Analytics Adds Intelligence to Search
We are well aware of the productive gains with faster search. The new frontier is to improve the quality and insights from search, achieved from integration of search with analytics.
Enterprise solutions are well practiced at extracting information from structured data. Search is designed to aggregate unstructured information from keywords and a wide variety of sources like emails, social blogs, etc. Merging structured stores with unstructured text derives meaningful, relevant summarizations as well as new insights. Analytics dissolves the barrier, providing integration founded on the data content itself—driving more complete and accurate intelligence for business users.
Analytics At Work
For instance, a sales executive sees the sales performance of "golf clubs" in the northeast region. When comparing the northeast golf club sales in the current quarter to those in the previous year, he sees that the sales have dropped by 20%. Now the sales executive wants to know why? He tries to analyze the information, searching with repeated queries, guessing with key words and topics, trying to extract the answer. He drills into the report, which only confirms the numeric calculation—but doesn't extend to describe contributing factors. Frustrated with not knowing what has happened in this region, he is left without an immediate answer.
The issue lies with the information presented to him, not with his approach. He is playing a corporate game of "go fish"—querying the system to find the one card containing the information he seeks—without knowing what is in the deck he is playing with. How could he know this?
Even if he had the complete list of terms that have been indexed to the system describing sales activity, competition and market indicators, he would still need to guess which of those keywords, or in what combination, would answer his specific question.
Utilizing data from all available sources (both structured and unstructured), and analyzing it, would have led him to know that competitor stores within the northeast region had lower-than-wholesale prices on a new club design to promote registration to their online network—for a limited time. They are not hearing the social noise, and while formal feedback comes through customer service, it hasn't yet reached our executive because it is in silos and not linked to the event context.
Having an analytically integrated environment in this case means integrating the reporting systems with the information from customer service, social networks, emails, competitive flyers and documents, with the content categorized (to competitor activity, region, golf clubs, online community), key facts extracted (limited-time sale at lower-than-wholesale prices). These content-derived entities, concepts and facts would be encapsulated in taxonomies and linguistic rules.
Instead of guessing what might be going on, our executive would now know. The query "What changed in golf clubs for the northeast last quarter?" could be asked. With analytics, a date calculation is made, to isolate the relevant time-stamped data. The content within that time span would be analyzed to identify the event. The output from the content categorization and the definitions from extraction-for-facts is input to search and indexing. Before adding a document to the search engine's index, the generated metadata to identify categories, concepts and facts is executed—defining facets for search. Our executive would receive the answer, with the issue of complex syntaxes or keyword guesses removed.
With analytics integrated with search, one solution would provide the relevant information without a need for separate steps of analysis and search for answers. The information available to the organization can flexibly be examined and explored from any number of perspectives—depending on the richness of the taxonomy and linguistic rules defined.
More Intelligent Search
The linkage between taxonomies and stores of data with rules describing relevance under specific scenarios can be managed. This taxonomy management can automatically map documents with one or more topics by looking at the context to understand the meaning and find or extract these from the text. As a result, related topics can be presented and explored, driving new discoveries.
Taken further, the combination of text analytics with text and data mining is a special area of intelligent applications. By adding structure to unstructured content, it is possible to integrate that content with traditional advanced analysis capabilities to further expand the coverage and usefulness of the models. Mining examines a document or data collection, as opposed to assessing a document at a time. Within a mining context, the technology identifies parts of speech, synonyms and naturally occurring topics—automatically. Such insight is useful to inform taxonomy definitions and sentiment rules and to identify new, emerging topics that occur.
What users are clamoring for is an intelligent system. Analytics-derived metadata makes frequently executed searches much faster and more relevant. The success of an enterprise search deployment relies on the automatic creation of metadata.
Analytics can extend business activity continuums. The intelligence from text and structured data can be embedded in rules, threshold processing, case prioritization and automatic delivery and notifications, and more—leveraging existing infrastructure investments.
The solution for any particular organization may involve a mix of search, analytics, complex event processing and rule-based/algorithmic inference engines. The end goal is the same: help business managers have a holistic picture of a business problem so that they can make the right decisions, with confidence.
Driven to solve customer business problems, privately held SAS works closely with customers in all stages of research and development to ensure that they get the most out of new offerings. More than 12,000 employees across more than 400 SAS offices provide local, hands-on support. Lean more about SAS at www.sas.com.