基于DBSCAN聚类改进随机森林算法的专利价值评估方法

对于专利价值的不确定性和影响因素的复杂性,以及评估工作中缺乏可操作性强并且科学高效的评估方法等问题,对价值评估指标体系进行分析,并使用随机森林算法选择最有效的指标集,同时基于DBSCAN(density-based spatial clustering of applications with noise)聚类选择高精度且一致性低的决策树子森林改进传统随机森林算法,使用改进前后的两种随机森林模型在专利数据样本上进行实验并比较。结果表明,改进的随机森林模型提升了传统模型的精度,在专利价值评估中具有一定的作用,总体上比较有效地反映了专利的价值度。

Research on Patent Value Evaluation Method Based on Random Forest Algorithm improved by DBSCAN Clustering

For the uncertainty of the patent value and the complexity of the influencing factors, as well as the lack of operability and scientific and efficient evaluation methods in the assessment work, this paper analyzed the value evaluation index system and used the random forest algorithm to select the most effective indicators. At the same time, based on DBSCAN clustering, this paper selected the high-precision and low-consistent decision tree sub-forest to improve the traditional random forest algorithm, and use the two random forest models before and after the improvement to experiment and compare on the patent data samples. The results indicated that the improved random forest model improves the accuracy of the traditional model and has a certain role in the evaluation of patent value. It generally reflects the value of patents effectively.