A Lightweight Neuro-Symbolic Pattern Recognition Framework for Edge Intelligence via NLP-Based Knowledge Graphs

Edge resource constraints: Edge intelligence for IoT and 6G must process massive heterogeneous data under strict latency, energy, and computational constraints.

Heavyweight edge model deployment: Deploying heavyweight deep learning models on resource-limited edge devices is challenging for pattern recognition tasks used in scheduling and decision-making.

Edge resource constraints: Edge intelligence for IoT and 6G must process massive heterogeneous data under strict latency, energy, and computational constraints.

Heavyweight edge model deployment: Deploying heavyweight deep learning models on resource-limited edge devices is challenging for pattern recognition tasks used in scheduling and decision-making.

Lightweight neuro-symbolic pattern recognition: The paper proposes a lightweight neuro-symbolic pattern recognition framework for edge intelligence that combines NLP-based entity and relation recognition with symbolic knowledge graph reasoning.

Edge knowledge graph construction: The framework constructs an edge-computing domain-specific knowledge graph using lightweight NER and relation extraction models.

BERT-BiLSTM-CRF NER: The framework uses a BERT-BiLSTM-CRF model for named entity recognition in the edge computing knowledge graph pipeline.

PCNNATT-BiLSTM relation extraction: The framework uses a PCNNATT-BiLSTM model for relation extraction in the edge computing knowledge graph pipeline.

Low-overhead KG semantic reasoning: Extracted structured knowledge forms a neuro-symbolic knowledge graph for semantic reasoning supporting resource localization, bottleneck analysis, and decision support.

NER F1 improvement:

Relation extraction F1 improvement:

Efficient edge deployment: