TechPat: Technical Phrase Extraction for Patent Mining

In recent years, due to the explosive growth of patent applications, mining has drawn extensive attention and interest. An important issue is that recognizing technologies contained in patents, which serves as a fundamental preparation for deeper analysis. To this end, article, we make focused study on constructing technology portrait each patent, i.e., recognize technical phrases concerned it, can summarize represent patents from perspective. Along line, critical challenge how analyze unique characteristics illustrate them with definite descriptions. Therefore, first generate detailed descriptions about existing based different criteria, including various previous works, practical experience, statistical analyses. Then, considering complex structure documents, such multi-aspect semantics multi-level relevances, further propose novel unsupervised model, namely TechPat, not only automatically massive but also avoid need expensive human labeling. After that, evaluate extraction results aspects. Specifically, evaluation metric called Information Retrieval Efficiency (IRE) quantify performance extracted new Extensive experiments real-world data demonstrate TechPat model effectively discriminate greatly outperform methods. We apply two application tasks, search classification, where experimental confirm wide prospects phrases. Finally, discuss generalization ability our proposed