AffRoDB: Curated Mappings of affiliation strings to ROR identifiers

AffRoDB is a curated dataset that contains mappings of raw affiliation strings from Crossref to Research Organization Registry (ROR) IDs. Each dataset record has been evaluated by multiple experts who mapped an affiliation string to a ROR ID that represents the respective organization or its parent. Each record includes the DOI of the respective publication, the affiliation string used, the expert judgments, and final IDs that have been assigned to the string based on the expert judgements. The aim of this dataset is to facilitate the evaluation and benchmarking of methodologies for institutional affiliation disambiguation by offering a high-quality, transparently created, and frequently updated ground truth dataset. This version of the dataset is based on the January 2024 release of the ROR database. Dataset preparation We randomly selected approximately ~1500 records from Crossref, each with non-empty affiliation lists. Each affiliation string was evaluated by two experts (PhDs or PhD candidates) who were tasked to find any relevant ROR ID and assign a label to the string: EXACT: If the ROR ID found corresponded precisely to the most specific organization mentioned in the string. ANCESTOR: If the ROR ID corresponded to a parent organization mentioned in the string. VAGUE: If no relevant ROR ID was detected or the expert could not confidently assign a corresponding ROR ID. In cases of disagreement between the two experts, a third expert reviewed the affiliation string and finalized the ROR IDs and their labels. By removing the affiliations labeled as VAGUE, we produced a dataset of raw affiliation strings along with their corresponding ROR links. A summary of key dataset statistics is provided in the table below. Raw affiliation strings 1,374 DOIs 1,261 ROR IDs 1,484 Unique ROR IDs 1,048 Exact matches 381 Ancestor matches 1,103 Dataset structure The dataset is provided in a JSONL format, where each line represents a single JSON object. Each JSON object contains four main fields: raw_affiliation_string: A string representing the raw affiliation data. extracted_dois: An array of DOIs (Digital Object Identifiers) from which the affiliation was extracted. expert_judgements: An array of objects representing evaluations by experts, where each object includes: expert_id: The ID of the expert, a positive integer. matches: An object with two fields: exact: An array of strings denoting exact matches. ancestor: An array of strings denoting matches to parent organizations. final_judgement: An object representing the final evaluation that includes: exact: An array of strings denoting exact matches. ancestor: An array of strings denoting matches to parent organizations.