperforming prior best model by 13.9% on average.
  arXiv  Detail & Related papers  (2020-04-23T08:17:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.