DECISION RULES FOR LINK, MERGE, AND UPDATE OPERATORS IN AN INCREMENTAL MULTI-AGENT FACT REPOSITORY

Abstract

Records arriving from distributed agents may be duplicates, supporting evidence, later states of an existing event, or genuinely new facts. Treating all similar records as duplicates causes information loss, while storing every record separately produces an inflated and poorly auditable repository. This paper formulates decision rules for distinguishing the Link, Merge, and Update operations within an incremental fact-repository method based on five lifecycle operators: Insert, Link, Merge, Update, and Close. The decision is made from referent identity, semantic similarity, temporal proximity, the role of the incoming evidence, and the current lifecycle status. A three-stage procedure first identifies candidate facts, then classifies the relation between the incoming fact and the candidate, and finally records the resulting transition in the repository history. A prototype demonstration on 20,000 service-request records reduced the active representation to 6,319 facts while retaining 19,544 evidential or semantic relations. The results show that repository compactness and evidence traceability can be achieved simultaneously only when Merge, Link, and Update are treated as different operations rather than a single deduplication step.