Knowledge Graph Research Articles

A Knowledge Graph Summarization Model Integrating Attention Alignment and Momentum Distillation

The integrated knowledge graph summarization model improves summary performance by combining text features and entity features. However, the model still has the following shortcomings: the knowledge graph data used introduce data noise that deviates from the original text semantics; and the text and knowledge graph entity features cannot be fully integrated. To address these issues, a knowledge graph summarization model integrating attention alignment and momentum distillation (KGS-AAMD) is proposed. The pseudo-targets generated by the momentum distillation model serve as additional supervision signals during training to overcome data noise. The attention-based alignment method lays the foundation for the subsequent full integration of text and entity features by aligning them. Experimental results on two public datasets, namely CNN / Daily Mail and XSum, show that KGS-AAMD surpasses multiple baseline models and ChatGPT in terms of the quality of summary generation, exhibiting significant performance advantages.

A novel approach for target deconvolution from phenotype-based screening using knowledge graph

Deconvoluting drug targets is crucial in modern drug development, yet both traditional and artificial intelligence (AI)-driven methods face challenges in terms of completeness, accuracy, and efficiency. Identifying drug targets, especially within complex systems such as the p53 pathway, remains a formidable task. The regulation of this pathway by myriad stress signals and regulatory elements adds layers of complexity to the discovery of effective p53 pathway activators. Recent insights into p53 activation have led to two main screening strategies for p53 activators. The target-based approach focuses on p53 and its regulators (MDM2, MDMX, USP7, Sirt proteins), but requires separate systems for each target and may miss multi-target compounds. Phenotype-based screening can reveal new targets but involves a lengthy process to elucidate mechanisms and targets, hindering drug development. Knowledge graphs have emerged as powerful tools that offer strengths in link prediction and knowledge inference to address these issues. In this study, we constructed a protein-protein interaction knowledge graph (PPIKG) and pioneered an integrated drug target deconvolution system that combines AI with molecular docking techniques. Analysis based on the PPIKG narrowed down candidate proteins from 1088 to 35, significantly saving time and cost. Subsequent molecular docking led us to pinpoint USP7 as a direct target for the p53 pathway activator UNBS5162. Leveraging knowledge graphs and a multidisciplinary approach allows us to streamline the laborious and expensive process of reverse targeting drug discovery through phenotype screening. Our findings have the potential to revolutionize drug screening and open new avenues in pharmacological research, increasing the speed and efficiency of pursuing novel therapeutics. The code is available at https://github.com/Xiong-Jing/PPIKG.

Dual-view embedding for hyper-relational knowledge graphs with hierarchical structure

Dual-view embedding for hyper-relational knowledge graphs with hierarchical structure

Type-based computation of knowledge graph statistics

Type-based computation of knowledge graph statistics

A knowledge graph completion model based on weighted fusion description information and transform of the dimension and the scale

A knowledge graph completion model based on weighted fusion description information and transform of the dimension and the scale

RDF Knowledge Graphs Designed with Axiomatic Methodology to Enhance Industry 4.0

Industry 4.0 has introduced a data-driven model of production and management of goods and services. This manufacturing paradigm leverages the potential of the Internet of Things (IoT), but finding the information necessary to drive manufacturing processes can be challenging. In this context, the authors propose an innovative approach based on axiomatic design to design RDF knowledge graphs from which to extract the information needed by decision makers. This approach derives from the possibility of providing RDF knowledge graphs with an equivalent matrix representation based on axiomatic design. It allows the selection of the most reliable data sources, thereby optimizing the knowledge graph construction process using matrix algebra, minimizing redundancy and improving the efficiency of query response. The goal of the presented methodology is to address the five critical aspects of Big Data (volume, velocity, variety, value, and veracity) by preordering the knowledge graph according to the information needs of business decision makers, thereby optimizing the use of the immense wealth of information made available by the Web in design.

Dual intent view contrastive learning for knowledge aware recommender systems

Knowledge-aware recommendation systems often face challenges owing to sparse supervision signals and redundant entity relations, which can diminish the advantages of utilizing knowledge graphs for enhancing recommendation performance. To tackle these challenges, we propose a novel recommendation model named Dual-Intent-View Contrastive Learning network (DIVCL), inspired by recent advancements in contrastive and intent learning. DIVCL employs a dual-view representation learning approach using Graph Neural Networks (GNNs), consisting of two distinct views: a local view based on the user-item interaction graph and a global view based on the user-item-entity knowledge graph. To further enhance learning, a set of intents are integrated into each user-item interaction as a separate class of nodes, fulfilling three crucial roles in the GNN learning process: (1) providing fine-grained representations of user-item interaction features, (2) acting as evaluators for filtering relevant relations in the knowledge graph, and (3) participating in contrastive learning to strengthen the model’s ability to handle sparse signals and redundant relations. Experimental results on three benchmark datasets demonstrate that DIVCL outperforms state-of-the-art models, showcasing its superior performance. The implementation is available at: https://github.com/yzxx667/DIVCL.

Learning behaviour prediction and multi-task recommendation based on a knowledge graph in MOOCs

ABSTRACT One challenging issue in improving the teaching and learning methods in MOOCs is to construct potential knowledge graphs from massive learning resources. Therefore, this study proposes knowledge graphs driving online learning behaviour prediction and multi-learning task recommendation in MOOCs. Based on the knowledge graphs supported by multi-entities and features, the authors designed a novel deep learning model to fully calculate and optimise iterative learning processes, explore key semantics and potential values of learning behaviours, integrate multi-task items and flexibly analyse learners’ multi-entities and related features. Assisted by knowledge concept semantics and the formation of knowledge graphs, multi-entities of learning behaviours are fused and associated, which might be the leading trend of knowledge graphs. Sufficient experiments have proved that the whole work might ensure learning behaviour prediction and multi-task selection that can enable the knowledge graph to improve learning behaviours in MOOCs.

Knowledge Integration and Analysis of Technological Innovation in Prefabricated Construction

Prefabricated construction (PC) plays a critical role in advancing the sustainable and high-quality transformation of the construction industry. Nevertheless, the fragmented and variable nature of technological innovations in PC complicates their acquisition, comprehension, and practical application, thereby hindering the process of innovation transformation. In response to these challenges, this study applies knowledge graph techniques to aggregate, correlate, and store knowledge pertaining to PC technological innovations. Specifically, using patent data from the past five years, and grounded in knowledge management and complex network theories, this study employs text mining, topic modeling, and association rule algorithms to perform clustering, evolutionary, and association analyses. The extracted entities and relationships obtained from the analyses are then stored in a Neo4j graph database for the construction and interactive visualization of a knowledge graph for PC technological innovation. According to the knowledge graph, a question-and-answer system framework is further proposed, providing practical application guidance. This research provides a comprehensive overview of the technological landscape, key nodes, and development trends in PC. It makes a meaningful contribution to knowledge management theory and complex network theory, advancing innovative applications in PC technology.

AT4CTIRE: Adversarial Training for Cyber Threat Intelligence Relation Extraction

Cyber Threat Intelligence (CTI) plays a crucial role in cybersecurity. However, traditional information extraction has low accuracy due to the specialization of CTI and the concealment of relations. To improve the performance of CTI relation extraction in the knowledge graph, we propose a relation extraction architecture called Adversarial Training for Cyber Threat Intelligence Relation Extraction (AT4CTIRE). Additionally, we developed a large-scale cybersecurity dataset for CTI analysis and evaluation called Cyber Threat Intelligence Analysis (CTIA). Inspired by Generative Adversarial Networks, we integrate contextual semantics to refine our study. Firstly, we use some wrong triples with incorrect relations to train the generator and produce high-quality generated triples as adversarial samples. Secondly, the discriminator used actual and generated samples as training data. Integrating the discriminator and the context-embedding module facilitates a deeper understanding of contextual CTI within threat triples. Finally, training a discriminator identified the relation between the threat entities. Experimentally, we set two CTI datasets and only one baseline that we could find to test the effect in the cybersecurity domain. We also took general knowledge graph completion tests. The results demonstrate that AT4CTIRE outperforms existing methods with improved extraction accuracy and a remarkable expedited training convergence rate.

FooDS: Ontology-based Knowledge Graphs for Forest Observatories

Wildlife research activities generate data on ecosystems and species interactions from varied independent projects. Forest Observatories are online platforms that curate, integrate, and analyze wildlife research data for forest monitoring. However, integrating data from disparate sources can be challenging due to data heterogeneity. This study, in collaboration with a research facility in the forest of Sabah, Malaysian Borneo, proposes a novel approach to integrate heterogeneous wildlife data for Forest Observatories. We used the Forest Observatory Ontology (FOO) to standardize wildlife data entities generated by sensors. Four semantically modeled wildlife datasets populated FOO, resulting in an ontology-based knowledge graph named FooDS (Forest Observatory Ontology Data Store). We evaluated FOO and FooDS using specialized open-source ontology scanners, domain experts’ feedback, and applied use cases. This study contributes FooDS, the first ontology-based knowledge graph for Forest Observatories, which provides accurate query responses, reasoning about data, and granular data acquisition from diverse datasets. FOO in turtle format, FOO’s documentation and FooDS in turtle format and their resource website are published at https://w3id.org/def/foo , https://w3id.org/def/fooDocs , https://w3id.org/def/fooDS , and https://ontology.forest-observatory.org .

An Ontology‐Augmented Digital Twin for Fiber‐Reinforced Polymer Structures at the Example of Wind Turbine Rotor Blades

A methodology for establishing a structural digital twin is proposed to facilitate the lifetime prediction of fiber‐reinforced polymer (FRP) structures, in this case, a wind turbine rotor blade. The digital twin incorporates production peculiarities and imperfections occurring during the manufacturing process of the FRP component. The methodology involves the computation of process‐defined effective elastic properties and residual stresses through numerical simulation of the resin cure cycle. The results are then transferred to a structural finite‐element model. By applying local wind conditions to this model, a comprehensive state of stress is obtained. This serves as a basis for a practical evaluation of material fatigue within the composite, leading to the prediction of the component's lifetime. The entire workflow is implemented in a Jupyter‐based application that uses an ontology with an appertaining knowledge graph to facilitate the transfer of intermediate results between the observation scales and process steps of the digital twin. In line with the principles of open science, the methodology utilizes open‐source software.

Constructing Knowledge Graphs for New Species in the Greater Mekong Subregion

Purpose: This study presents a methodology for developing a knowledge graph by integrating World Wildlife Fund (WWF) species discovery reports in the Greater Mekong Subregion. Design/methodology/approach: The research employs a four-phase approach: constructing a taxonomic data model, implementing a graph database, developing a web application architecture, and creating an interactive user interface. Findings: The knowledge graph integrates 2,783 species across 27 taxonomic groups, comprising 12,089 nodes and 30,303 connections. The system enables complex queries through SPARQL endpoints, revealing patterns in species distribution and conservation status from 1997 to 2022. Research limitations: Primary limitations include geographical constraints to the Greater Mekong Subregion and limited integration with external biodiversity databases. Practical implications: The system provides researchers with a graph search engine for exploring species data, supporting scientific research and conservation planning. Originality/value: This study presents a novel approach to organizing biodiversity data by combining taxonomic hierarchies with conservation metrics, establishing a framework for integrating species discovery data.

Personalized federated knowledge graph embedding with client-wise relation graph

Personalized federated knowledge graph embedding with client-wise relation graph

A knowledge graph-based intelligent planning method for remanufacturing processes of used parts

Intelligent remanufacturing process planning is crucial for the efficient and high-quality remanufacturing of used parts with complex failure characteristics. However, due to the varied failure characteristics of used parts, the diversity of remanufacturing processes, and complex non-linear relationships among remanufacturing process elements, relying solely on mathematical programming or manual empirical is difficult to effectively model and optimise the remanufacturing process planning. To this end, a knowledge graph-based intelligent planning method for remanufacturing processes is proposed to enhance efficiency and quality by combining mathematical programming and knowledge reuse. Firstly, with failure characteristics as decision nodes, a full-element remanufacturing process ontology model is constructed, linking used parts, failure characteristics, and corresponding process plans. The BERT-BiLSTM-CRF model extracts remanufacturing process entities, and a remanufacturing process knowledge graph (RPKG) is constructed. Secondly, an intelligent decision-making model based on graph multi-node path retrieval is proposed. Aim to minimise carbon emissions, time, and cost, combining feature similarity calculations and nearest neighbour search (NNS) to efficiently retrieve the optimal process plan for each failure characteristic. Then, the optimal process plans are merged based on process constraints to create the complete plan. Finally, a concrete case is given to verify the effectiveness and advantages of this method.

Chinese Mathematical Knowledge Entity Recognition Based on Linguistically Motivated Bidirectional Encoder Representation from Transformers

We assessed whether constructing a mathematical knowledge graph for a knowledge question-answering system or a course recommendation system, Named Entity Recognition (NER), is indispensable. The accuracy of its recognition directly affects the actual performance of these subsequent tasks. In order to improve the accuracy of mathematical knowledge entity recognition and provide effective support for subsequent functionalities, this paper adopts the latest pre-trained language model, LERT, combined with a Bidirectional Gated Recurrent Unit (BiGRU), Iterated Dilated Convolutional Neural Networks (IDCNNs), and Conditional Random Fields (CRFs), to construct the LERT-BiGRU-IDCNN-CRF model. First, LERT provides context-related word vectors, and then the BiGRU captures both long-distance and short-distance information, the IDCNN retrieves local information, and finally the CRF is decoded to output the corresponding labels. Experimental results show that the accuracy of this model when recognizing mathematical concepts and theorem entities is 97.22%, the recall score is 97.47%, and the F1 score is 97.34%. This model can accurately recognize the required entities, and, through comparison, this method outperforms the current state-of-the-art entity recognition models.

Evaluating the effectiveness of prompt engineering for knowledge graph question answering

Many different methods for prompting large language models have been developed since the emergence of OpenAI's ChatGPT in November 2022. In this work, we evaluate six different few-shot prompting methods. The first set of experiments evaluates three frameworks that focus on the quantity or type of shots in a prompt: a baseline method with a simple prompt and a small number of shots, random few-shot prompting with 10, 20, and 30 shots, and similarity-based few-shot prompting. The second set of experiments target optimizing the prompt or enhancing shots through Large Language Model (LLM)-generated explanations, using three prompting frameworks: Explain then Translate, Question Decomposition Meaning Representation, and Optimization by Prompting. We evaluate these six prompting methods on the newly created Spider4SPARQL benchmark, as it is the most complex SPARQL-based Knowledge Graph Question Answering (KGQA) benchmark to date. Across the various prompting frameworks used, the commercial model is unable to achieve a score over 51%, indicating that KGQA, especially for complex queries, with multiple hops, set operations and filters remains a challenging task for LLMs. Our experiments find that the most successful prompting framework for KGQA is a simple prompt combined with an ontology and five random shots.

Hot Spots and Frontiers of Livestreaming Communication Research in China in the Past 10 Years---- Bibliometric Analysis and Knowledge Graph Visualization Using Cite Space

Chinese research on livestreaming communication lacks systematic theoretical analysis and comprehensive understanding; hotspots and frontiers are still unclear. Based on the relevant literature in the past ten years (2013-2023) on CNKI, this paper explores the development trends, research status, and keyword clustering in this field. It was found that academic articles on livestreaming communication in China show a significant growth trend. The past decade's research can be divided into three stages of development, and the establishment of strong collaboration among prolific authors, represented by Yu Guoming, Liu Jia, Hua Jie, and Yan Sanjiu, remains elusive. Eight hotspots emerged in live streaming communication research during the four evolutionary phases: webcasting, livestreaming reports, mainstream media, e-commerce live streaming, sports events, media convergence, livestreaming bandwagons, and live streaming platforms. Moreover, five major research frontiers were identified: mobile live streaming, live streaming with goods, slow live streaming, live streaming with goods, and e-commerce live streaming. Regarding knowledge structure, in the cluster analysis, webcasting, e-commerce live streaming, and livestreaming coverage scale are more significant and are the current areas of concern for scholars. This study helps scholars in livestreaming communication to discover new frontiers and distinguish key knowledge subfields.

Research on the Construction of Oracle Bone Inscriptions Knowledge Graph and its Application in Historical Language Analysis

Research on the Construction of Oracle Bone Inscriptions Knowledge Graph and its Application in Historical Language Analysis

Homo Sapiens Chromosomal Location Ontology: A Framework for Genomic Data in Biomedical Knowledge Graphs

The Homo sapiens Chromosomal Location Ontology (HSCLO) is designed to facilitate the integration of human genomic features into biomedical knowledge graphs from releases GRCh37 and GRCh38 at multiple resolutions. HSCLO comprises two distinct versions, HSCLO37 and HSCLO38, each tailored to its respective human genome release. This ontology supports the efficient integration and analysis of human genomic data across scales ranging from entire chromosomes to individual base pairs, thereby enhancing data retrieval and interoperability within large-scale biomedical datasets. Unlike existing ontologies that primarily focus on genomic feature identification or annotation, HSCLO is specifically engineered to optimize the interoperability and scalability of genomic data within biomedical knowledge graphs. The utility and performance of HSCLO are demonstrated through a case study involving the integration of high-resolution chromatin interaction data, which reveals significant improvements in query efficiency and data linkage. HSCLO represents a valuable resource for advancing research in disease genetics, personalized medicine, and other domains that require complex genomic data integration.