Heterogeneous Dependency Graph-Guided Attentionfor Patent Representation Learning
A novel approach to patent representation learning has been introduced through the Patent Heterogeneous Attention Graph Encoder (PHAGE), which addresses the limitations of pre-trained language models by incorporating the dependency hierarchy among claims. This method constructs a typed graph to differentiate between legal citations and technical relations, thereby enhancing the accuracy of patent classification and retrieval.
WPN Brief
- What Happened
A novel approach to patent representation learning has been introduced through the Patent Heterogeneous Attention Graph Encoder (PHAGE), which addresses the limitations of pre-trained language models by incorporating the dependency hierarchy among claims. This method constructs a typed graph to differentiate between legal citations and technical relations, thereby enhancing the accuracy of patent classification and retrieval.
- Why It Matters
The development of PHAGE is significant as it improves the handling of complex patent data, potentially leading to more reliable outcomes in patent analysis and innovation, which is crucial for legal and technological advancements in the field.