Anomaly detection based on a deep graph convolutional neural network for reliability improvement DOI Creative Commons
Gang Xu,

Jie Hu,

Xin Qie

и другие.

Frontiers in Energy Research, Год журнала: 2024, Номер 12

Опубликована: Янв. 16, 2024

Effective anomaly detection in power grid engineering is essential for ensuring the reliability of dispatch and operation. Traditional methods based on manual review expert experience cannot be adapted to current rapid increases project data. In this work, address issue, knowledge graph technology used build an dataset. Considering over-smoothing problem associated with multi-level GCN networks, a deep skip connection framework attributed networks called DIET proposed ultra-high voltage (UHV) projects. Furthermore, distance-based object function added conventional function, which gives ability process multiple attributes same type. Several comparative experiments are conducted using five state-of-the-art algorithms. The results receiver operating characteristic area under curve (ROC-AUC) indicator show 12% minimum improvement over other methods. Other evaluation indicators such as precision@ K recall@ indicate that can achieve better rate less ranking. To evaluate feasibility model, parameter analysis number layers also performed. relatively few needed good small datasets.

Язык: Английский

Construction of Knowledge Graphs: Current State and Challenges DOI Creative Commons
Marvin Hofer, Daniel Obraczka, Alieh Saeedi

и другие.

Information, Год журнала: 2024, Номер 15(8), С. 509 - 509

Опубликована: Авг. 22, 2024

With Knowledge Graphs (KGs) at the center of numerous applications such as recommender systems and question-answering, need for generalized pipelines to construct continuously update KGs is increasing. While individual steps that are necessary create from unstructured sources (e.g., text) structured data databases) mostly well researched their one-shot execution, adoption incremental KG updates interplay have hardly been investigated in a systematic manner so far. In this work, we first discuss main graph models introduce major requirements future construction pipelines. Next, provide an overview build high-quality KGs, including cross-cutting topics metadata management, ontology development, quality assurance. We then evaluate state art with respect introduced specific popular some recent tools strategies construction. Finally, identify areas further research improvement.

Язык: Английский

Процитировано

16
GaitMGL: Multi-Scale Temporal Dimension and Global–Local Feature Fusion for Gait Recognition DOI Open Access
Zhipeng Zhang, Siwei Wei,

Liya Xi

и другие.

Electronics, Год журнала: 2024, Номер 13(2), С. 257 - 257

Опубликована: Янв. 5, 2024

Gait recognition has received widespread attention due to its non-intrusive mechanism. Currently, most gait methods use appearance-based methods, and such are easily affected by occlusions when facing complex environments, which in turn affects the accuracy. With maturity of pose estimation techniques, model-based have more their robustness environments. However, current mainly focus on modeling global feature information spatial dimension, ignoring importance local features influence Meanwhile, temporal these usually single-scale extraction, does not take into account inconsistency motion cycles limbs a human body is walking (e.g., arm swing leg pace), leading loss some limb information. To solve problems, we propose network based Global–Local Graph Convolutional Network, called GaitMGL. Specifically, introduce new spatio-temporal extraction module, MGL (Multi-scale Temporal Spatial Extraction Module), consists GLGCN (Global–Local Network) MTCN Network). models both features, extracts global–local MTCN, other hand, takes cycles, facilitates multi-scale convolution capture motion. In short, our GaitMGL solves problems at single scale that exist existing networks. We evaluated method three publicly available datasets, CASIA-B, Gait3D, GREW, experimental results show demonstrates surprising performance achieves an accuracy 63.12% dataset exceeding all

Язык: Английский

Процитировано

7
A Review on the Large Language Model Augmented Knowledge Graph Question Answer: Task, Model, Advance and Outlook DOI Creative Commons
Rongdong Yu, Dou Wang, Xiaoyan Jia

и другие.

Lecture notes in electrical engineering, Год журнала: 2025, Номер unknown, С. 333 - 347

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Domain- and Language-Adaptable Natural Language Interface for Property Graphs DOI Creative Commons
Ioannis Tsampos, Emmanouil Marakakis

Computers, Год журнала: 2025, Номер 14(5), С. 183 - 183

Опубликована: Май 9, 2025

Despite the growing adoption of Property Graph Databases, like Neo4j, interacting with them remains difficult for non-technical users due to reliance on formal query languages. Natural Language Interfaces (NLIs) address this by translating natural language (NL) into Cypher. However, existing solutions are typically limited high-resource languages; adapt evolving domains annotated data; and often depend Machine Learning (ML) approaches, including Large Models (LLMs), that demand substantial computational resources advanced expertise training maintenance. We these limitations introducing a novel dependency-based, training-free, schema-agnostic Interface (NLI) converts NL queries Cypher querying Graphs. Our system employs modular pipeline-integrating entity relationship extraction, Named Entity Recognition (NER), semantic mapping, triple creation via syntactic dependencies, validation against an automatically extracted Schema Graph. The distinctive feature approach is reduction in candidate pairs using analysis schema validation, eliminating need generation ranking. design enables adaptation across supports single- multi-hop queries, conjunctions, comparisons, aggregations, complex questions through explainable process. Evaluations real-world demonstrate reliable translation results.

Язык: Английский

Процитировано

0
Enhanced Heterogeneous Graph Attention Network with a Novel Multilabel Focal Loss for Document-Level Relation Extraction DOI Creative Commons
Yang Chen, Bowen Shi

Entropy, Год журнала: 2024, Номер 26(3), С. 210 - 210

Опубликована: Фев. 28, 2024

Recent years have seen a rise in interest document-level relation extraction, which is defined as extracting all relations between entities multiple sentences of document. Typically, there are mentions corresponding to single entity this context. Previous research predominantly employed holistic representation for each predict relations, but approach often overlooks valuable information contained fine-grained mentions. We contend that prediction and inference should be grounded specific rather than abstract concepts. To address this, our paper proposes two-stage mention-level framework based on an enhanced heterogeneous graph attention network extraction. Our employs two different strategies model intra-sentential inter-sentential mentions, yielding local mention representations global prediction. For inference, we propose better the long-distance semantic relationships design entity-coreference path-based strategy conduct inference. Moreover, introduce novel cross-entropy-based multilabel focal loss function class imbalance problem simultaneously. Comprehensive experiments been conducted verify effectiveness framework. Experimental results show significantly outperforms existing methods.

Язык: Английский

Процитировано

3
Enhancing SPARQL Query Generation for Knowledge Base Question Answering Systems by Learning to Correct Triplets DOI Creative Commons
Jiexing Qi, Chang Su, Zhixin Guo

и другие.

Applied Sciences, Год журнала: 2024, Номер 14(4), С. 1521 - 1521

Опубликована: Фев. 14, 2024

Generating SPARQL queries from natural language questions is challenging in Knowledge Base Question Answering (KBQA) systems. The current state-of-the-art models heavily rely on fine-tuning pretrained such as T5. However, these methods still encounter critical issues triple-flip errors (e.g., (subject, relation, object) predicted (object, subject)). To address this limitation, we introduce TSET (Triplet Structure Enhanced T5), a model with novel pretraining stage positioned between the initial T5 and for Text-to-SPARQL task. In intermediary stage, new objective called Triplet Correction (TSC) to train corpus derived Wikidata. This aims deepen model’s understanding of order triplets. After specialized pretraining, undergoes query generation, augmenting its query-generation capabilities. We also propose method named “semantic transformation” fortify grasp syntax semantics without compromising pre-trained weights Experimental results demonstrate that our proposed outperforms existing three well-established KBQA datasets: LC-QuAD 2.0, QALD-9 plus, QALD-10, establishing performance (95.0% F1 93.1% QM 75.85% 61.76% 51.37% 40.05% QALD-10).

Язык: Английский

Процитировано

1
Construction of an Event Knowledge Graph Based on a Dynamic Resource Scheduling Optimization Algorithm and Semantic Graph Convolutional Neural Networks DOI Open Access
Xing Liu, Long Zhang, Qiusheng Zheng

и другие.

Electronics, Год журнала: 2023, Номер 13(1), С. 11 - 11

Опубликована: Дек. 19, 2023

Presently, road and traffic control construction on most university campuses cannot keep up with the growth of universities. Campus roads are not very wide, crossings do have lights, there no full-time management personnel. Teachers students prone to forming a peak flow people when going from classes. This has led constant stream accidents. It is critical conduct comprehensive analysis this issue by utilizing voluminous data pertaining school incidents in order safeguard lives faculty students. In case domestic universities, fewer studies studied knowledge graph methods for safety incidents. event construction, reasonable release recycling computational resources inefficient, existing entity–relationship joint extraction unable deal ternary overlapping entity boundary ambiguity problems relationship extraction. response above problems, paper proposes method on-campus events improved dynamic resource scheduling algorithms multi-layer semantic convolutional neural networks. The experiment’s results show that proposed increases GPU CPU use 25% 9%. On public dataset, model’s F1 scores triples increase 1.3% NYT dataset 0.4% WebNLG dataset. can help relevant personnel dealing unexpected reduce impact opinion.

Язык: Английский

Процитировано

3
Anomaly detection based on a deep graph convolutional neural network for reliability improvement DOI Creative Commons
Gang Xu,

Jie Hu,

Xin Qie

и другие.

Frontiers in Energy Research, Год журнала: 2024, Номер 12

Опубликована: Янв. 16, 2024

Effective anomaly detection in power grid engineering is essential for ensuring the reliability of dispatch and operation. Traditional methods based on manual review expert experience cannot be adapted to current rapid increases project data. In this work, address issue, knowledge graph technology used build an dataset. Considering over-smoothing problem associated with multi-level GCN networks, a deep skip connection framework attributed networks called DIET proposed ultra-high voltage (UHV) projects. Furthermore, distance-based object function added conventional function, which gives ability process multiple attributes same type. Several comparative experiments are conducted using five state-of-the-art algorithms. The results receiver operating characteristic area under curve (ROC-AUC) indicator show 12% minimum improvement over other methods. Other evaluation indicators such as precision@ K recall@ indicate that can achieve better rate less ranking. To evaluate feasibility model, parameter analysis number layers also performed. relatively few needed good small datasets.

Язык: Английский

Процитировано

0