Knowledge Representation and Reasoning

Knowledge Representation and Reasoning (KR&R) encompasses the design of formal models and inference mechanisms to capture, organise and derive new insights from structured, semi-structured or uncertain information. Over the past decades, KR&R has progressed from symbolic logics and semantic networks to probabilistic and evidential frameworks, rule-based expert systems and expressive ontologies. Its aim is to reconcile expressivity—enabling rich conceptual hierarchies, exceptions and contextual nuances—with computational tractability. Core KR&R technologies include logic programmes, belief functions, rule bases and knowledge graphs, each paired with reasoning engines that perform deductive, non-monotonic or hybrid inference. Practical applications range from multi-sensor data fusion and environmental monitoring under uncertainty to semantic data integration for large-scale repositories and transparent decision support. Recent research has emphasised scalable conflict resolution in evidence aggregation, semantically coherent integration of heterogeneous sources and the fusion of symbolic and statistical methods to support adaptive, interpretable reasoning.

Research from Nature Portfolio

Recent studies have advanced evidence-theory models by proposing a novel basic probability assignment generation method that uses Mahalanobis distance, cosine similarity and belief entropy to resolve conflicting data, yielding more rational fusion outcomes and faster convergence. Parallel work has introduced an air-quality evaluation model that integrates evidence weighting and decision credibility into an improved Dempster–Shafer framework, demonstrating enhanced reliability and uncertainty expression when processing hourly pollution measurements. Most recently, an interpretable belief rule-base approach incorporating constrained reference value optimisation and projection-based evolutionary strategies has been developed for complex system health-state assessment, preserving expert semantics while boosting inference accuracy under uncertainty.

Research from all publishers

A hybrid query-answering paradigm combines a lightweight Datalog reasoner with a full-featured ontology engine in a pay-as-you-go strategy, delegating expensive inference only when necessary to ensure scalable completeness over ontological data. An ontology-driven integration system for personal data lakes has achieved automated schema matching and metadata management, retaining native data formats while supporting dynamic, unified querying across heterogeneous sources. In the biomedical domain, transformer-based contextual embeddings have been fine-tuned on ontology descriptions to predict candidate concept alignments, followed by structural extension and logic-based repair, resulting in superior alignment accuracy under both unsupervised and semi-supervised settings.

Knowledge Representation and Reasoning publication trend

The graph below shows the total number of articles in knowledge representation and reasoning across all publications each year (not limited to Nature Index journals).

Technical terms

Basic probability assignment (BPA): A function in Dempster–Shafer theory that allocates support mass to subsets of hypotheses, quantifying evidence for each subset.

Evidential reasoning: An aggregation methodology that combines multiple belief distributions or rule outputs into a unified belief structure under uncertainty.

Belief rule base (BRB): A knowledge-representation scheme using weighted rules with belief degrees and evidential transformation for inference from imprecise inputs.

Ontology: A formal, machine-interpretable specification of domain concepts, properties and relationships, supporting semantic interoperability.

Datalog: A rule-based query language used for efficient logical inference and query rewriting over relational and graph data sources.

Pay-as-you-go query answering: A hybrid reasoning approach that performs lightweight inference by default and escalates to full ontology reasoning only as required.

References

1. A new basic probability assignment generation and combination method for conflict data fusion in the evidence theory. Scientific Reports (2023).
2. An ambient air quality evaluation model based on improved evidence theory. Scientific Reports (2022).
3. A complex system health state assessment method with reference value optimization for interpretable BRB. Scientific Reports (2024).
4. PAGOdA: Pay-As-You-Go Ontology Query Answering Using a Datalog Reasoner. Journal of Artificial Intelligence Research (2015).
5. SemLinker: automating big data integration for casual users. Journal of Big Data (2018).
6. BERTMap: A BERT-Based Ontology Alignment System. Proceedings of the AAAI Conference on Artificial Intelligence (2022).