PatSTEG: Modeling Formation Dynamics of Patent Citation Networks via The Semantic-Topological Evolutionary Graph (2402.02158v1)

Abstract: Patent documents in the patent database (PatDB) are crucial for research, development, and innovation as they contain valuable technical information. However, PatDB presents a multifaceted challenge compared to publicly available preprocessed databases due to the intricate nature of the patent text and the inherent sparsity within the patent citation network. Although patent text analysis and citation analysis bring new opportunities to explore patent data mining, no existing work exploits the complementation of them. To this end, we propose a joint semantic-topological evolutionary graph learning approach (PatSTEG) to model the formation dynamics of patent citation networks. More specifically, we first create a real-world dataset of Chinese patents named CNPat and leverage its patent texts and citations to construct a patent citation network. Then, PatSTEG is modeled to study the evolutionary dynamics of patent citation formation by considering the semantic and topological information jointly. Extensive experiments are conducted on CNPat and public datasets to prove the superiority of PatSTEG over other state-of-the-art methods. All the results provide valuable references for patent literature research and technical exploration.

• The paper introduces PatSTEG, a novel model combining semantic patent text analysis with topological network features to predict citation dynamics.
• The methodology leverages the CNPat dataset, demonstrating superior performance over state-of-the-art models on AUC, AP, and nDCG metrics.
• The research provides actionable insights by attributing citations to specific technological aspects, thereby enhancing patent analysis and strategic decision-making.

Introduction

Patent documents are foundational to industrial research and development. They safeguard intellectual property and provide insights into technological advancements. Understanding the intricate structure of patent citation networks, therefore, has significant implications for stakeholder strategies. Traditional approaches often neglect the rich textual content within patents, focusing predominantly on the network's topological traits. The paper "PatSTEG: Modeling Formation Dynamics of Patent Citation Networks via The Semantic-Topological Evolutionary Graph" addresses this oversight by proposing a comprehensive model that incorporates both semantic content and network topology to paper the dynamics of patent citation networks.

Methodology

The authors commence by forming a real-world dataset of Chinese patents, named CNPat, mined from patent texts and their citations, creating a citation network. The core innovation of their model, referred to as PatSTEG, is its treatment of the citation network's evolution, effectively capturing both temporal dynamics and textual attributes. PatSTEG performs a joint semantic-topological analysis which distinguishes the proposed model from earlier work. This dual approach is crucial as it tackles the challenge of sparse patent citations by enriching the network representation with semantic details from the text.

The semantic component of PatSTEG focuses on the textual analysis of patents, highlighting the usage of titles, abstracts, and claims to discern the technological significance within the patents. This is married with the model's capability to learn the topological structure of citation relationships, guided by an evolutionary process that aims to predict and rationalize the formation of links in the citation network.

Empirical Validation

To validate their approach, the authors conduct extensive experiments on the newly created CNPat dataset as well as established public databases. PatSTEG consistently outperforms state-of-the-art models across various metrics, such as Area Under the Curve (AUC), Average Precision, and Normalized Discounted Cumulative Gain (nDCG). These strong numerical results affirm the advantages of integrating semantic and topological information for citation prediction and analysis.

Significance

A striking feature of PatSTEG is its ability to attribute patent citations to specific technological aspects. This multi-aspect link prediction provides a nuanced understanding of why certain patents cite others, enabling patent analysts to decipher patterns in technology influence and development trends. The interpretability of these semantic-topological links is illustrated through selected case studies from the CNPat dataset, where the reasons behind patent citations are visualized and explored.

Conclusion

The paper's proposed framework, PatSTEG, revolutionizes our approach to patent citation analysis. The ability of PatSTEG to leverage both the rich text within patent documents and the complex topology of citation networks yields a powerful tool for forecasting citation dynamics and for mining deeper insights from patents. The research lays a robust foundation for future exploration in this domain, with potential to extend the model to domains beyond patents that exhibit similar structural and content complexities.