In this paper we describe the design of a reputation framework for an information management system under active development. The integration of a reputation framework with an IMS is a novel combination that can produce a distinctly more e↵ective business intelligence tool. claim One or more assertions made of a datum. data source A computer system that stores data such as a database or file system.
Neustar is a data analytics and intelligence services company that operates
several large database systems. To eciently manage these numerous, disparate
systems, we are developing an Information Management System (IMS) that maps
technical data models using a standard set of ontologies. The IMS is an online
community for employees where they can share, classify and discover metadata
about various Neustar data sources. Its main purpose is to assist users in
achieving two main objectives: a) reducing costs by utilizing existing information and
b) increasing revenues by creating new information [
With these objectives in mind, users must have the ability to make value judgments about data sources relative to one another. Such data sources may number in the hundreds and the datum contained therein may number in the tens of thousands. The majority of these entities will lack significant value for data science, and those that are valuable will risk being lost in a deluge of information. Therefore it is imperative that the system establish a bias towards meaningful datum by highlighting interestingness. A well-crafted reputation framework can excel at doing exactly this. Glossary data steward A individual or group of individuals holding domain-specific knowledge of an information system. datum An instance of metadata mapped to an atomic data field. This includes, for example, columns in a relational database or entities defined in an XML schema. interestingness A scalar value indicating the suitability for inclusion in further analysis. reputation A qualitative measure that informs a value judgment about a datum or user.
IMS Information Management System. 3
The framework is comprised of several reputation models, each of which computes one or more scores for a resource type. A fixed set of claims serve as inputs to each model which assigns numerical values to them and passes them through a series of mathematical filter processes. Models are distinguished by their input selection, process configuration, and output scores. The IMS utilizes a fixed ontology to define claims that include appropriate business and technical classifications for data within the subject systems. The essential claims of the datum model are summarized in Table 1. The IMS also incorporates techniques to simplify crowd sourcing the classification of datum by data steward s. However, we have concluded from early usage, that a simple classification process is insucient. Classifications can be subjective, and classification sparseness results in under utilization of the system. As a consequence, methods to encourage accurate and complete classification will be implemented to enrich the overall ecacy of the system. Name Description classified Data steward classified datum from the business domain ontology described Data steward entered a description discussed User participated in a discussion topic about datum emailed User emailed the link to the datum page to another user flagged User informed the data steward about insucient or inaccurate details watched User will be notified of future updates by other users wanted User requested access to the datum from the data steward
In the IMS, datum is an atomic unit of data. Its’ classification results in queriable metadata, and can relate to a column in a relational database or an element, attribute or phrase in a document. At the time of writing, the system had over 7,000 fields from merely four data sources. Even at this watermark, the task of finding interesting datum is impractical for any user community. As more data sources are imported into the system, this task will be become impossible even if the datum population grows sub-linearly. Therefore it is imperative that the system is capable of identifying and highlighting interesting datum to facilitate user objectives.
In Figure 1 we describe the simplified model for calculating the
interestingness reputation score for datum. Our approach is informed by [
We have described a realistic blue print for a reputation system that is on the
roadmap of our IMS. Once implemented we think that it will dramatically
improve the quality of information that is retrievable by users, thus increasing its’
e↵ectiveness as a platform for information management and data science. We
have left outcome analysis of the approach and results for a future paper. Also
on the roadmap is a meaningful gamification system inspired by [