Flying Into Intelligent Search
An apocryphal story about a pilot trying to land at the Seattle Tacoma Airport (SeaTac) in heavy fog holds some lessons for intelligent search. The pilot of a small plane has no visibility due to the fog. Depending on who’s telling this story, an electrical malfunction disabled the instruments, the instruments simply don’t work, or the pilot is only rated for visual flight. Thus, the pilot can’t figure out how to get the plane to SeaTac.
Then the pilot glimpses a person working in a tall building and, through a small break in the fog, holds up a sign reading “Where am I?” The person responds with a sign reading “You are in an airplane.” This immediately tells the pilot the precise coordinates so that navigation to SeaTac is easily and swiftly accomplished.
This feat of intelligence, the punchline of the story, is that the answer given by the person in the building was entirely correct but completely useless, proving that the building was Microsoft’s and the person worked in customer support. Knowing that the aircraft was at the Microsoft building, the pilot had the intelligence needed to land safely at SeaTac.
That’s one form of intelligence, although the story always makes me wonder why any pilot who wasn’t instrument rated would be attempting a flight in such bad weather. Doing so just isn’t smart. Is being smart the same as being intelligent? Practical knowledge can power street smarts that are more effective than implied intelligence. Another story comes to mind. The driver of a car listens to the voice of the GPS that says “Go straight ahead for 300 meters, then turn right.” The driver, however, knows that you can take a shortcut through a store parking lot and get to the destination faster and without stopping at the light on the corner. Some of the newer systems incorporate this travel intelligence, probably infuriating the store owner through whose parking lot vehicles are transiting.
The trick to intelligent search is to blend the various forms of knowledge to achieve the desired result in the most efficient and effective manner. Most organizations have information stored in multiple locations and in multiple formats. To create an intelligent search, these diverse pieces of information need to be connected. At Sinequa, explains Scott Parker, Senior Product Marketing Manager, the focus is on helping companies become information-driven. Ubiquitous connectivity is a big part of this. One suggestion he has is to connect information along topical lines, which both brings out collective expertise and makes it transparent. This is particularly beneficial in geographically distributed organizations, allowing employees to tap into the expert knowledge available to them so they can learn new skills with internal expert guidance.
Language can be tricky, and this is another area where Sinequa can help. Determining the language being used, analyzing the lexical construction of words, automatically extracting entity types, and text mining are key to determining meaning. Just think about how Polish people polish resumes. Knowing the difference between Polish and polish is obvious to humans but might trip up a computer. Entity extraction for concepts and names of people, places, and companies keeps apple the fruit from being confused with Apple the company. Text mining, when integrated with the indexing engine, normalizes words, terms, and phrases to see patterns.
Machine learning contributes to the information-driven process by analyzing and structuring content, modifying search results, and recommending additional content. It’s self-learning on a massive scale. Most importantly, when looking at any search and analytics platform, to be successful, it must align with end user goals. It must present a user experience that is aesthetically pleasing and understandable by employees.
Companies and Suppliers Mentioned