检索结果-内蒙古大学图书馆

A Survey of the Application of Neural Networks to Event Extraction

Tsinghua Science and Technology 2025年第2期30卷 748-768页

作者： Jianye Xie Yulan Zhang Huaizhen Kou Xiaoran Zhao Zhikang Feng Lekang Song Weiyi Zhong College of Computer Science and Technology China University of Petroleum(East China)Qingdao 266000China Shandong Provincial University Laboratory for Protected Horticulture Weifang University of Science and TechnologyWeifang 261000China School of Computer Science and Engineering Nanjing University of Science and TechnologyNanjing 210000China School of Computer Science Qufu Normal UniversityRizhao 276800China

Event extraction is an important part of natural language information extraction,and it’s widely employed in other natural language processing tasks including question answering and machine reading ***,there is a lack of recent comprehensive survey papers on event *** the past few years,numerous high-quality and innovative event extraction methods have been proposed,making it necessary to consolidate these new developments with previous work in order to provide a clear overview for researchers and serve as a reference for future *** addition,event detection is a fundamental sub-task in event extraction,previous survey papers have often overlooked the related work on event ***,this paper aims to bridge these gaps by presenting a comprehensive survey of event extraction,including recent advancements and an analysis of previous research on event *** resources for event extraction are first introduced in this research,and then the numerous neural network models currently employed in event extraction tasks are divided into four types:word sequence-based methods,graph-based neural network methods,external knowledge-based approaches,and prompt-based *** compare and contrast them in depth,pointing out the flaws and difficulties with existing ***,we discuss the future of event extraction development.

关键词： event extraction natural language processing event extraction methods graph neural network prompt-based learning

来源：评论

学校读者我要写书评

暂无评论

Knowledge graph Completion With Pattern-based methods

引用

IEEE ACCESS 2025年 13卷 5815-5831页

作者： Sabet, Maryam Pajoohan, Mohammadreza Moosavi, Mohammad Reza Yazd Univ Comp Engn Dept Yazd *** Iran Shiraz Univ Sch Elect & Comp Engn Dept Comp Sci & Engn & IT Shiraz *** Iran

Knowledge graphs (KGs) are popularly used to develop several intelligent applications. Revealing valuable knowledge hidden in these graphs opened up a branch of research, known as KG reasoning, aiming at predicting the missing links. Some methods take advantage of external information such as entity description but at the cost of more computational complexity. Besides, most of the current techniques focus solely on local information in the KG. However, the learning process can utilise valuable global information in the entire graph. In this paper, we propose a Pattern-based Knowledge graph Completion (PKGC) method that consists of three phases. The first phase utilizes multi-source information and expands the KG using entity description as external information with efficient natural language processing (NLP) techniques. In the second phase, we mine frequent patterns from the expanded KG, extract connections between them and assign entities to the patterns that construct the abstraction layer. based on the extracted patterns, connections, and entity assignments, a flow network is constructed on the abstraction layer in the third phase. We use global internal information, namely patterns, by adapting the minimum-cost circulation problem to the flow network. This way the links in a larger neighborhood are involved in the inference. We conducted experiments on the link prediction task and evaluated the training time on two benchmark datasets, WordNet and Freebase. Experiments have demonstrated that the proposed method is superior to the state-of-the-art methods and that pattern extraction is effective for knowledge graph completion tasks.

关键词： Cognition Knowledge graphs Vectors Semantics Tail Translation natural language processing Training Tensors Surveys Frequent pattern mining flow network knowledge graph completion minimum-cost circulation problem

来源：评论

学校读者我要写书评

暂无评论

Emotion Classification With Visibility graphs

引用

IEEE Signal processing Letters 2025年 32卷 2055-2059页

作者： Şimşek, Ecem Topcu, Atakan Koç, Emirhan Saritas, Emine Ulku Koç, Aykut Bilkent University Dept. of Electrical and Electronics Engineering Ankara06800 Turkey Bilkent University UMRAM Ankara06800 Turkey

Transformers have gained prominence in natural language processing due to their representational capabilities and performances. Transformers process natural language as a sequence on finite context windows;however, global relationships among words beyond these windows cannot be completely modeled via sequence processing only. graph neural network (GNN) based models have been proposed to alleviate this problem, as they provide geometric extensions to neural networks, enabling models to learn associations within a text. However, regular graph-based methods ignore the sequential nature of underlying texts. In this paper, we propose EmoVis, the first generic graph-based neural network that utilizes visibility graphs, which converts classical time-series information to graph representations. We cast the problem as an emotion classification task, enabling the proposed model to learn associations between the labels and words in a sentence. Moreover, EmoVis can be used as a highly modular graph-based extension to any transformer-based model, significantly improving their performance and learning capabilities in various languages. We experimentally show that EmoVis enables transformer-based models to outperform the state-of-the-art baselines across three diverse datasets in different languages in the SemEval2018 competition datasets and the GoEmotions dataset. © 1994-2012 IEEE.

关键词： natural language processing systems

来源：评论

学校读者我要写书评

暂无评论

Universal embedding for pre-trained models and data bench

引用

NEUROCOMPUTING 2025年 619卷

作者： Cho, Namkyeong Cho, Taewon Shin, Jaesun Jeon, Eunjoo Lee, Taehee Pohang Univ Sci & Technol POSTECH Ctr Math Machine Learning & its Applicat CM2LA Dept Math Pohang 37673 Gyeongbuk South Korea Samsung SDS 125 Olymp Ro 35 Gil Seoul 05510 South Korea

The transformer architecture has shown significant improvements in the performance of various natural language processing (NLP) tasks. One of the great advantages of transformer-based model is that they allow for the addition of an extra layer to a pre-trained model (PTM) and fine-tuning, rather than requiring the development of a separate architecture for each task. This approach has provided great promising performance in NLP tasks. Therefore, selecting an appropriate PTM from the model zoo, such as Hugging Face, becomes a crucial task. Despite the importance of PTM selection, it still requires further investigation. The main challenge in PTM selection for NLP tasks is the lack of a publicly available benchmark to evaluate model performance for each task and dataset. To address this challenge, we introduce the first public data benchmark to evaluate the performance of popular transformer-based models on diverse ranges of NLP tasks. Furthermore, we propose graph representations of transformer-based models with node features that represent the matrix weight on each layer. Empirical results demonstrate that our proposed graph neural network (GNN) model outperforms existing PTM selection methods.

关键词： Transfer learning Pretrained models graph neural networks

来源：评论

学校读者我要写书评

暂无评论

GSEA: Global Structure-Aware graph Neural Networks for Entity Alignment 13th

GSEA: Global Structure-Aware Graph Neural Networks for Entit...

引用

13th International Conference on natural language processing and Chinese Computing

作者： Wang, Cunda Wang, Weihua Liang, Qiuyu Yu, Jie Gao, Guanglai Inner Mongolia Univ Coll Comp Sci Hohhot Peoples R China Natl & Local Joint Engn Res Ctr Intelligent Infor Hohhot Peoples R China Inner Mongolia Key Lab Multilingual Artificial In Hohhot Peoples R China Natl Univ Def Technol Coll Comp Sci Changsha Peoples R China

ISBN: (纸本)9789819794331;9789819794348

Entity alignment (EA) aims to identify the same real-world entities presented in different knowledge graphs (KGs), which is the most crucial step in integrating multi-source KGs. Existing entity alignment methods based on graph Neural Networks mainly focus on aggregating neighborhood structure information within the original graph to generate entity embeddings. However, these methods fail to effectively leverage the global structural information of entities. In this paper, we propose a novel method based on Global Structure-Aware graph Neural Networks to capture the global structure between entities. Specifically, our method learns global embeddings from the augmented graph reconstructed by Singular Value Decomposition. Moreover, we employ contrastive learning to maximize the consistency between global and local embeddings while mitigating the over-smoothing problem. Extensive experimental results on benchmark datasets, along with further analysis, demonstrate the superiority and effectiveness of our method. The source code is available at https://***/wonderCS1213/GSEA.

关键词： Knowledge graph Entity alignment graph neural networks Contrastive learning

来源：评论

学校读者我要写书评

暂无评论

Multi-level feature subgraph aggregation graph neural network

引用

INFORMATION FUSION 2025年 121卷

作者： Xie, Chengxin Song, Qiya Huang, Jingui Chen, Wei Pan, Liangrui Hunan Normal Univ Coll Informat Sci & Engn Changsha 410000 Peoples R China Hunan Normal Univ Hunan Prov Key Lab Intelligent Comp & Language Inf Changsha 410000 Peoples R China Hebei Univ Architecture Coll Informat Engn Zhangjiakou 075000 Peoples R China Hunan Univ Coll Comp Sci & Elect Engn Changsha 410000 Peoples R China

Heterogeneous graph Convolutional Networks (HGCNs) have become a pivotal technology for pattern mining in heterogeneous graphs. However, existing methods for heterogeneous graphs often overlook the distinct interactions between homogeneous and heterogeneous neighbors. Additionally, they fail to capture high-level features typically present in graphs, such as global topology, semantics, and attributes. To address these limitations, we propose a general and precise multi-level feature subgraph aggregation model for heterogeneous graphs (SAGNN). The SAGNN model first applies nonlinear transformations to map features of various node types into latent subspaces. A meta-path-based subgraph generation module is then used to decompose the heterogeneous graph into both homogeneous and heterogeneous subgraphs. To enable effective aggregation of node information within these subgraphs, we introduce a innovative subgraph aggregation strategy. Finally, weighted coefficients are applied to integrate node information from both homogeneous and heterogeneous subgraphs, ensuring the comprehensive fusion of multi-level features. Extensive experimental results demonstrate that the proposed SAGNN effectively extracts and fuses diverse features within heterogeneous graph structures, thereby improving performance in tasks such as node classification and clustering.

关键词： Heterogeneous graph Homogeneous graph Multi-level features Subgraph aggregation

来源：评论

学校读者我要写书评

暂无评论

Effective Knowledge graph Embedding with Quaternion Convolutional Networks 13th

Effective Knowledge Graph Embedding with Quaternion Convolut...

引用

13th International Conference on natural language processing and Chinese Computing

作者： Liang, Qiuyu Wang, Weihua Yu, Lie Bao, Feilong Inner Mongolia Univ Coll Comp Sci Hohhot Peoples R China Natl & Local Joint Engn Res Ctr Intelligent Infor Hohhot Peoples R China Inner Mongolia Key Lab Multilingual Artificial In Hohhot Peoples R China Natl Univ Def Technol Coll Comp Sci Changsha Peoples R China

ISBN: (纸本)9789819794362;9789819794379

Convolutional Neural Networks (CNNs) have demonstrated effectiveness in knowledge graph embedding, but existing CNN-based methods encounter two main challenges. firstly, CNN-based models with simple architectures are unable to extract latent features. Secondly, these models enhance feature extraction by adding extra modules, which inevitably increases training cost with limited performance improvement. To address these challenges, we go beyond traditional CNNs and propose a novel knowledge graph embedding model, which utilizes the powerful capability of Quaternion Convolutional Neural Networks (QCNNs) for effective representation learning. Specifically, we learn representations of entities and relations in a quaternion space and utilize QCNN to extract the inherent information of the entity-relation matrix. We evaluate the performance of our model on multiple knowledge graph completion benchmark datasets. Experimental results show that our model achieves effective improvements compared to existing CNN-based models. Moreover, in terms of training time, our model is faster than other outstanding models. The code of all experiments is available at https://***/llqy123/ConvQE.

关键词： Quaternion algebra Knowledge graph embedding Link prediction Convolutional neural network

来源：评论

学校读者我要写书评

暂无评论

A large language model-based building operation and maintenance information query

引用

ENERGY AND BUILDINGS 2025年 334卷

作者： Li, Yan Ji, Minxuan Chen, Junyu Wei, Xin Gu, Xiaojun Tang, Juemin Nanjing Univ Posts & Telecommun Sch Commun & Informat Engn Nanjing 210003 Jiangsu Peoples R China Anhui Jianzhu Univ Anhui Prov Key Lab Intelligent Bldg & Bldg Energy Hefei 230022 Anhui Peoples R China Nanjing Yangtze River Urban Architectural Design C Nanjing 210002 Jiangsu Peoples R China

In smart buildings, operation and maintenance (O&M) are critical for achieving energy optimization and ensuring comfort. However, as smart buildings incorporate more systems and generate increasingly complex data, efficiently querying this information has become a major challenge for O&M tasks. Traditional methods require complex queries, considering data standards and building structure information, which is demanding, while current artificial intelligence (AI) querying technologies lack accuracy and interactivity. To address this, this paper proposes a large language model (LLM)-based information query paradigm for building O&M. Under this paradigm, the first step is data processing, which adopts the isolation forest and the prompt-guided data structure reconstruction methods. The second step is O&M LLM fine-tuning, which fine-tunes the Qwen-32B LLM by a constructed O&M hybrid dataset. The third step is knowledge enhancement for the fine-tuned LLM by constructing an external O&M knowledge base and connecting it to the model using the retrieval-augmented generation (RAG) technique. Additionally, a search algorithm is proposed to integrate the O&M knowledge graph (KG), enabling the retrieval of pertinent O&M information. The fourth step is alignment fine-tuning for the knowledge enhanced-LLM which utilizes human feedback to further promote its interactive effectiveness. Experiments show that the proposed paradigm achieved execution accuracy rates of 95.5% in the O&M information query evaluation and it outperforms state-of-the-art (SOTA) LLMs including GPT-4, Llama-70B, CodeLlama-70B. Feedback from practical use and surveys by twenty O&M engineers and senior management confirm the interactivity and practicality of this paradigm in real-world smart building scenarios.

关键词： Building operation and maintenance Information query Large language model Knowledge graph Retrieval-augmented generation

来源：评论

学校读者我要写书评

暂无评论

Advances in neural text generation: A systematic review (2022-2024) 7

Advances in neural text generation: A systematic review (202...

引用

7th workshop for Young Scientists in Computer Science and Software Engineering, CS and SE@SW 2024

作者： Slobodianiuk, Artem V. Semerikov, Serhiy O. Kryvyi Rih State Pedagogical University 54 Universytetskyi Ave. Kryvyi Rih50086 Ukraine Institute for Digitalisation of Education The NAES of Ukraine 9 M. Berlynskoho Str. Kyiv04060 Ukraine Zhytomyr Polytechnic State University 103 Chudnivsyka Str. Zhytomyr10005 Ukraine Kryvyi Rih National University 11 Vitalii Matusevych Str. Kryvyi Rih50027 Ukraine Academy of Cognitive and Natural Sciences 54 Universytetskyi Ave. Kryvyi Rih50086 Ukraine

Recent years have witnessed significant advancements in neural text generation driven by the emergence of large language models and growing interest in this field. This systematic review aims to identify and summarize current trends, approaches, and methods in neural text generation from 2022 to 2024, complementing the findings of a previous review covering 2015-2021. Following the PRISMA methodology, 43 articles were selected from the Scopus database for analysis. The review reveals a shift towards innovative model architectures like Transformer-based models (GPT-2, GPT-3, BERT), attention mechanisms, and controllable text generation. While BLEU, ROUGE, and human evaluation remain the most popular evaluation metrics, new metrics like BERTScore have emerged. Datasets span diverse domains and data types, with growing interest in unlabeled data. Applications have expanded to areas such as table-to-text generation, knowledge graph-based generation, and medical text generation. Although English dominates, there is increasing research on low-resource languages. The findings highlight the rapid evolution of neural text generation methods, the broadening of application areas, and promising avenues for future research. © 2025 Copyright for this paper by its authors.

关键词： applications datasets deep learning evaluation metrics low-resource languages natural language processing neural text generation systematic review

来源：评论

学校读者我要写书评

暂无评论

Evaluating LLMs Capabilities at natural language to Logic Translation: A Preliminary Investigation 6

Evaluating LLMs Capabilities at Natural Language to Logic Tr...

引用

6th International workshop on Artificial Intelligence and Formal Verification, Logic, Automata, and Synthesis, OVERLAY 2024

作者： Brunello, Andrea Ferrarese, Riccardo Geatti, Luca Marzano, Enrico Montanari, Angelo Saccomanno, Nicola University of Udine Italy G.A.P. Srl Italy

Translating natural language (NL) into logical formalisms like first-Order Logic (FOL) has long been a challenge across multiple disciplines, including mathematics, computer science, and education. Traditional computational linguistics methods have struggled with this task due to the complexity and ambiguity of natural language. However, advancements in natural language processing (NLP), particularly the introduction of Large language Models (LLMs), have opened up new possibilities for tackling this challenge. Despite their potential, a systematic approach to evaluating the performance of LLMs in NL-to-FOL translation is still lacking. In this study, we take a first step towards filling in this gap. We examine a large dataset based on students’ efforts in formalizing natural language statements from the book "language, Proof, and Logic". based on this dataset, we propose a preliminary evaluation pipeline to assess LLM performance in NL-to-FOL translation tasks, considering both syntactic and semantic aspects. We then apply this pipeline to evaluate two recent LLMs, Meta’s Llama 3.1 (8B) and Google DeepMind’s Gemma 2 (9B). Our findings validate the proposed approach, revealing key similarities and differences between LLM-generated and student-produced formulas, and provide valuable insights into the current capabilities of LLMs in this domain. © 2024 Copyright for this paper by its authors.

关键词： Computer circuits

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：