Adaptive Resonance Theory Based Two-Stage Chinese Name Disambiguation

Wang, Xin; Liu, Yuanchao; Wang, Xiaolong; Liu, Ming; Liu, Bingquan

doi:10.4236/ijis.2012.24011

Cited by 2 publications

(1 citation statement)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The method uses features to calculate the similarity between articles and then cluster them. For example, Xu et al [7] used a center-based approach after hierarchical clustering to find nodes that deviated from the results, and further used SVM method and then clustered them; Wang et al [8] handled the name ambiguity task based on the two-stage disambiguation theory of adaptive resonance, which first clusters the string information of the names to be disambiguated and then merges the similar clusters. Although the approach in [8] is better than the traditional supervised ones, due to the limitation of features similarity, it is difficult to determine the similarity threshold, the disambiguation accuracy of this type of method is relatively low.…”

Section: Related Workmentioning

confidence: 99%

A Scholar Disambiguation Method Based on Heterogeneous Relation-Fusion and Attribute Enhancement

Deng

2020

IEEE Access

View full text Add to dashboard Cite

Name disambiguation becomes increasingly important in information retrieval in the era of big data, while on how to further improve the accuracy of disambiguating duplicate names, the existing models are encountering many problems or facing many challenges, such as 1) how to capture and sufficiently use the global structure features of the network; 2) how to handle the complexity of feature data sets; 3) how to differentiate different types of feature relations, and 4) how to deal with the feature information missing etc. All these challenges are very important issues in further improving the accuracy. In order to address the above issues, this paper proposed a novel model which is called HRFAENE (Heterogeneous Relation Fusion and Attribute Enhanced Network Embedding Model). This model considers both feature network structure information (multiple relations) and document attribute features. The feature network structure information is represented by a scalable loss function which is designed based on pairwise constraints. The document attribute features are comprehensively extracted through multiple heterogeneous information networks which are constructed based on strong features. In order to better identify the disambiguation entity, this model uses weak features as node attributes in strong feature networks and iteratively learns network structure information and node attribute information. The experimental results show that the proposed model significantly outperforms the existing models in the disambiguation accuracy and has good stability, indicating that the model is very effective as expected and can be applied in reality.

show abstract

Section: Related Workmentioning

confidence: 99%