A knowledge network uses structure, operators and links, to represent analytic knowledge. Non-analytic knowledge is represented using operators that store probability of behaviour. The operators have their contents filled during a Learn phase, and become active in broadcasting information during the Run phase.
A distribution operator attaches to a single variable, and stores a probability distribution for the variable. The distribution operator has a Probability Control, so adjustment of the range can be performed by varying the probability. There is an initial probability, which is the result of mining the entire database. The current probability takes into account any reduction in range, on this or other related variables.
More about the distribution operator.
A relation operator attaches to two or more variables, and stores a probability map for the relation existing between those variables. A change to the range of one of the variables will cause a change in the distribution of the other variables (by removing some of the population which the distribution represents) and may lead to a change in the range of the other variables, or at least to a change in the distribution of the other variables.
More about the relation operator.
A histogram operator attaches to two or more variables, and stores the sum of values on one of its pins. It can be used, for example, for summing sales over a number of different variables, such as date, region, product. Later, the data can be sliced in various ways, including any subrange on any of the dimensions.
More about histograms
A productmap operator connects to two or more variables, acting as dimensions on the productmap hypercube. The productmap has cells, corresponding to elements on the dimensions. These cells can be empty, can contain a product object, or a list of product objects. The product objects are linkable, so various types of logic can be attached to the products. See Product Knowledge.
The combination of distribution and relation or productmap operators allows a web of experience to be mined from a database or web log and stored for operational use. An example might be the vehicle buying habits of a population, with dimensions of Age, Gender, Net Worth and Vehicle Type. Recognition of a customer's attributes along one or more axes allows greater precision in prediction of behaviour. A customer's selection of a particular vehicle type may allow us to surmise something about them. The distribution and relation operators store experience, and are embedded in a network of analytic operators. The analytic operators can cut and subset the experience, so the combination is far more powerful than either acting alone.