制造业众包平台服务资源的特征构建及聚合匹配

Feature construction and aggregation matching of service resources in manufacturing crowdsourcing platform

针对制造业众包平台中服务资源的特征构建缺乏规范、聚合匹配不准确的问题,提出了基于双向编码表示(BERT)模型的特征构建方法和基于多粒度特征交互的聚合匹配模型(MFIM)。前者改进了经典的BERT模型,采用停止符掩码操作和基于分类任务的学习策略,实现了服务资源所具有的语义特征建模,精准衡量了平台商所具有的服务资源能力水平,为大规模的语义特征体系的构建提供了有力支撑。后者针对服务资源的语义粒度差异较大的问题,采用了多层次的扩张卷积结构,并构建了跨粒度信息交互的关联矩阵。这有效解决了服务资源的聚合匹配及语义辨识问题。实验表明,所提出的一系列研究方法能够有效地建模服务资源文本,为服务资源的深入挖掘和精准的匹配检索提供了一种新的思路和手段。

For the problem of non-standardized and inaccurate feature construction and aggregation matching of service resources in manufacturing crowdsourcing platforms,a feature construction method based on the Bidirectional Encoding Representation(BERT)model and a Multi-granularity Feature Interactive Matching model(MFIM)were proposed.The former improved the classical BERT model by adopting the learning strategy of stopper mask operation and classification task to realize the semantic features modeling possessed by the service resources,which provided a strong support for the construction of large-scale semantic feature system.The latter addressed the problem of differences in the semantic granularity of service resources,employed a hierarchical dilation convolution structure and constructed a correlation matrix for cross-granularity information interaction,which effectively solved the problem of aggregated matching of service resources.Experiments showed that the proposed series of research methods could effectively model the service resource texts,providing a new idea and means for deeper mining and accurate matching retrieval of service resources.