Incremental Inference for Semantic Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Maintaining and updating large semantic data models with new triples is inefficient due to the need to re-apply inference rules across the entire corpus, leading to significant processing time, especially when adding new triples to datasets like LUBM8000 which contains over a billion triples.
Innovation Solution
Implementing incremental inference logic that fires inference rules in multiple join steps, selecting from new triples while applying rules to a union of old and new triples, and removing duplicates after each inference round to efficiently update the semantic model without re-deriving existing inferred triples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If inference rules are applied to the entire corpus of triples including new triples, then the semantic model is updated with newly inferred triples, but the processing time becomes extremely long
Solution Approach 1:
The patent segments the inference process into two distinct phases: (1) computing newly inferred triples by applying inference rules only to the union of old and new triples, and (2) updating the semantic model by combining old inferred triples with newly inferred triples. This segmentation allows the system to avoid re-applying inference rules to the entire corpus, significantly reducing processing time while maintaining completeness of the entailment update.
Solution Approach 2:
The patent performs preliminary computation of newly inferred triples before the final model update. By pre-computing the set of newly inferred triples from the union of old and new triples, the system prepares the necessary information in advance, allowing the subsequent update operation to be performed efficiently without re-processing the entire corpus.
2Manufacturing precision
If inference rules are re-applied to the entire corpus, then all inferred triples are re-derived, but existing inferred triples are re-derived unnecessarily
Solution Approach 1:
The patent extracts and isolates only the newly inferred triples that result from adding new triples to the semantic model. By computing the difference between the inferred triples from the union of old and new triples and the original inferred triples, the system extracts only the necessary new information, avoiding redundant re-derivation of existing inferred triples while maintaining derivation accuracy.
Solution Approach 2:
The patent applies partial action by computing inference rules only on the necessary subset of triples (the union of old and new triples) rather than the entire corpus. This partial application of inference rules is sufficient to identify all newly inferred triples, achieving the required accuracy without the excessive computation of re-processing all old triples.
3Loss of information
If the semantic model is updated with new triples, then the data model reflects current information, but the processing overhead increases significantly
Solution Approach 1:
The patent segments the update process into computing newly inferred triples and then merging them with the existing semantic model. This segmentation ensures that information freshness is maintained by incorporating all necessary new inferences, while processing overhead is reduced by avoiding redundant computation of triples that have not changed.
Data Source
AI summary
Systems, methods, and other embodiments associated with incremental inference are described. One example method includes updating existing or old triples in a semantic model with triples resulting from the addition of new triples. The updating is performed by separating inference rules into joining steps that are performed on first and second predicates for the inference rule. A first joining step joins results of execution of the first predicate on the new triples with the results of execution the second predicate on the union of the old and new triples to produce newly inferred triples. A second joining step joins results of execution of the first predicate on the union of the old and new triples with the results of execution the second predicate on the new triples to produce newly inferred triples.