[December 29, 2004]

Extracting Meaning From Content

By Bjørn Olstad, Chief Technology Officer, and Lee Phillips, Director of Intelligence Solutions, FAST

Over the past four years, the Internet has changed many tried-and-true business propositions. From ‘category killers’ like Amazon, to new paradigms like eBay and Priceline, to new-media giants Google and Yahoo!, this new medium continues to create new messages about business. Its central impact involves more than removing ‘the middle’ of transactions; it has literally created new forums and new forms of speech -- both commercial and non-commercial. With cyclonic force, it has become the most common publishing platform on the planet. In its wake, it has left conventional thinking and traditional processes devastated.

The changing stakes of an increasingly Web-centric world are clearly evidenced by recent evolutional developments in the market for business information -- the raw material of the gathering/analyzing/communicating paradigm commonly referred to as ‘intelligence.’

Historically, the market for information was driven by a combination of two factors: 1) the intrinsic worth of content and 2) managed, exclusive access. The economic model of publishers embodied this value proposition -- information is the critical fuel for decisions; better fuel produces better decisions; therefore premium information (better, more exclusive fuel) has intrinsically greater value. Previously, publishers evolved new media, or forms of distribution gradually -- carefully managing the balance between obsolescence and innovation.

What’s going on?

But now the value of content is changing. With increasing volumes to consider, it isn’t just about content anymore. Content-analysis technologies making their way into the marketplace will significantly reshape the processes of gathering, analyzing and communicating the intelligence that drives decisions. These new procedures for analyzing unstructured content allow companies to bring structure and standard reporting techniques to both the constant flow and their deep archives of unstructured information.

The forces driving changes in the value of information are well known: volume, and the continuing drive to deploy low-latency, highly-responsive processes in every organization. These forces have created a persistent drumbeat for what some refer to as the ‘real-time organization.’ Accelerating the decision process in the face of overwhelming increases in information volume creates a genuine difficulty -- how to interpret more information in less time. This difficulty is complicated by a third force -- a perception of increasing uncertainty. Whether viewed from the perspective of the organization, the market or industry, or global conditions, the perceived risk of overlooking meaningful information is clearly increasing, reflecting greater risks from geopolitics and increasingly information-intensive industries.

What does it mean?

In most practical applications, interpreting meaning is complicated by the variability found in commonly held senses of meaning. Variability is inherent in complex concepts with multiple ‘dimensions’ of meaning -- for example, brand value has at least two dimensions of meaning (customer satisfaction and willingness to recommend). Multiple perspectives also create variability -- from the differences in meaning, or connotation, present in any group of individuals. For example, your concept of threat may reflect a stronger sense of scale (massive versus personal) than mine -- which may reflect greater sensitivity to immediacy.

Methods and procedures that do not account for, nor explicitly treat, the sources of variation will be less efficient (statistically speaking, show greater variance) and may well be biased (statistically, inevitably ‘off-target’). Methods that incorporate these sources of variation into the interpretation process will also be capable of addressing important operational questions, such as “How strong is the consensus opinion of a team of analysts?” or “Among a group of possible meanings, which is best supported by the evidence?” In short, ignoring these sources of variation increases the risks of oversimplification, lack of imagination and substituting collective for collaborative thinking

Among the methods available for interpreting meaning from unstructured text, one with published, benchmarked performance and the flexibility to analyze meaning from any number of perspectives is a technique termed ‘semantic orientation’ -- recently summarized by Turney and Littman in their research paper, “Measuring Praise and Criticism: Inference of Semantic Orientation from Association.”

Recently, a large European services provider employed this strategy to activate the power of customer loyalty. Their first insight was to distinguish two sources of loyalty variation -- customer satisfaction and willingness to recommend. Next, using a ‘semantic orientation’ technique recently summarized by Turney and Littman, they created a scoring engine to identify extremes of satisfaction and recommendation. Integrated into an enterprise search and alerting platform, they analyze volumes of text feedback from sources heretofore ignored. The results -- better monitoring (‘brand dashboards’) and timely new dialog (‘customer early warning’) -- help convert potential brand energy (satisfaction) into kinetic brand energy (recommendation).

Interpreting meaning adds value

Tremendous increases in the flow of unstructured information have left many organizations and their intelligence processes ill-prepared for both threat and opportunity. Without a means to rapidly locate data and assess and communicate its meaning, the risk of missing something really important increases substantially. Companies need solutions that can accurately extract meaning and trends from unstructured data sources. The underlying technology for interpreting meaning has applications far beyond market intelligence. These solutions, which are powered by search technology, are just beginning to enter the market and offer companies the ability to add structure to the mountains of information that are building up on corporate servers all over the world.