Knowledge graphs (KGs) have become a widely adopted standard for knowledge representation in the semantic web. Currently, significant efforts have been invested in constructing a KG with a primary focus on named entity recognition and relation extraction. Several domain knowledge graphs have already been established, including those for food and manufacturing.
Existing knowledge graph construction methods are mostly designed for English and do not address specialized domains. They often lack consideration for the characteristics of data in other languages and specific fields.
To solve the problems, a research team led by Xuanhua Shi published their new research on 15 October 2025 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The team proposed propose a novel entity-type-enriched cascaded neural network (E²CNN) that considers the overlap triple problem and entity-type information to construct a Chinese financial knowledge graph.
In their research, they analyzed the challenges and data characteristics of the Chinese financial relation extraction task. To address the prevalent issue of overlapping relations, where the same entity appears in multiple relations within a single sentence, they employed a cascaded neural network. Additionally, they found a connection between the relations and the associated entity types in the dataset, so they integrated entity type information into the relation extraction model.
First, the span-type-identification module identifies the entities and corresponding types and obtains entity-enriched representations by incorporating the predicted types’ information into the original corpus. Then in the cascaded neural network module, the subject extraction module predicts both the beginning and end position of the subject, leveraging type-enriched information, object extraction detects all objects corresponding to the subjects obtained by the subject extraction under each predefined relation.
Future work can focus on finding more efficient span-type identification models to reduce the erroneous information introduced by this module and improve the overall model performance.
DOI:10.1007/s11704-024-3983-6