One-to-many comparative summarization for patents

• Published in: Scientometrics 2022-04, Vol.127 (4), p.1969-1993
• Main Authors: Liu, Zheng, Zhang, Jialing, Qin, Tingting, Qu, Yanwen, Li, Yun
• Format: Article
• Language: English
• Subjects: Computer Science; Feature extraction; Greedy algorithms; Information Storage and Retrieval; Infringement; Library Science; Optimization; Patent applications; Sentences
• ISSN: 0138-9130; 1588-2861
• DOI: 10.1007/s11192-022-04307-8

Patents bring technology companies commercial values in modern business operations. However, companies have to bear the high cost of handling patent applications or infringement cases. A common yet expensive task among these jobs is to analyze relevant patent literature. Lengthy and technically complicated patents require a large number of human efforts. This paper focuses on automatically analyzing the similar contents between a patent and its relevant literature, relevant patents specifically, to help experts review the similarities among these patents. We formulate this as a one-to-many document comparison problem by generating a comparative summary of a given patent and its relevant patents. We extract essential technical features from semantic dependency trees based on sentences in claims and construct a multi-relational graph to model the relevance between features and patents. The key to generating the comparative summary is selecting comparative essential technical features, which we formulate as an optimization problem and solve by a fast greedy algorithm. Experiments on real-world datasets and case studies demonstrate the effectiveness and efficiency of the proposed methods.