Author Name Disambiguation via Multi-Stage Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The ambiguity in author names across various publications due to inconsistent reporting and lack of standards hinders effective search and discovery of relevant publications attributed to individual authors, as different authors may share names and the same author may publish under multiple variants.
Innovation Solution
A computer-implemented system and method for disambiguating author names by clustering author name mentions based on similarity and feature sets, using a multi-stage process that groups names into buckets, forms compatibility sets, and merges clusters based on feature similarity, with human input for verification and assignment to author identities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If author names are reported consistently with standards, then search and discovery of publications would be improved, but current inconsistent reporting and lack of standards causes author ambiguity
Solution Approach 1:
The patent segments the author name disambiguation process into multiple stages: (1) grouping author name mentions into buckets based on name similarity, (2) forming compatibility sets within buckets, (3) clustering based on feature similarity, and (4) assigning clusters to author identities. This segmentation allows complex disambiguation to be handled through manageable sequential steps, reducing overall system complexity while improving accuracy.
Solution Approach 2:
The patent introduces intermediary structures including name buckets, compatibility sets, and feature sets as mediating elements between raw author name mentions and final author identity assignments. These intermediaries simplify the disambiguation process by breaking down the complex task into smaller sub-tasks that can be processed independently and systematically.
2Quantity of substance
If multiple author name variants are collected, then completeness of author attribution is improved, but ambiguity and difficulty in matching increase
Solution Approach 1:
The patent performs preliminary actions by pre-grouping author name mentions into buckets and compatibility sets before the main clustering process. This preliminary organization pre-sorts the data by name similarity and compatibility, making the subsequent feature-based clustering more efficient and reducing the difficulty of matching variants to author identities.
Solution Approach 2:
The patent changes parameters by using multiple feature sets (publication features, author name features, co-authorship features) to represent author name mentions. By transforming the problem into a multi-parameter space and using similarity thresholds on these parameters, the system can effectively match variants to author identities despite name inconsistencies.
3Measurement precision
If comprehensive feature comparison is performed, then author disambiguation accuracy is improved, but computational cost increases
Solution Approach 1:
The patent segments the computational task into hierarchical stages where name-based grouping is performed first (which is computationally lighter), followed by feature-based clustering only within smaller subsets. This segmentation dramatically reduces the total number of comparisons needed while maintaining high disambiguation accuracy.
Solution Approach 2:
The patent applies partial action by performing feature comparisons only on author name mentions that have passed through the preliminary name-based filtering and compatibility set formation. This partial application of computational resources to the most promising candidates achieves high accuracy while significantly reducing overall computational cost compared to exhaustive comparison.
Data Source
AI summary
Described herein are computer-implemented systems and methods for automatically disambiguating author names for a plurality of publications so as to create clusters of author name mentions that are with high probability associated with a single author identity for each cluster. Also described are systems and methods for assigning the clusters to respective unique author identities, automatically and/or based on human input (e.g., as received from authors, co-authors, or administrative curators).