What Are They Saying About Me?

Today at the The 2008 IEEE/WIC/ACM International Conference on Web Intelligence (WI-08) in Sydney, Australia, IBM Research is presenting a case study entitled "Multi-Taxonomy: An Approach to Determine Perceived Brand Characteristics from Web Data." Here's an inside perspective on how brand reputation and management is changing from IBM Researcher and guest blogger Scott Spangler.

A strong brand/company image is a critical asset to any corporation. Back in the “good old days” you could find out pretty reliable current information about how your brand image was faring by reading clippings from newspapers. Those days are long gone.

With the emergence of web and Consumer Generated Media (CGM), such as blogs, news forums, message boards, and other web sites, the number of public places where a company’s image can be discussed and influenced has exploded exponentially. How can a company possibly manage their brand image in such a fluid and dynamic environment?

Enter the next generation COBRA (Corporate Brand and Reputation Analysis) technology, an approach to mining a wide range of CGM content to discover how the social media-based community perceives a brand. The solution processes a diverse set of structured and unstructured information generated directly from CGM content. It creates order out of chaos by systematically generating multiple taxonomies from the input data. These taxonomies are then used singly and in combination to better understand important brand characteristics as they are perceived by the public. These insights can then be used to enhance marketing and strategic decision making.

In our presentation, we describe how we used our approach to help Kraft Foods find the right message to leverage the perception of their Vegemite brand on the web to design a new advertising approach that built upon the natural inherent strengths of the brand.

The beauty of this new approach is that it does not make any assumptions about the kind of discussion that is taking place or what words the customer might use to describe their feelings. Instead, it allows the data drive the analysis to conclusion, while exploiting domain knowledge wherever possible to enhance the relevance of the analysis.

Behind COBRA’s robust ability to discover, summarize, and communicate what large amounts of web content is saying about a brand lies the use of multiple taxonomies (aka perspectives) in concert. Each taxonomic generation method splits the aggregate data set up in a different way, and the combination of these different approaches leads to insight at the intersections. In other words, when we find an unusual overlap between two concepts taken from different taxonomies, that’s the clue that very often leads to a solution to the brand perception equation.

To take a specific example from Kraft we might use a taxonomic concept around breakfast food brands and another taxonomy around ingredients or attributes of the brands. This revealed a high correlation between “folate” and “vegemite,” indicating one potential way in which vegemite is seen by the consumer as significantly different from the competition.

The COBRA software automatically creates multiple taxonomies and performs thousands of comparisons between categories to determine where the interesting relationships are. This information can then be presented to the analyst in a way that is easy to absorb so that they can drill into the data to find out exactly why the relationship exists, and how it is being discussed. From there, brand insight is usually just a click away.