Author Disambiguation via Semantic and Network Representation Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ambiguity in author names across academic publications leads to errors in literature management and social network analysis, as existing methods struggle to accurately disambiguate authors with the same name due to the complexity and diversity of publication information.
Innovation Solution
A method combining network representation and semantic representation learning using heterogeneous networks and word2vec models to extract relationship and semantic features, followed by clustering and similarity threshold matching to accurately disambiguate authors with the same name.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If rule-based matching methods are used for author disambiguation, then the process is simple and easy to implement, but the accuracy is low due to the complexity and diversity of publication information
Solution Approach 1:
The patent replaces rule-based mechanical matching with representation learning methods. Specifically, it uses word2vec to transform publication information (title, abstract, authors, institutions, journals) into semantic vectors, and employs network representation learning (DeepWalk, LINE) to capture relational patterns. This substitution enables the system to handle the complexity and diversity of publication information automatically, achieving high disambiguation accuracy without requiring complex manual rule design.
Solution Approach 2:
The patent changes the representation parameters of publication information from raw text forms to transformed semantic representations. By applying word2vec and network embedding techniques, the system transforms discrete publication attributes into continuous vector spaces where semantic similarities can be captured. This parameter transformation enables more effective clustering and disambiguation while maintaining computational efficiency.
2Measurement precision
If representation learning methods are used to characterize publication information, then the disambiguation accuracy is improved, but the system complexity increases
Solution Approach 1:
The patent segments the publication information into distinct feature categories: title, abstract, authors, institutions, and journals. Each segment is processed independently through word2vec transformation, and then the results are integrated. This segmentation allows the system to handle complex publication data in manageable parts, reducing overall system complexity while maintaining high accuracy through comprehensive feature utilization.
Solution Approach 2:
The patent employs a universal representation learning framework that can handle multiple types of publication information (title, abstract, authors, institutions, journals) through a single word2vec model. This multi-functional approach consolidates what would otherwise require separate processing systems into one unified framework, improving accuracy while controlling complexity through shared computational resources.
3Measurement precision
If clustering methods are applied to group similar theses, then author disambiguation is achieved, but discrete publications with less obvious features难以处理
Solution Approach 1:
The patent introduces semantic representations as an intermediary between raw publication information and clustering results. Word2vec vectors serve as mediators that capture semantic similarities, while network embedding representations act as intermediaries that capture relational patterns. These intermediary representations enable the system to handle discrete publications with less obvious features by transforming them into a unified semantic space where clustering can effectively operate, thereby improving both accuracy and adaptability.
Data Source
AI summary
The present invention discloses a method for disambiguating between authors with a same name on basis of network representation and semantic representation. The method includes: extracting semantic and discrete features of each publication in a target publication library; calculating a similarity between the theses based on the discrete features to obtain a relationship similarity matrix of the theses; if the publication has no common author or institution with other theses, it is added into an discrete publication set; calculating a semantic similarity matrix of the theses based on the semantic features of the theses; and adding theses which do not contain the semantic features in the target publication library to the discrete publication set; performing weighted summation on the relationship similarity matrix and the semantic similarity matrix to obtain a publication similarity matrix and clustering the same; adding theses which do not belong to any cluster to the publication discrete set; and allocating the theses in the discrete publication set to corresponding clusters by using a method based on similarity threshold matching. The present invention enables disambiguation between the authors of the same name of theses with high accuracy.