Research And Application Of Massive Rule-based Reasoning Sys- Tem By Extended RETE Algorithm

The production system, as is easy for understanding and modification, is widely used in Artificial Intelligence. However, due to its bad matching efficiency,it’s not practicable in large amount of data. RETE algorithm, which speed up the match efficiency by sharing condition elements and reserving temporary matching results, has become one of the most widely used reasoning algorithms for production systems. However, as the widely emerging of massive, rapidly changing or imperfect data as well as fuzzy logics, the production reasoning algorithms, typically for the RETE algorithm, are now facing considerable challenges.To efficiently solve the problems of massive data and logics that the traditional production system faces, this paper chooses a typical reasoning algorithm, i.e. RETE algorithm,giving a survey about the principles, state of arts and problems of massive data it faces. Then proposes a new rule-based reasoning engine in which two key technologies are discussed:1. Approximate RETE Algorithm:Introduced the tendency and feasibility of ap-proximate method. Proposed an approximate RETE algorithms from the angles of fact propagation. The approximate algorithm significantly improves the effi-ciency of reasoning,making them quite applicative for massive data reasoning.2. Parallel RETE Algorithm:Summarized the tendency and current issues of parallel reasoning, proposed a sub-rule splitting method, which splits rules into sub-rules, and implicated the parallel algorithm in the MapReduce architecture.Finally, we realized the proposed rule-based reasoning engine,respectively apply-ing it into the self-adaptive middleware framework and the on-demand massive data mining platform. Both experiments and applications have verified the efficiency of the intelligent reasoning framework.