What problem does it solve? Bibliographic exports from WoS, CNKI, Scopus, and PubMed contain noise from wildcard overflow, cross-language homographs, and weakly related records, which distorts bibliometric statistics and CiteSpace visualizations. This Skill provides a 16-stage pipeline that semantically cleans literature datasets, maintains a traceable mapping table, and exports analysis-ready files. ## Core Features & Use Cases - Semantic Noise Cleaning: Detects source platform and format, builds a mapping table as the single operation hub, and judges each record through a three-layer noise model (word-form, topic, boundary) with iterative multi-round refinement. - Quality Assurance Loop: Learns from manual review decisions, maintains a reusable noise pattern library, detects duplicates, and runs a six-dimension critical audit comparing expected versus actual noise rates. - Analysis & Export: Exports cleaned data in original WoS plain text format for direct CiteSpace import, plus keyword burst detection (Kleinberg algorithm), basic statistics, and network pruning references (Pathfinder, MST). - Use Case: A researcher exports 5,000 records from WoS using a wildcard query like meme*, then uses this Skill to remove homograph noise (memetic algorithms, French même), deduplicate records, audit cleaning quality, and produce a burst-analysis-ready dataset. ## Quick Start Start cleaning my exported literature file by detecting its source platform and building the mapping table for semantic noise removal.
