利用文本相似度计算的装配工作指令评估

An Assessment Method for Assembly Work Instruction Using Text Similarity Computation

针对现有四种装配工艺指令相似度评估方法(即词重叠、Jaccard、TF-IDF和潜在语义分析LSA)缺乏系统性量化评估的问题,提出采用十个工作指令集的45个基于文本的装配指令作为案例对现有四种文本比较方法进行分析。统计假设测试表明,Jaccard方法模拟装配工作指令的相似度高于其他三种方法至少3.75%,且对同义或多义词不敏感;而LSA方法对同义词或多义词的敏感度总体上低于其他三种方法至少6.02%,且可用于检索自由文本的装备工作指令,避免了TF-IDF方法依赖于装配工作指令数据库的弊端。故而LSA方法更适合用于装配工作指令的评估。

Considering the lack of systematic and quantitative evaluation of four existing similarity evaluation methods(i.e. word overlapping, Jaccard, TF-IDF and latent semantic analysis(LSA)) for assembly process instructions, a case study of 45 text-based assembly instructions from 10 work instruction sets is presented to analyze four existing text comparison methods. Statistical hypothesis tests show that the similarity of simulation assembly work instructions of Jaccard method is at least 3.75% higher than that of the other three methods,and the Jaccard method is insensitive to synonyms or polysemous words. In addition, the sensitivity of LSA to synonyms or polysemous words is at least 6.02% lower than that of the other three methods. Meanwhile, LSA method can also be used to retrieve free-text equipment work instructions, avoiding the disadvantage of TF-IDF method depending on assembly work instruction database. Therefore, LSA method is more suitable for the evaluation of assembly work instructions.