Results of the summarizing process for the arXiv paper: 2107.04880v1

In the field of patent research, there has been a limited amount of work conducted on building knowledge graphs to explore the rich semantic relationships within patent documents. Previous studies have focused on using natural language processing methods to construct engineering knowledge graphs or predict emerging technologies based on labeled networks. However, a gap exists in integrating patent knowledge graphs with the latest natural language processing techniques, such as pretraining models, attention mechanisms, and graph neural networks. To address this gap, this paper introduces a novel approach called PatentMiner for predicting potential patents by leveraging knowledge graphs (KG) and graph attention mechanisms. The methodology involves constructing a patent knowledge graph over time by performing named entity recognition and relation extraction from patent documents. Subsequently, Common Neighbor Method (CNM), Graph Attention Networks (GAT), and Context-enhanced Graph Attention Networks (CGAT) are proposed to conduct link prediction within the constructed knowledge graph to uncover potential triples. Furthermore, each patent is defined on the knowledge graph through co-occurrence relationships, representing each patent as a fully connected subgraph containing all its entities and co-occurrence relationships within the graph. A new patent prediction task is introduced where a fully connected subgraph with newly added prediction links is predicted as a new patent. Experimental results demonstrate that the proposed patent prediction approach effectively predicts new patents, with Context-enhanced Graph Attention Networks outperforming baseline methods. However, there is still significant room for improvement in the proposed patent vacancy prediction task. Overall, this study highlights the importance of utilizing knowledge graphs and advanced natural language processing techniques for mining hidden semantic relationships in patents and predicting future innovations in intellectual property.

• - Limited work on building knowledge graphs in patent research
• - Gap in integrating patent knowledge graphs with advanced natural language processing techniques
• - Introduction of PatentMiner approach for predicting potential patents using knowledge graphs and graph attention mechanisms
• - Methodology involves constructing a patent knowledge graph through named entity recognition and relation extraction
• - Use of Common Neighbor Method, Graph Attention Networks, and Context-enhanced Graph Attention Networks for link prediction within the knowledge graph
• - Definition of each patent on the knowledge graph through co-occurrence relationships
• - Introduction of a new patent prediction task based on fully connected subgraphs with prediction links
• - Experimental results showing effectiveness of the proposed patent prediction approach, with Context-enhanced Graph Attention Networks performing well
• - Importance of utilizing knowledge graphs and advanced natural language processing techniques for mining hidden semantic relationships in patents and predicting future innovations

Summary- People are working on using pictures to learn more about patents. - They want to connect these pictures with fancy computer language skills. - A new way called PatentMiner helps find possible future patents using pictures and special attention techniques. - They make a picture of the patents by finding important words and how they are related. - They use different methods to guess which patents might come next based on the picture. Definitions- Patents: Exclusive rights given by a government to an inventor for their invention. - Knowledge graphs: Visual representations of information showing how things are connected or related. - Natural language processing: Using computers to understand, interpret, and generate human language data. - Named entity recognition: Identifying important words or phrases in text that refer to specific entities like people, places, or organizations. - Relation extraction: Finding connections between different pieces of information in text.

In the world of intellectual property, patents play a crucial role in protecting and promoting innovation. However, with millions of patents being filed every year, it can be challenging to keep track of all the information contained within them. This is where patent research comes into play – using various techniques and tools to analyze and extract valuable insights from patent documents. One area that has seen limited exploration in patent research is the use of knowledge graphs. These are powerful tools that allow for the representation and exploration of complex relationships between entities or concepts. In this context, knowledge graphs can be used to uncover hidden semantic relationships within patents, providing a deeper understanding of their content. Previous studies have primarily focused on using natural language processing (NLP) methods to construct engineering knowledge graphs or predict emerging technologies based on labeled networks. However, there is still a gap in integrating these advanced NLP techniques with patent knowledge graphs. To address this gap, a team of researchers has introduced a novel approach called PatentMiner for predicting potential patents by leveraging knowledge graphs and graph attention mechanisms. The methodology involves constructing a patent knowledge graph over time by performing named entity recognition and relation extraction from patent documents. The first step in this process is identifying relevant entities within the text using named entity recognition (NER). These could include inventors' names, company names, technology terms, etc. Next, relation extraction techniques are applied to identify connections between these entities such as co-occurrence relationships or other semantic associations. Once the initial patent knowledge graph is constructed, three different methods are proposed for conducting link prediction within the graph: Common Neighbor Method (CNM), Graph Attention Networks (GAT), and Context-enhanced Graph Attention Networks (CGAT). These methods utilize advanced NLP techniques such as pretraining models and attention mechanisms to uncover potential triples – sets of three related entities – within the graph. Furthermore, each individual patent is defined on the knowledge graph through its co-occurrence relationships. This means that each patent is represented as a fully connected subgraph containing all its entities and co-occurrence relationships within the larger graph. The researchers also introduce a new patent prediction task where a fully connected subgraph with newly added prediction links is predicted as a new patent. This allows for the identification of potential future innovations based on existing patents and their semantic relationships. Experimental results demonstrate that the proposed approach effectively predicts new patents, with Context-enhanced Graph Attention Networks outperforming baseline methods. However, there is still room for improvement in the proposed patent vacancy prediction task, highlighting the need for further research in this area. Overall, this study highlights the importance of utilizing knowledge graphs and advanced NLP techniques for mining hidden semantic relationships in patents. By doing so, it enables researchers to gain deeper insights into patent content and predict future innovations in intellectual property. The integration of these techniques has great potential to enhance traditional patent research methods and aid in decision-making processes related to innovation and intellectual property protection.