PACE: Publisher-Adaptive Content Extraction via Agentic Automation
- Source
- arxiv.org
- Author
- Zhanlin Liu, Munirathnam Srikanth
- Date
Context
PACE uses a to learn how a publisher's pages are organized, then reuses that knowledge without calling the model for every article. The paper describes analyzing representative pages to identify content locations, exclusion rules and fallback logic. Those settings feed a fixed extraction program that processes further pages from the same publisher.
This moves AI work into setting up the extractor. In tests on 71 held-out news pages from 16 publishers, article-text extraction, measured by text overlap with reference text, scored 0.976 against 0.980 for the manually built Fundus parsers and higher than the general-purpose tools tested, such as trafilatura at 0.954. The mechanism is useful where layouts repeat: a costly page-analysis step produces rules that can serve many subsequent pages.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
- multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
articleFrom Pixels to Pairs: A Comprehensive Benchmark of LLM-Based Key-Value Extraction in Noisy Document SettingsZahra Anvari, Vassilis Athitsos
articleAgentic Data Engineering for LLMs: System Design and a Controlled Empirical StudyRuomeng Ding, Qianli Shen, Zhaoyang Han, Meixin Chen, Daoyuan Chen, Yaliang Li
articleHarness Engineering in LLM Tool Use via Agent-Native Reusable Tool PrimitivesHaibo Jin, Suijin Wang, Xucheng Yu, Haojing Luo, Haohan Wang
Checking sign-in…
Loading comments…