A Granular Computing Methodology to Improve Record Quality for Master Data Management

主数据管理（MDM）是一个企业流程，可确保公司主数据资产的准确性、一致性、适当管理、隐私性、语义一致性和责任性。主数据是描述公司核心实体的一组一致的业务标识符和属性。公司MDM解决方案的一个关键方面是为每个感兴趣的客户创建“黄金记录”。这些黄金记录使用分布式数据存储库中的异构信息集进行链接、同步和聚合。随着存储成本的降低和大数据技术的成熟，公司可以存储更复杂、更庞大、更多样（模式化、半结构化、非结构化、社交媒体等）的标识符和属性。 黄金记录的自动创建提出了七个重大挑战：㈠汇总在不同信息具体程度上获得的数据; ㈡补偿不完整或缺失的属性值; ㈢减弱不精确或不准确数据的影响; ㈣尽量减少误报率，即避免合并属于不同客户的记录; ㈤确保属于同一客户的记录实际上已合并，即尽量减少假阴性率; ㈥减少最终用户手工核对的次数; ㈦减少处理自动化过程中出现的异常情况的临时规则的数量。 通过与我们的工业合作伙伴InfoMagnetics Technologies Corporation密切合作，我们建议通过为MDM设计和开发粒度计算方法来解决上述挑战，以提高客户记录质量。我们将采用新的数据聚合技术，具有不同程度的信息特异性和不同的数据存储库，以提高组装的黄金记录的整体质量。

Master Data Management (MDM) is a corporate process that ensures the accuracy, uniformity, proper stewardship, privacy, semantic consistency, and accountability of a company's master data assets. Master data is the consistent set of business identifiers and attributes that describe the core entities of a company. A key aspect of a company's MDM solution is the creation of a "golden record" for every client of interest. These golden records are linked, synchronized, and aggregated using sets of heterogeneous information across distributed data repositories. As storage costs decrease and big data technologies mature, companies can store more complex, voluminous, and varied (schematized, semi-structured, unstructured, social media, and so on) identifiers and attributes. The automated creation of golden records presents seven significant challenges: (i) aggregating data acquired at varying levels of information specificity; (ii) compensating for incomplete or missing attribute values; (iii) attenuating the effects of imprecise or inaccurate data; (iv) minimizing the false positive rate, that is, avoid merging records belonging to different clients; (v) ensuring that records belonging to the same client are, in fact, merged, that is, minimizing the false negative rate; (vi) reducing the number of manual, end user, reconciliations; (vii) and reducing the number of ad hoc rules to deal with anomalies from the automated process. In close collaboration with our industrial partner, InfoMagnetics Technologies Corporation, we propose to address the above challenges via the design and development of a Granular Computing Methodology for MDM in order to improve client record quality. We will employ novel aggregation techniques on data, with varying levels of information specificity and across disparate data repositories, in order to improve the overall quality of the assembled golden records.