Data Driven Backward Chaining

Data-Driven Backward Chaining White Paper

With highly technical audiences, questions arise regarding a rule engine's logic capabilities. Is the rules engine support inference logic? Is the engine based on the Rete Algorithm? Does it support both forward and backward chaining? Ultimately, these questions seek to determine the performance, depth and breadth of a rules engine's logic capability.

Haley Systems' family of rules engines, provide the fastest, most scalable, and most functional business rules engines in the industry since they support an optimized version of the Rete Algorithm with both forward and backward chaining.

Backward chaining is a method of optimizing performance by identifying goals and reducing the number of deductions required with forward-chaining only systems. For further information on backward chaining, refer to the abstract below for the whitepaper available for download listed on this page.

Abstract

CLIPS cannot effectively perform sound and complete logical inference in most real-world contexts. The problem facing CLIPS is its lack of goal generation. Without automatic goal generation and maintenance, forward chaining can only deduce all instances of a relationship. Backward chaining, which requires goal generation, allows deduction of only that subset of what is logically true which is also relevant to ongoing problem solving.

Goal generation can be mimicked in simple cases using forward chaining. However, such mimicry requires manual coding of additional rules which can assert an inadequate goal representation for every condition in every rule that can have corresponding facts derived by backward chaining. In general, for N rules with an average of M conditions per rule the number of goal generation rules required is on the order of N*M. This is clearly intractable from a program maintenance perspective.

We describe the support in Eclipse for backward chaining which automatically asserts goals as it checks rule conditions. Important characteristics of this extension are that it does not assert goals which cannot match any rule conditions, that two equivalent goals are never asserted, and that goals persist as long as, but no longer than, they remain relevant.

Note:
The document referenced here was published in the Proceedings of the Second Annual CLIPS Conference, NASA Johnson Space Center, Houston TX, September 1991, and written:

• While Microsoft Windows 3.x and DOS were popular,
• Before NASA discontinued funding to support or enhance CLIPS,
• Before Haley Systems (f.k.a. The Haley ) introduced HaleyRules

Nonetheless, the basic truths of rule-based programming and expert systems in this white paper remain unchanged.