Enabling understanding of markets and customers

Dr. Gautam Shroff, Head, TCS Innovation Labs - Delhi

Unearthing hidden patterns in data can help to discover knowledge that can improve business strategy or optimize operational processes. Traditional applications of knowledge discovery in data focus on tasks such as:

• Segment customers, suppliers, markets, or employees to
  identify behaviour which can help in planning retain and
  reward schemes

• Selective targeting of advertising and marketing campaigns by determining what information needs to be pushed to which consumers, based on, for example, similarities in buying patterns.

• Detect anomalies and other rare events, such as credit card and insurance claim fraud, illegal duplication of mobile phone SIMs, and even terrorist activities.
  

• Identify problems and opinions, using data such as from customer feedback, blogs, and emails. Assess overall situations and trends by fusing evidence from multiple sources to determine high level explanations from large volumes of ground level evidence.

Our research activities focus on understanding customers and markets through knowledge discovery tasks. In this context we are focusing on two main directions of research

• Estimating perceptions of customers based on sentiments and opinions about brands or specific problems faced by customers
  

• Detecting adversarial behaviour in the context of intelligence, law enforcement, and defence applications as well as fraud detection in other industries such as telecom or financial services.

Estimating and Tracking Perceptions

In today’s online world it is increasingly possible to solicit and receive direct customer feedback about product’s services. Further, conversations in forums such as blogs and online discussion forums can yield insight into the brand perception of a company or its products. However, traditional business intelligence tools and predictive analytics techniques deal only with structured data. Traditional analytics is unable to derive much value from unstructured inputs such as the large volumes of customer feedback that is available due to enhanced connectivity and increased socialization through Web 2.0. This year we have developed OPTRA, a Causal Analytics Platform that analyses customer feedback to capture Opinions, Problems and Trends, so as to determine future Requirements that can be Actions. Using OPTRA, the Voice of the Customer can discerned from large volumes of feedback, yielding information on what customers are saying about which products and why. We have successfully deployed the OPTRA platform to analyze large volumes of customer feedback from telecom equipment manufacturers, to regularly provide crisp summaries that can be worked on to improve services or products. We have also incorporated our text-analytics technology in TCS’ Listening platform, which seeks to derive brand perception from online Web 2.0 forums and discussion boards.

Another focus area within perception analysis is Broadcast Analytics, where the goal is to track news stories to discern how specific events are perceived differently around the world. Our research seeks to analyze TV, radio, print and electronic media in an integrated manner. Apart from media and government agencies, broadcast analytics holds promise for marketing and corporate communication department’s of large organizations.

Detecting Anomalous Behaviour

Data available on online social networks, through email exchanges in an organization, telephone call data records, or financial transactions is accessible publicly or to agencies that need them; making it open to automated analysis through visual link analysis tools. However, such analysis is time consuming and effort intensive. We have developed a platform, called iLinks, to facilitate Semi-automated Visual Link Analysis over very large social networks, and have currently applied the technology to real-life data for law enforcement and related applications. Our tool enhances the state-of-art in that it enables automatic analysis of patterns as well as automatic detection of outlier notes, whereas most commercial as well as research tool for link analysis focus entirely enabling on visual detection of patterns or suspicious behaviour by the human analyst.

Future outlook

Our current research efforts are focusing on estimating perceptions, be they about products or news events, as well as detecting events themselves (especially rare events). With the ever increasing amount of data being generated online, as well as the increasing digitization of human transactions from the financial sphere to communications, the volume of events being generated is growing exponentially. Searching for patterns in large-scale event-streams is therefore another capability that we plan to incorporate into our tools in the future. With this in mind we have already initiated exploratory work in the emerging areas of information fusion, event-stream processing and complexevent processing, wherein there have been significant advances in academic research in the past few years, along with some commercial implementations.