Telecommunication networks face significant challenges in performing root cause analysis due to their complexity, data heterogeneity, and high-volume event streams. This paper presents a comparative evaluation of the Semanticenhanced Programmable Graph (SPG) paradigm, implemented via the OpenSPG engine, which integrates the structural advantages of property graphs with semantic constraints inspired by RDF ontologies. Through a practical telecom use case, we demonstrate SPG's strengths in event-centric modeling and causal reasoning. However, our analysis reveals notable trade-ofs, particularly in semantic interoperability, reasoning capabilities, and tooling maturity compared to established RDF standards. We conclude that while SPG is a promising model for performance-driven, internal domains like telecom operations, its current limitations reduce the universality that is a hallmark of semantic web solutions. Our findings ofer critical insights into the applicability of OpenSPG and highlight key considerations for developing future semantic-enabled network management solutions.
In complex systems such as telecommunications networks, maintaining operational reliability requires not only detecting issues early but also understanding their underlying causes. Root cause analysis (RCA) in the telecom domain faces several challenges due to the scale, complexity, and heterogeneity of modern network infrastructures1. Telecom networks generate significant volumes of event data from diverse and distributed sources, often in inconsistent formats and without unified semantics. Events are highly interdependent, with cascading efects across diferent layers (e.g., management, control, and data), making it dificult to isolate the origin of faults. Correlating this data across diferent sources, understanding service and customer impact, and preserving contextual information such as interface roles or link functions are particularly dificult. Additionally, the dynamic and real-time nature of network operations requires RCA methods that are not only accurate but also scalable and responsive. Integrating domain knowledge and semantic context remains a key challenge for building explainable and automated RCA systems in telecom. Addressing these limitations demands graph-based approaches that can efectively represent event dependencies and domain knowledge.
In the domain of graph-based knowledge representation, two primary models are RDF [
ceur-ws.org
network data into a knowledge graph using shared ontologies, validated through Edge-Cloud analytics
use cases. The NORIA-O model [
RDF enables reasoning and interoperability through subject–predicate–object triples and ontologies,
while LPG provides flexibility and scalability with key-value properties on nodes and edges. However,
RDF’s query ineficiencies [
OpenSPG is a knowledge graph engine based on the SPG approach to improve industrial applicability
[
As both RDF and SPG ofer distinct modeling philosophies and capabilities, we compare them along
several criteria relevant to building explainable and performant root cause analysis systems in telecom
(see Table 1). From this comparison as well as a proof of concept implementation for a preliminary
router operations use case given in the previous paper [
OpenSPG aims to support rapid development of enterprise-level knowledge systems by modeling the
full knowledge lifecycle [
OpenSPG emphasizes event-centric domain modeling, using context-specific and action-oriented relationships to represent entities and their interactions in dynamic systems such as networks. The relationships in OpenSPG engine are more focused on practical domain modeling rather than formal 2https://github.com/OpenSPG 3https://github.com/beyzayaman/openspg logical reasoning. They are designed to capture the semantic meaning of entities and events in specific contexts (such as supply chains, business processes, or network systems). OWL relationships (properties) are defined within a formal ontology and support logical reasoning. It provides a more rigorous logical framework for relationships, including subclassing, inverse properties, and cardinality constraints. OpenSPG engine relationships are often more action-oriented (e.g., “locateAt”, “mannerOf”) and tailored for real-world systems where event-driven analysis is key. Both support inheritance (is_a in OpenSPG engine, rdfs:subClassOf in RDFS/OWL) and enforce domain and range constraints for semantic consistency.
Our use of SPG’s built-in schema and property constraints (e.g., enforcing node types and expected properties) helped uncover obvious data issues like missing identifiers or incorrect links. However, these validations are relatively shallow compared to what RDF technologies like SHACL can provide. SHACL allows declarative, expressive constraints (e.g., cardinality, value ranges, conditional constraints) and supports reusable constraint templates. Thus, currently OpenSPG lacks the mature data validation capabilities of RDF+SHACL, which are crucial in high-integrity domains like telecom.
In terms of uplift of data, although OpenSPG supports loading data from relational tables (MySQL), the alignment of column names to schema properties was found to be non-trivial in our use case. Schema mismatches or missing property definitions led to runtime errors or silent omissions in graph construction. Successful data ingestion required iterative tuning of schema-property alignment and index strategies.
On the other hand, limited interoperability complicates integration with external linked data sources and the reuse of public RDF datasets, primarily due to the absence of shared vocabularies and standard query protocols like SPARQL. While this constraint is often acceptable in closed, performance-oriented environments, it sacrifices the cross-system compatibility central to the semantic web vision.
In the implementation of our use case, we generated sample instances for each class and loaded them into OpenSPG. A total of 9 classes (including events, entities, and concept types) and 35 relations were defined for the use case. Among these relations, 4 were logically inferred rather than physically created, meaning they were derived through reasoning instead of being explicitly stored in the data. Due to the built-in vectorization for natural language querying, the data insertion into Neo4j was time-consuming (e.g., 15 minutes for 40KB on an 8GB RAM system). Additionally, query performance degraded significantly when primary keys were omitted, highlighting some current limitations in OpenSPG’s query optimization. Overall, despite our initial findings, our experience of using OpenSPG for our network scenario holds promise, and further exploration will be undertaken as the engine and supporting documentation evolves and matures. Despite its current immaturity, OpenSPG ofers value for projects already based on property graphs seeking enhanced semantic capabilities.
Feature SPG-Schema Ontology-based Schema Schema Definition Hybrid model: Combines elements of RDF schemas Strict ontology-driven: Uses RDF Schema (RDFS) and and Property Graphs with controlled semantics. OWL for defining classes, properties, and constraints.
Data Model Property Graph with semantic enhancements (sup- Triple-based (subject-predicate-object) with strict onports labels, relationships, and typed attributes). tological constraints.
Reasoning & Inference Supports programmable reasoning (SPG-Reasoner) Strong reasoning using RDFS entailment, OWL DL, using KGDSL, enabling custom rules and hybrid AI- and SPARQL inferencing. Supports logical inference symbolic reasoning. but can be computationally expensive.
Query Language KGDSL (Knowledge Graph DSL) + Property Graph SPARQL (standard for RDF queries).
queries (Cypher, Gremlin).
Challenges - More complex than traditional Property Graphs. - - Performance bottlenecks when reasoning. - Difi
Steep learning curve due to hybrid schema. cult to manage for fast-changing data.
Flexibility vs. Control Balanced: Provides schema control, allows extensions. Highly controlled: Strict ontology-based constraints. Expressiveness Supports semantic constraints but retains flexibility High expressiveness due to OWL-based reasoning, for graph traversal. with complex class hierarchies, rules, and inference. This research received funding from Research Ireland [grant number 24/IRDIF/13114] and ADAPT Researchers are partially funded by the European Regional Development Fund through the ADAPT Centre for Digital Content Technology [grant number 13/RC/2106]. We appreciate the helpful discussions and collaboration from Anatolii Parerva, Bunyarit Puangthamawathanakun, Francesco Landolfi, and Shengyang Chen.
During the preparation of this work, the authors used Chat-GPT-4 in order to: Grammar and spelling check. After using these tool(s)/service(s), the authors reviewed and edited the content as needed and take full responsibility for the publication’s content.