business intelligence and analytics: from big data to big impact

Chen et al./Introduction: Business Intelligence Researchservice (SaaS), infrastructure as a service (IaaS), or platformas a service (PaaS). Only a few leading IT vendors are currentlypositioned to support high-end, high-throughput BI&Aapplications using cloud computing. For example, AmazonElastic Compute Cloud (EC2) enables users to rent virtualcomputers on which to run their own computer applications.Its Simple Storage Service (S3) provides online storage webservice. Google App Engine provides a platform for developingand hosting Java or Python-based web applications.Google Bigtable is used for backend data storage. Microsoft’sWindows Azure platform provides cloud services such asSQL Azure and SharePoint, and allows .Net frameworkapplications to run on the platform. The industry-led web andcloud services offer unique data collection, processing, andanalytics challenges for BI&A researchers.In academia, current web analytics related research encompassessocial search and mining, reputation systems, socialmedia analytics, and web visualization. In addition, webbasedauctions, Internet monetization, social marketing, andweb privacy/security are some of the promising researchdirections related to web analytics. Many of these emergingresearch areas may rely on advances in social network analysis,text analytics, and even economics modeling research.Network AnalyticsNetwork analytics is a nascent research area that has evolvedfrom the earlier citation-based bibliometric analysis to includenew computational models for online community and socialnetwork analysis. Grounded in bibliometric analysis, citationnetworks and coauthorship networks have long been adoptedto examine scientific impact and knowledge diffusion. Theh-index is a good example of a citation metric that aims tomeasure the productivity and impact of the published work ofa scientist or scholar (Hirsch 2005). Since the early 2000s,network science has begun to advance rapidly with contributionsfrom sociologists, mathematicians, and computerscientists. Various social network theories, network metrics,topology, and mathematical models have been developed thathelp understand network properties and relationships (e.g.,centrality, betweenness, cliques, paths; ties, structural holes,structural balance; random network, small-world network,scale-free network) (Barabási 2003; Watts 2003).Recent network analytics research has focused on areas suchas link mining and community detection. In link mining, oneseeks to discover or predict links between nodes of a network.Within a network, nodes may represent customers, end users,products and/or services, and the links between nodes mayrepresent social relationships, collaboration, e-mail exchanges,or product adoptions. One can conduct link mining usingonly the topology information (Liben-Nowell and Kleinberg2007). Techniques such as common neighbors, Jaccard’scoefficient, Adamic Adar measure, and Katz measure arepopular for predicting missing or future links. The linkmining accuracy can be further improved when the node andlink attributes are considered. Community detection is also anactive research area of relevance to BI&A (Fortunato 2010).By representing networks as graphs, one can apply graphpartitioning algorithms to find a minimal cut to obtain densesubgraphs representing user communities.Many techniques have been developed to help study thedynamic nature of social networks. For example, agent-basedmodels (sometimes referred to as multi-agent systems) havebeen used to study disease contact networks and criminal orterrorist networks (National Research Council 2008). Suchmodels simulate the actions and interactions of autonomousagents (of either individual or collective entities such asorganizations or groups) with the intent of assessing theireffects on the system as a whole. Social influence and informationdiffusion models are also viable techniques forstudying evolving networks. Some research is particularlyrelevant to opinion and information dynamics of a society.Such dynamics hold many qualitative similarities to diseaseinfections (Bettencourt et al. 2006). Another networkanalytics technique that has drawn attention in recent years isthe use of exponential random graph models (Frank andStrauss 1986; Robins et al. 2007). ERGMs are a family ofstatistical models for analyzing data about social and othernetworks. To support statistical inference on the processesinfluencing the formation of network structure, ERGMsconsider the set of all possible alternative networks weightedon their similarity to an observed network. In addition tostudying traditional friendship or disease networks, ERGMsare promising for understanding the underlying networkpropertities that cause the formation and evolution ofcustomer, citizen, or patient networks for BI&A.Most of the abovementioned network analytics techniques arenot part of the existing commercial BI&A platforms. Significantopen-source development efforts are underway from thesocial network analysis community. Tools such as UCINet(Borgatti et al. 2002) and Pajek (Batagelj and Mrvar 1998)have been developed and are widely used for large-scalenetwork analysis and visualization. New network analyticstools such as ERGM have also been made available to theacademic community (Hunter et al. 2008). Online virtualcommunities, criminal and terrorist networks, social andpolitical networks, and trust and reputation networks are someof the promising new applications for network analytics.MIS Quarterly Vol. 36 No. 4/December 2012 13