Putting intelligent search to work
The progression of enterprise search is one of the most profound developments of the contemporary data ecosystem. While expanding to utilize almost every dimension of cognitive computing, it reinforces some of the most pivotal aspects of knowledge management, business intelligence, and real-time querying.
In the process, search has moved beyond a heavy reliance on mere keywords to encompass something much more: seamless (and seemingly effortless) natural language interactions between enterprise users and their data.
According to Qlik CTO Mike Potter, “The transition from a query to a question is really the evolution of search in terms of your ability to get the most out of the data itself. It’s more than just questions and answers: It’s more about a conversation.” Conversational engagements with data systems via cognitive search rely on several aspects of AI, including these:
♦ Conversational AI: Formerly known as natural language interaction, conversational AI enables users to ask questions in natural language and get responses the same way.
♦ Symbolic reasoning: This rules-based form of AI’s traditional knowledgebase relies heavily on taxonomies, knowledge graphs, and fundamental understanding of business concepts.
♦ Machine learning: Often employed with deep neural networks, various machine learning approaches—including both supervised and unsupervised learning—are essential for analyzing data to produce rapid search responses.
♦ Neuro-symbolic AI: The fusion of AI’s statistical or machine learning foundation with its knowledge or symbolic reasoning side overcomes the shortfalls of each respective approach to optimize search capabilities.
AI enables search to surmount its basic keyword limitations to become a nuanced form of simplistic linguistic interactions,and the overarching utility provided by search itself has also burgeoned into more than previously offered, even a short while ago.
“A lot of search vendors over the years used to talk about search as ‘finding a needle in a haystack,’” said John Rueter, head of marketing at Katana Graph. “Using predictive AI and knowledge graphs, it’s not about finding a needle in the haystack, it’s actually telling you there’s a needle in the haystack. Graph and AI can be predictive in not only discovering something but actually quantifying it, telling you about it, and measuring it with a real precision and accuracy.”
Multiple search dimensions
The close relationship between search and AI is largely responsible for the latter’s propulsion of the former. According to Ryan Welsh, CEO of Kyndi, “It’s not search, then AI search. Search is an AI task. It’s always been an artificial intelligence task dating back to the early days of artificial intelligence.” The abundance of AI currently at an enterprise’s disposal has broadened the usefulness of search beyond what was traditionally a keyword approach that eventually encompassed thesauri. Today, the numerous dimensions of search include exact or approximate matches for results, as well as ranking results when multiple results are returned.
As previously noted, the most prominent dimensions are conversational and cognitive search. “Conversational means you either type in a natural language query or you talk to the computer like you talk to [Amazon] Alexa and you get answers,” said Keshav Pingali, CEO and co-founder of Katana Graph. “Cognitive is basically about using knowledge about the user and the word to return better results.” A basic example of cognitive search is giving a user weather results for his or her location, without the user specifying that place when inquiring about the weather.