Knowledge Graphs in Principle and Practice: Extraction, Consolidation, Storage/Inference, and Access [4/8]
Episode notes
Module IV: The Construction Pipeline
This module addresses the practical "cleaning" and transformation of raw data into structured knowledge.
The Four-Phase Pipeline: Extraction, Consolidation, Storage/Inference, and Access.
Knowledge Acquisition: Named Entity Recognition & Disambiguation (NERD) and Relation Extraction.
Entity Resolution (ER): The "deduplication" challenge.
Strategies: Blocking (to reduce search space) and Similarity Metrics (Jaccard, Levenshtein, Jaro-Winkler).
Methodologies: Comparing Rule-based/Classical ML vs. Deep Learning (DeepMatcher).
Benchmarking and Evaluation: Mastering Mean Reciprocal Rank (MRR) and Hits@K, while avoiding data leakage in datasets like FB15k-237 and WN18RR.
HALLUCINATION CHECK: The bots say the next episode is module "V" instead of "5". They remain quirky as ever.