Journal of ICT Standardization

Study on Phrase Processing of English Texts by Neural Machine Translation

2025-11-18T15:18:56+01:00

The current research on neural machine translation (NMT) rarely involves phrase processing, which leads to poor translation quality. This paper first gives a brief introduction to NMT and the Transformer model. Then, a statistical machine translation (SMT)-based phrase processing method that adds phrases in different suffix forms to the source-end sentences was proposed to improve translation quality. Experiments were conducted on the China Workshop on Machine Translation 2018 (CWMT2018) dataset (Chinese-English) and the WMT2014 dataset (English-German). The results showed that, among the three suffix forms, only adding the target phrase sequence in the suffix form was conducive to improving the translation quality of the Transformer model: the mean bilingual evaluation understudy (BLEU) value increased by 0.0254 on the Chinese-English dataset and by 0.0105 on the English-German dataset compared with the baseline model. Compared with NMT models such as seq2seq, the Transformer model combined with phrase processing obtained the best BLEU value, and the resulting translation was more in line with the reference translation. The results verify that the proposed method is reliable and can be applied in practice.

Standardized Framework for AI-Enabled Decision Support Systems for Sustainable Sports Facility Management

2026-04-19T18:39:48+02:00

In current sports facility management, the lack of unified IoT data standards and the semantic fragmentation of multi-source heterogeneous data make it difficult for AI decision support systems to form a consistent and effective data foundation. This paper proposes a graph neural network (GNN) model based on cross-modal semantic alignment. Taking multi-source sensor data as input, it maps heterogeneous data to a shared semantic embedding space through a unified encoding mechanism. A heterogeneous graph structure is constructed based on device topology and functional coupling. Under the guidance of a graph attention mechanism, weighted propagation and fusion of semantic information between nodes are completed. Simultaneously, contrastive learning constraints are introduced to compress the representation distance between semantically similar nodes, thereby achieving semantic consistency reconstruction of cross-system data chains. Finally, a unified feature representation is output to support upper-level decision analysis. Experimental results show that the system’s interoperability consistency score improves from 0.52 to 0.89, cumulative probability increases from 0.30 to 0.82 when the alignment error is 0.2, and semantic alignment accuracy reaches 92.4%, indicating that the proposed method achieves significant improvements in cross-system data semantic consistency reconstruction and feature representation capabilities. Research shows that this method provides a feasible path for data chain integration and intelligent decision-making in sustainable sports facility management.

Federated Learning-Enabled Analysis of Digital Inequality and Inclusive Modeling for Native-AI Telecom Networks Using Social Mobility Data

2026-04-19T19:17:04+02:00

Artificial intelligence is becoming a native capability of telecom networks, which requires AI models to be trained across multiple administrative and service domains while preserving data sovereignty and trust. To support network-level digital inclusion and differentiated service provisioning, this paper proposes a federated learning-based mobile network digital inequality modeling framework that integrates social mobility data. First, heterogeneous multi-source data–including mobile network usage records, geo-temporal mobility traces, and socioeconomic indicators from different network or organizational domains–are collected to build an AI-ready data plane without exposing raw user data. Second, a distributed feature-engineering and training scheme is designed in which each participating domain locally trains a gradient-boosting decision tree model and contributes encrypted model updates to a secure aggregation procedure; differential privacy is applied to enhance AI model governance and regulatory compliance in multi-vendor/multi-tenant telecom environments. Third, a network-facing digital inequality assessment service is constructed to quantify access and usage gaps among population segments, so that intent-based management or policy-based resource allocation can target under-served groups. Experiments on five cities show that the proposed framework achieves a validation accuracy of 91.5%; low-income users consume less than 40% of the network usage time of high-income users and, when the privacy budget ε=2, the risk of data leakage is reduced by 73.2%. These results demonstrate that privacy-preserving, federated, and explainable AI can be embedded as a native capability of telecom networks to provide actionable analytics for digital inclusion policies.

From Lump-Sum to E-Taxation with a Behavioral Analysis of Payment Method Choices Among Informal Retailers

2026-02-03T14:56:21+01:00

In the transition from lump-sum to tax declaration, retailers face two choices: continue using cash or adopt e-taxation with electronic invoices. Using a behavioral approach, this study applies Institutional Pressure and the Theory of Planned Behavior (TPB) to explain payment choices in traditional markets. Partial least squares structural equation modeling (PLS-SEM) was used to test relationships with survey data from 365 retailers. The findings show coercive and mimetic pressures affect TPB mediators, whereas normative pressure influences attitude and subjective norms but not perceived control. In the case of e-payment intentions, attitude, subjective norms, and perceived control are significant predictors. For the intention to continue cash use, attitude and perceived control are significant, whereas subjective norms are not. Familiarity and security with cash sustain traditional behavior alongside modern adoption. The sample was limited to one locality, reducing generalizability.

Design of a Multi-Tenant Real-Time Inference Framework Based on OpenStack and SR-IOV GPU Virtualization

2026-02-06T15:44:52+01:00

The deployment of real-time artificial intelligence inference services on shared cloud infrastructure poses significant challenges due to resource contention, latency variability, and tail-latency amplification. While cloud platforms offer scalability and flexibility, conventional accelerator sharing mechanisms often fail to provide the determinism required by latency-sensitive inference workloads. This paper presents a standards-based, multi-tenant cloud inference framework that integrates OpenStack orchestration with Single Root I/O Virtualization (SR-IOV)-enabled graphics processing unit (GPU) partitioning to achieve predictable and isolated real-time inference execution. In the proposed architecture, each tenant is assigned an exclusive GPU virtual function, enabling hardware-level isolation while remaining fully compatible with native OpenStack scheduling and resource management mechanisms. A comprehensive experimental evaluation is conducted on a private OpenStack cloud to assess inference latency distribution, tail behavior, scalability, robustness to background network and control-plane activity, and throughput-latency trade-offs. Experimental results show that median inference latency remains stable across single-tenant and multi-tenant configurations, while P95 and P99 tail latencies exhibit no measurable amplification under concurrent execution. The system scales linearly with the number of available GPU virtual functions, maintaining consistent latency behavior until hardware capacity is reached. Additional experiments demonstrate that background network traffic and control-plane operations introduce negligible impact on inference latency. Throughput analysis reveals a well-defined saturation knee, enabling clear identification of safe operating regions for real-time inference services. By leveraging mature ICT standards and open-source cloud infrastructure, this work provides a reusable reference architecture for deploying latency-sensitive inference services in private and hybrid clouds. The results highlight the effectiveness of hardware-assisted accelerator isolation in balancing performance determinism, scalability, and operational simplicity, and offer practical guidance for future system design and standardization efforts.

A Standardized Framework for Evaluating Language Quality of Chinese Large Language Models

2026-04-17T21:19:07+02:00

Chinese large language models (CLLMs) are rapidly transitioning from research to deployed infrastructure across multiple sectors. Yet current evaluation practice remains fragmented and benchmark-centric, conflating language quality with general capability and weakening comparability across models. This is especially critical in standards-oriented contexts, where language quality must be treated as a multidimensional construct encompassing linguistic correctness, semantic adequacy, discourse coherence, style appropriateness, factual grounding, safety compliance, and robustness. In this study, we propose a standardized framework for evaluating CLLM language quality as a pre-standardization, which integrates five core elements, specifically a layered quality model, scenario-driven test specification, multi-source evidence collection (automated metrics, expert review, calibrated LLM-as-a-judge), transparent scoring and conformity schemes, and governance structures aligned with AI standards practice. Unlike conventional approaches, the framework explicitly separates language quality from broader capability, introduces a hierarchical indicator system tailored to Chinese linguistic phenomena, and defines a reproducible evaluation workflow with quality assurance and version control. To operationalize the proposal, we also develop a reference architecture for quality dimensions, indicators, scoring logic, reporting structure, and pre-standard deliverables. It further demonstrates how the framework can support both research benchmarking and practical deployment scenarios, including enterprise acceptance testing and sector-specific profile extension. The resulting framework provides a technically grounded and standards-oriented blueprint for CLLM language-quality evaluation, with potential value as both a de facto industrial evaluation specification and a foundation for future formal standardization in the information and communication technology (ICT) and artificial intelligence (AI) quality ecosystem.

5G-Advanced Network Slicing for Smart Grid Communications: Intelligent Resource Scheduling Under Energy-Efficiency and Performance Trade-Offs

2026-05-05T18:41:03+02:00

5G-Advanced network slicing is emerging as a promising communication framework for smart-grid services with diverse latency, reliability, bandwidth, and criticality requirements. In smart-grid communication infrastructures, however, slice scheduling must balance service differentiation with energy efficiency, fairness, and resilience under dynamic operating conditions. This paper presents an interpretable intelligent scheduling framework, where intelligence refers to state-aware, service-aware, energy-aware, and resilience-aware adaptation rather than purely black-box learning. The framework jointly considers slice admission, radio-resource allocation, edge-resource allocation, activity-state control, and disturbed-mode adaptation. The problem is formulated as a dynamic multi-objective scheduling problem incorporating delay, reliability, service utility, energy consumption, fairness, and resilience. On this basis, a hierarchical scheduling method is developed for normal, bursty, and degraded operating conditions. Evaluation under representative smart-grid scenarios, including mixed-service operation, demand-response events, distributed energy resource (DER) coordination surges, and degraded-capacity conditions, shows that the proposed method achieves a better overall balance among service-level agreement (SLA) satisfaction, energy efficiency, fairness, and resilience than benchmark strategies. The results indicate that intelligent 5G-Advanced slice scheduling is a promising standards-aligned approach for service-differentiated and dependable smart-grid communications.