Abstract: In order to provide abundant source of information and depth information for product innovation design, a method for mining patent harmful performance knowledge was proposed based on semantic association. First, harmful performances of the products were defined and classified, and the distribution and semantic characteristics of harmful performances in patents were analyzed accordingly. Second, combined with industrial patent full-text information, word2vec was employed to construct a text vector space model, then the sematic distances of words were calculated by cosine algorithm for merging synonyms. Third, part-of-speech and dependency syntax were introduced to obtain harmful performance keywords from technological background of the patents, and four classification rules were proposed to make association relationship among structural objects and harmful performances, then a database of product harmful performances was built. Finally, focusing on keywords of design scheme components, the related harmful performances were retrieved, and the occurrence probability of different harmful performances of the components was calculated using frequency formula. The domestic shower product patent was selected as an application case and compared with the other three algorithms to verify the feasibility and effectiveness of the proposed method.

Key words: harmful performance, knowledge mining, dependency syntax, semantic association

摘要： 为向产品设计过程提供更加丰富的知识来源，进一步获取专利深层信息，提出了基于语义关联的专利有害性能知识挖掘方法。首先，对产品有害性能进行定义和分类，据此对产品专利分布以及语义特点进行分析；其次，结合行业专利全文信息，引入word2vec算法构建文本向量空间，并利用余弦算法计算不同单词的语义距离，进而实现同义词合并；再次，利用词性组合以及依存关系获取行业专利技术背景中有害性能关键词，结合四种分类规则获取结构对象与有害性能的关联关系，并构建有害性能数据库；最后，针对设计方案元件关键词，检索关联有害性能知识，结合频次公式计算元件不同有害性能发生的概率。选择国内花洒产品专利为应用实例，与其他三种算法进行比较，验证了所提方法的可行性及有效性。

关键词: 有害性能, 知识挖掘, 依存关系, 语义关联