Business Rule Technology

Backward chaining

A style of rule processing where rules are selected for execution based on a goal-seeking strategy. Each rule breaks a goal into subgoals, or satisfies a subgoal(s). When trying to satisfy a goal, the rule engine looks for rules that conclude that goal, and then checks for any whose conditions are true. For example, a system that evaluates a loan application might have a goal to determine the risk the applicant poses. A rule such as:

	If applicant.creditScore > 675 and
	   applicant.salary > 50,000 and
	   applicant.debtLoad < bank.debtLimit
	Then applicant.creditRisk = Low

can be used if the system is currently trying to evaluate the credit risk. If the three bits of information about the applicant are already known (are in working memory), the rule immediately succeeds or fails, depending on their values. If any are not known, they are set up as subgoals. Other rules may then be able to establish whether the condition holds or not, or may be able to obtain the data in some other manner.

 

Chaining

Chaining occurs in a rule system when one rule makes a change in working memory that leads to another rule executing. While this can be used to define sequential processing (writing rules so that the change one rule makes explicitly triggers one specific other rule), that is considered bad form because it does not use the rule technology correctly. One of the advantages of a rule system is the modularity of the knowledge. When you trust the rule engine to select the best rule at each point, you avoid the complex interdependencies and difficult maintenance of a heavily nested set of IF statements in a traditional programming language.

Complex event processing

Complex event processing (CEP) is a style of rule processing where a rule’s condition specifies patterns within a stream of events, generally with a time element. It is like forward chaining, specialized for time-based reasoning. Patterns include expressions like “When an Event1 occurs within 5 min of an Event2,” or “When 3 instances of Event1 occur within a 30 min window.” These systems are often used in very fast-moving situations, like real-time securities trading. A CEP system can execute actions directly, or can recognize significant patterns that it adds as data to a more traditional working memory.

Forward chaining

The forward-chaining style of rule processing is data driven, rather than goal driven. The rules are interpreted as “When the condition becomes true (because of a change in working memory), then do ...” Take a rule such as:

	If phase = "risk scoring" and 
	   applicant.creditScore > 675 and
	   applicant.salary > 50,000 and
	   applicant.debtLoad < bank.debtLimit
	Then set applicant.creditRisk = Low and
		 set phase = "summarize application"

The inference engine applies this rule when all the clauses in the condition have become true and there is no better rule to run. Its actions place two new facts into working memory, which might trigger other rules.

RETE algorithm

In a forward-chaining system, any rule whose condition has become true is eligible to execute. There will typically be many ready to execute, and it is the job of the rule engine to select the best one to actually execute. The RETE algorithm is a method for compiling a rule base into an efficient network. When a change takes place in working memory, the network determines the right rule to execute, using the minimal number of tests, and with the maximal reuse of shared patterns. Almost every forward-chaining rule engine today uses some variant of the RETE algorithm.

Rule

A rule is a small unit of knowledge or computation that contains a condition and an action. The condition defines when the rule is applicable. The condition is defined as a pattern of working memory elements in the form of a Boolean expression. The action defines what should happen when the rule executes. Actions can add, delete, and modify working memory elements. They can also call methods that cause processing to occur outside working memory. The order in which rules execute is controlled by the rule engine.

Rule base

Sometimes also called a knowledge base. A rule base contains definitions for the data elements that can appear in the working memory. It includes the rule definitions. It might include configuration settings that govern how the rule engine works. A rule base can be seen as the program for the business decision.

Rule engine

A rule engine is technically the code that selects and executes the next rule. The term also refers to the entire application that supports creation and maintenance of rule bases, instrumentation of execution for debugging, and other functions needed for a full rule-based application. Most rule bases today are designed to support applications integrated within a traditional language such as Java. Most also have some form of stand-alone mode for rule-development purposes.

Working memory

A rule system will pay attention only to the data known to it. Working memory is exactly what it sounds like—the collection of data items that are available for examination by the rules. This data may have come into the system from outside, and also includes data created internally by rule conclusions.