Abstract: To improve the product key design knowledge discovery system of cloud service platform and further improve the matching efficiency of requirements and services,a product design key features recognition method based on Bidirectional Encoder Representations from Transformers (BERT) and random Lasso was proposed.First,the user feedback of real products was adopted in the experiment and was annotated manually.Based on BERT model,the output layer was built,and the named entity recognition model in the design field was trained to realize the automatic recognition of explicit design features.Experimental results showed that the proposed method could achieve better performance,precision and recall,and F1 scores were 90.55%,97.16% and 93.68% respectively.Simultaneously,a new idea of knowledge transfer was proposed.In the current big data environment,the key design features contained in it could be mined and reused by using random lasso algorithm,so as to realize the accurate positioning of implicit design features.

Key words: industrial design, user requirements, bidirectional encoder representations from transformers, named entity recognition, random Lasso, product design

摘要： 为完善云服务平台产品设计知识发现系统,同时进一步提升需求与服务的匹配效率,提出一种基于转换器的双向编码表征(BERT)和随机Lasso的产品关键设计特征识别方法。首先,实验采用真实产品用户反馈数据集并对其进行人工标注,以BERT预训练语言模型为基础,建立输出层以训练设计领域命名实体识别模型,实现对显性设计特征的自动识别。实验表明,所提方法可以实现较好的性能,精确率、召回率、F1分数分别为90-55%、97-16%和93-68%。同时,提出一种知识迁移思想,在当前大数据环境下,利用随机Lasso算法挖掘其中蕴含的关键设计特征并加以重用,实现了对隐性设计特征的精确定位。

关键词: 工业设计, 用户需求, 基于转换器的双向编码表征, 命名实体识别, 随机Lasso, 产品设计