Semi-supervised Entity Alignment via Joint Knowledge Embedding Model and Cross-graph Model

Abstract

Entity alignment aims at integrating complementary knowledge graphs (KGs) from different sources or languages, which may benefit many knowledge-driven applications. It is challenging due to the heterogeneity of KGs and limited seed alignments. In this paper, we propose a semi-supervised entity alignment method by joint Knowledge Embedding model and Cross-Graph model (KECG). It can make better use of seed alignments to propagate over the entire graphs with KG-based constraints. Specifically, as for the knowledge embedding model, we utilize TransE to implicitly complete two KGs towards consistency and learn relational constraints between entities. As for the crossgraph model, we extend Graph Attention Network (GAT) with projection constraint to robustly encode graphs, and two KGs share the same GAT to transfer structural knowledge as well as to ignore unimportant neighbors for alignment via attention mechanism. Results on publicly available datasets as well as further analysis demonstrate the effectiveness of KECG. Our codes can be found in https: //github.com/THU-KEG/KECG.

The Knowledge Graph Track at OAEI

“…With our ongoing evaluation efforts -the KG track will be part of OAEI 2020 again -we provide a testbed for new and improved solutions in that field. Moreover, a few approaches for the task of knowledge graph matching have been published in the recent past (e.g., [20,27,29]), which have been evaluated on different datasets (and in closed domain settings only), hence, their results are not directly comparable. By providing a generic benchmark including both open and closed domain settings, we enable a more systematic comparison for such works in the future.…”

Towards Knowledge Graphs Federations: Issues and Technologies

“…At present, there are mainly three categories of EA models. The first group of studies merely harness the KG structure to align entities, and representative work includes [4,6,14]. The second category of methods iteratively label most likely aligned entity pairs as the additional training data and progressively enhance the alignment performance [22,41,42].…”

REA: Robust Cross-lingual Entity Alignment Between Knowledge Graphs

“…Then several methods, such as ITransE [50], IPTransE [50] and BootEA [32] leverage the bootstrapping strategy to address the lack of labeled data. Further, semi-supervised approaches [18,26,27] were developed to utilize the unlabeled data for enhancing the performance of entity alignment. Moreover, several works consider to jointly model the structure and attribute information of KGs [6,31,34], or the structure and relation information [29,51] or the structure, relation and attribute information together [49]; 2) Graph Neural Network-based methods.…”

“…Existing methods are often designed to address the alignment problem in a supervised way with human-designed features [23] or with entity representations learned from KG embedding approaches [6,7,31,38]. Also, a few semi-supervised methods [18,26,27] were proposed to make use of unlabeled data to enhance the performance of supervised entity alignment. Supervised or semi-supervised alignment methods have made remarkable discovery of semantically related entities.…”