Distant Annotation for Training Data Preparation
summary
Claims that distant supervision produces labeled relation-extraction triples by aligning a knowledge base (Freebase/Wikidata/Wikipedia) with raw text, turning subject–object pairs into training examples. Implements a fast string-matching pipeline using Aho-Corasick (C-backed, Python wrapped) to locate pairs at scale and reports it handles on the order of 100k subject pairs / 1M articles so annotation runs in hours not weeks, producing noisy but usable training data.
AI-written summary · about
related talks
01h 30mRelation extraction from texts
30mEfficient data labelling with weak supervision
26mDeep Learning Architectures for Semantic Relation Detection Tasks
37mCreating Knowledgebases from unstructured text
58mRelation Extraction from Online PH News Source
34mExposing Dark Data in the enterprise with custom NLP
33mWeak supervision: a new paradigm for unreliable data
29mAnnotating data the right way