基于PSO-SOM算法的专利价值的分类研究

Research on Patent Value Classification Based on PSO-SOM Algorithm

根据世界知识产权组织统计的数据,我国的专利申请数量多年来名列前茅,专利授权数量也每年都在增长,因此学术界及社会各界越来越关注专利价值的研究。传统的专利价值评价方法大都是采用定性和定量结合的决策方法,比如:层次分析法、综合评价法等,已经不能满足当今大规模大体量数据的分析需求。利用大数据环境下的机器学习方法不但可以降低人力成本,还可以提高分类的准确率和效率。本文提出一种基于粒子群优化算法–自组织映射网络(PSO-SOM)的专利价值分类模型,依据专利价值指标,从incoPat专利数据库选取了5000条专利数据进行实证研究。通过PSO-SOM聚类得到了有效的专利价值标签,利用随机森林算法对初始专利价值进行指标重要性排序,并逐个依次将指标引入朴素贝叶斯模型中进行分类,能够有效提高朴素贝叶斯分类模型的准确率和效率。

According to statistics from the World Intellectual Property Organization, the number of patent applications in China has been among the top for many years, and the number of patent authorizations is also increasing every year. Therefore, the academic community and various sectors of society are increasingly concerned about the study of patent value. The traditional patent value evaluation methods mostly use a combination of qualitative and quantitative decision-making methods, such as the Analytic Hierarchy Process, Comprehensive Evaluation Method, and so on, which can no longer meet the analysis needs of large-scale and large-volume data today. The use of machine learning methods in the big data environment can not only reduce labor costs but also improve classification accuracy and efficiency. This article proposes a patent value classification model based on particle swarm optimization and self-organizing mapping network (PSO-SOM). Based on patent value indicators, 5000 pieces of patent data were selected from the incoPat patent database for empirical research. Effective patent value labels were obtained through PSO-SOM clustering, and the initial patent value was ranked in importance using the random forest algorithm. The indicators were introduced into the Naive Bayes model for classification one by one, which can effectively improve the accuracy and efficiency of the Naive Bayes classification model.