REVIEW & COMMENTARY
Chen Liang, Chen Lili, Xu Haiyun, Wei Chao, Su Na, Shang Weijiao
[Purpose/Significance] Patent mining is a significant means to achieve technical intelligence from patent documents. Driven by recent advancement of AI technologies, not only have patent mining methods achieved competitive performance in terms of automation, intelligence, mining depth and accuracy, but also they have revealed a new paradigm of integrating datasets and algorithms together, which indicates a comprehensive survey of related research achievements and future development trends is urgent.[Method/Process] This paper linked the main steps of literature review into a closed loop “searching→screening→sorting→checking→expanding and re-searching” and kept updating and iterating. Review scope covered the related papers, patents, datasets, algorithm competition, patent information service platforms, and even code hosting and model hosting Websites. In a meanwhile, relevant information from expert interviews, competition player seminar and academic achievement reviews was also included in the narrative content.[Result/Conclusion] It finds that the types and quantity of patent resources are growing more rapidly than before, which paved the way for training and evaluating algorithms and models in an uniform standard. Forefront patent mining methods are closely following the pace of intelligent technology development to achieve technological upgrades and performance enhancements, while traditional methods such as statistical learning, manual rules, and software tools have been optimized and developed in the balance of learning costs, practical costs and method performances. The research scope of patent mining has achieved full coverage from data processing, normalization to basic patent service supporting and technological intelligence analysis, and the exploration of legal judgement prediction for patent lawsuit have been launched.