Tapping Ubiquitous Collective Wisdom
Data collected in complex environments often represents a knowledge-base of experiences of people, systems and events. Often, “what if” analysis may be desired in order to gauge the effect of certain actions – without actually implementing the actions. Example datasets include vehicle repairs (the data may represent the collective experience of many technicians), patient records (which may provide information on epidemiology), and sales campaigns (which may reflect revenues affected by a large number of factors). Each of the datasets represents a wealth of information which could be used for predictive and prescriptive purposes. In these examples, we may want to use the stored knowledge as follows: with vehicle repairs data, to help identify the problem and the recommended repairs for a new vehicle repair case; with the patient records, help in diagnosing the ailment and a guideline to prescribe for a new patient encounter; with the sales data, help with designing a new campaign to increase revenues. Unfortunately, “what if” or predictive analysis is only as good as the mathematical models used. Attempts to model such data mathematically, and thereby leverage the collective knowledge, has been of limited value due to the inherent complexity of the systems. But now –
Ubiquiti technologies provide an alternative means to utilize organizational knowledge already collected and available. Our approach is to create a knowledge repository via ETL, or extract-transform-load, of the raw data. Thereafter, for each new instance, similar cases in the knowledge repository are found rapidly (by means of a statistical similarity function), and normalized results are reported to the user. The user interface itself is kept simple, and is modeled on the familiar Web search-engines. Powerful statistical search operates behind the scenes to provide the needed information obtained from the available data. Since Ubiquiti technologies can extract information even from raw datasets which may contain data-entry errors, the knowledge repository is set up quite easily. Our technology does not require word lists to be created or maintained; similarly, there is no need to create or maintain “lessons learned” or matching criteria etc. – just the raw records are sufficient. Our approach automatically adapts and “learns” new situations since the information reflected in the data changes as new records are added or removed, and the data itself is searched to obtain the results.