AICC: Parse HTML Finer, Make Models Better -- A 7.3T AI-Ready Corpus Built by a Model-Based HTML Parser

2025-11-20

AI Summary

Existing HTML parsing methods (e.g., Trafilatura) rely on heuristic rules, resulting in poor fidelity in reconstructing structured content such as mathematical formulas, code blocks, and tables—thereby degrading the quality of large language model (LLM) training data. To address this, we formulate HTML extraction as a semantic-aware sequence labeling task and propose a two-stage formatting pipeline. We develop MinerU-HTML, a lightweight (0.6B-parameter) parser that jointly performs semantic classification and Markdown conversion. Evaluated on MainWebBench, it demonstrates superior structural preservation. Leveraging this parser, we construct AICC—a multilingual, AI-ready corpus of 7.3 trillion tokens—significantly improving pretraining efficacy: under identical filtering criteria, models trained on AICC achieve a mean accuracy of 50.8% across 13 benchmarks, outperforming TfCC by +1.08 percentage points and surpassing both RefinedWeb and FineWeb. This confirms that high-fidelity HTML parsing is a critical enabler of LLM performance.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Knowledge Acquisition from the Web
• Natural Language Processing: (Large) Language Models

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

While web data quality is crucial for large language models, most curation efforts focus on filtering and deduplication,treating HTML-to-text extraction as a fixed pre-processing step. Existing web corpora rely on heuristic-based extractors like Trafilatura, which struggle to preserve document structure and frequently corrupt structured elements such as formulas, codes, and tables. We hypothesize that improving extraction quality can be as impactful as aggressive filtering strategies for downstream performance. We introduce MinerU-HTML, a novel extraction pipeline that reformulates content extraction as a sequence labeling problem solved by a 0.6B-parameter language model. Unlike text-density heuristics, MinerU-HTML leverages semantic understanding and employs a two-stage formatting pipeline that explicitly categorizes semantic elements before converting to Markdown. Crucially, its model-based approach is inherently scalable, whereas heuristic methods offer limited improvement pathways. On MainWebBench, our benchmark of 7,887 annotated web pages, MinerU-HTML achieves 81.8% ROUGE-N F1 compared to Trafilatura's 63.6%, with exceptional structured element preservation (90.9% for code blocks, 94.0% for formulas). Using MinerU-HTML, we construct AICC (AI-ready Common Crawl), a 7.3-trillion token multilingual corpus from two Common Crawl snapshots. In controlled pretraining experiments where AICC and Trafilatura-extracted TfCC undergo identical filtering, models trained on AICC (62B tokens) achieve 50.8% average accuracy across 13 benchmarks, outperforming TfCC by 1.08pp-providing direct evidence that extraction quality significantly impacts model capabilities. AICC also surpasses RefinedWeb and FineWeb on key benchmarks. We publicly release MainWebBench, MinerU-HTML, and AICC, demonstrating that HTML extraction is a critical, often underestimated component of web corpus construction.

Problem

• Improving HTML-to-text extraction quality for web data
• Preserving structured elements like formulas and codes
• Enhancing downstream model performance through better parsing

Innovation

• Uses a 0.6B language model for sequence labeling
• Employs two-stage formatting pipeline for semantic elements
• Achieves superior structured element preservation in extraction