Consistency publishing and retrieval method of isomorphic vector and lexical index

Abstract

The present application relates to information retrieval index updating technology, and discloses a consistency publishing and retrieval method of heterogeneous vector and lexical index. According to the knowledge base, path and source content abstract, a revision token is generated, and the sub-blocks carrying the same token are temporarily stored in the vector and the lexical index; the number and abstract proof are read back and generated respectively. When the numbers of both ends are equal and greater than zero, and the abstracts are the same, the expected publishing state is compared, the publishing list is atomically updated in the relational database transaction, and the old revision retention period is recorded; when the first insertion conflict occurs, the coverage is refused. The retrieval request filters the double channel with the same complete list snapshot and fuses the candidate, and when the coverage is incomplete, a retryable state is returned. After the retention period, the old revision is re-verified and not restored to the current revision, and the double-end record is cleaned up. The method is used for heterogeneous indexes that cannot share storage transactions, reduces the mixing and concurrent coverage of new and old revisions, and supports old snapshot reading within the set retention window.

Images