基于产品生命周期的专利技术主题演化分析

Analysis of Patent Technology Topic Evolution Based on Product Life Cycle

专利技术主题演化蕴含着产品技术的发展脉络,如何对其进行准确把握对产品的技术研究人员至关重要。针对现有的专利技术主题演化分析研究中的不足,提出了一种基于产品生命周期的专利技术主题演化分析方法:针对现有时间特征划分方式会导致产品发展前期主题信息被淹没的问题,提出一种产品生命周期划分方法,利用专利增长规律和局部主题语义相似度划分产品生命周期。然后,针对现有主题挖掘方法对专利文献的适用性差、主题技术特点不明显等问题,提出MW-LDA(multiple weighted latent Dirichlet allocation)的专利技术主题提取方法,多角度改进词汇权重并构造复合权值,优化LDA模型生成特征词的过程,使其更适用于专利文本。接着,提出产品专利技术主题演化分析方法,利用阶段间的主题关联强度,实现产品生命周期不同阶段的技术主题演化分析。最后,在由电动车辆动力装置的专利构成的语料库上的实验结果以及对比实验表明,本文方法能够有效地发现专利的技术主题,并分析产品的发展态势,可为科学研究和科技政策制定提供支持。

The evolution of the patent technology subject contains the development context of product technology.Grasping it accurately is very important for product technology researchers.In view of the shortcomings of the existing patent technology subject evolution analysis,a patent technology subject evolution analysis method based on product life cycle is proposed.First,aiming at the problem of the existing time feature partitioning method leading to an inundation of topic information in the early stage of product development,a product life cycle partitioning method is proposed that uses the patent growth law and local topic semantic similarity to divide the product life cycle.Second,aiming at the problems of poor applicability of the existing subject mining methods to the patent literature and the lack of obvious technical characteristics of the subject,a Multiple Weighted Latent Dirichlet Allocation(MW-LDA)patent subject extraction method is proposed.MW-LDA is used to improve the term weights from multiple angles and construct compound weights to optimize the process by which an LDA model generates feature words,making it more applicable to patent texts.Then,a method for the evolution analysis of the product patent technology topic is proposed,and the evolution analysis of the product patent technology topic in different stages of the product life cycle is realized using the strength of topic correlation between phases.Finally,the experimental results and comparative experiments on a corpus of patents of electric vehicle power units show that the method can effectively identify the technical topics of patents and analyze the development trend of products,which can provide support for scientific research and science and technology policymaking。