Article
Computer Science, Artificial Intelligence
Yong Shi, Yang Xiao, Pei Quan, MingLong Lei, Lingfeng Niu
Summary: This paper proposes a new DSRE framework A2DSRE, which addresses the issues in distant supervision relation extraction by introducing dependency trees and knowledge graph supervision, effectively reducing data noise and improving the accuracy of relation paths.
Article
Computer Science, Artificial Intelligence
Chong Chen, Tao Wang, Yu Zheng, Ying Liu, Haojia Xie, Jianfeng Deng, Lianglun Cheng
Summary: Fault diagnosis is crucial in operating and maintaining industrial assets. A fault diagnosis knowledge graph can assist engineers in conducting maintenance tasks. However, manually labeling the corpus from multiple sources is time-consuming, and the presence of noisy sentences hampers the performance of relation extraction modeling.
ADVANCED ENGINEERING INFORMATICS
(2023)
Article
Computer Science, Artificial Intelligence
Changsen Yuan, Heyan Huang, Chong Feng
Summary: The Graph Convolutional Network (GCN) is a universal method for relation extraction by capturing sentences' syntactic features, but the quality of dependency parsing affects its performance. The Multi-Graph Cooperative Learning model (MGCL) proposed in this article focuses on extracting reliable syntactic features from different graphs to improve sentence representation, achieving state-of-the-art performance in relation extraction.
ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY
(2021)
Article
Computer Science, Information Systems
Yu-Ming Shang, Heyan Huang, Xin Sun, Wei Wei, Xian-Ling Mao
Summary: Relation ties are critical for distant supervised relation extraction. This study proposes a novel force-directed graph method to comprehensively learn relation ties. By constructing a global co-occurrence graph and introducing the concept of attractive force and repulsive force, the method accurately models the correlation and mutual exclusion between relations. Experimental results show that it outperforms existing baselines and can be used to augment relation extraction systems.
ACM TRANSACTIONS ON INFORMATION SYSTEMS
(2023)
Article
Computer Science, Information Systems
Xiang Ying, Zechen Meng, Mankun Zhao, Mei Yu, Shirui Pan, Xuewei Li
Summary: This paper proposes an enhanced representation method that addresses the issue of wrong labeling in distant supervised relation extraction. By incorporating enhanced representations into a gated graph convolutional network, the proposed method achieves significant improvement on two popular datasets.
WORLD WIDE WEB-INTERNET AND WEB INFORMATION SYSTEMS
(2023)
Article
Chemistry, Multidisciplinary
Xiaoyan Meng, Tonghai Jiang, Xi Zhou, Bo Ma, Yi Wang, Fan Zhao
Summary: This paper introduces a noise detection-based relation extraction approach (NDRE) to automatically detect noisy labels and dynamically correct them, alleviating both instance-level and bag-level noisy problems in distant supervised relation extraction. Experimental results show that the proposed approach achieves significant improvements over existing baselines on a public benchmark dataset.
APPLIED SCIENCES-BASEL
(2021)
Article
Computer Science, Artificial Intelligence
Jing Zhang, Meilin Cao
Summary: This paper proposes hierarchical attention-based networks that can de-noise at both sentence and bag levels. In the calculation of bag representation, we provide weights to sentence representations using sentence-level attention that considers correlations among sentences in each bag. The proposed method has shown significant advantages in relation extraction tasks according to experimental results.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Shirong Shen, Shangfu Duan, Huan Gao, Guilin Qi
Summary: Distant supervision relation extraction (DSRE) aims to train a classifier by aligning knowledge base triples with large-scale corpora, but mislabeled instances and limited interactions between background information pose challenges. This study introduces a novel ER-HG model that leverages five types of background information and an attention mechanism to improve information integration and mitigate noise effects. Experimental results on two datasets show significant performance improvement over state-of-the-art methods in held-out metric and robustness tests.
JOURNAL OF WEB SEMANTICS
(2021)
Article
Computer Science, Artificial Intelligence
Yanru Zhou, Limin Pan, Chongyou Bai, Senlin Luo, Zhouting Wu
Summary: A distant supervision relation extraction method with self-selective attention is proposed in this study, which uses convolution and self-attention mechanism to encode instances, making full use of the correlation information between instances and achieving better results.
Article
Computer Science, Information Systems
Yu-Ming Shang, Heyan Huang, Xin Sun, Wei Wei, Xian-Ling Mao
Summary: This paper introduces a novel distant supervised relation extraction model, which utilizes a pattern-aware self-attention network to discover relational patterns and process them in pre-trained Transformers. By applying the probability distribution as a constraint in the first Transformer layer, fine-grained pattern information in the pre-trained Transformer is enhanced without sacrificing global dependencies.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Qiji Zhou, Yue Zhang, Donghong Ji
Summary: We explore distantly supervised relation extraction using knowledge-guided latent graphs and an iterative graph learner. Input sentences are assumed to contain latent graphs with useful structural information linking mentions and relations in the distantly supervised data. These graphs are embedded into input sentences with default initialized graphs. Pre-trained Knowledge Bases (KB) are also utilized to guide the latent space of a Variational Graph Auto-Encoders (VGAE) module. The iterative graph learner optimizes the latent graph structures and their corresponding node embeddings, leading to improved performance in downstream relation extraction tasks.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Chemistry, Multidisciplinary
Xuxin Chen, Xinli Huang
Summary: Distant supervision for relation extraction (DSRE) automatically acquires annotated data by aligning corpus with knowledge base, reducing manual annotation cost, but it is affected by noisy data. This paper introduces negative training to filter out noisy data and improves model performance by using entity attributes.
APPLIED SCIENCES-BASEL
(2022)
Article
Computer Science, Artificial Intelligence
Meizhen Liu, Fengyu Zhou, Jiakai He, Xiaohui Yan
Summary: Distant supervision neural relation extraction faces the challenges of multi-labels and long-tail distribution. Existing studies mainly focus on developing attention mechanism to reduce noise, but fail to explore the semantic correlations between entity pairs and context. This paper proposes a Knowledge Graph ATTention (KGATT) mechanism to address these issues, which consists of two modules: fine-alignment and inductive. With the mutual reinforcement of these two modules, our model enriches the representation of instance bags, improves overall performance, and mitigates the long-tail phenomenon.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Yan Xiao, Yaochu Jin, Ran Cheng, Kuangrong Hao
Summary: This study introduces a new framework for relation extraction using Transformer block and multi-instance learning, which effectively addresses the challenges in relation extraction. Experimental results demonstrate that the proposed approach outperforms state-of-the-art algorithms on the selected dataset.
Article
Computer Science, Artificial Intelligence
Jiasheng Wang, Qiongxin Liu
Summary: Distant supervision reduces manual labor by automatically labeling data. Existing relation extraction methods under distant supervision ignore the repetition of entity pairs in sentences and the varying noise between sentence bags. A novel method with position feature attention and selective bag attention is proposed to address these issues, demonstrating effectiveness in experimental results.
Article
Automation & Control Systems
Giuseppe Cuccu, Julian Togelius, Philippe Cudre-Mauroux
Summary: The proposed method allows for learning compact state representations and policies separately and simultaneously for policy approximation in vision-based applications. By separating image processing from action selection, better understanding of each task and potentially finding smaller policy representations is possible. The study uses a compact encoder and two novel algorithms for state representation learning, along with a new evolutionary algorithm to address the issue of producing large inputs as the dictionary size increases.
AUTONOMOUS AGENTS AND MULTI-AGENT SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Akansha Bhardwaj, Albert Blarer, Philippe Cudre-Mauroux, Vincent Lenders, Boris Motik, Axel Tanner, Alberto Tonon
Summary: Microblogging services like Twitter are important sources of information. We proposed a successful system ArmaTweet for semantic event detection on Twitter streams, and explored and empirically evaluated various approaches for event detection on microposts, with results showing ArmaTweet outperforming other methods in most cases and a combined approach offering highest recall without adversely affecting precision.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2021)
Article
Computer Science, Information Systems
Ruslan Mavlyutov, Philippe Cudre-Mauroux
Summary: This paper introduces CINTIA, an efficient data structure for storing and querying interval data, achieving high memory locality and outperforming current solutions. Additionally, a low-latency Big Data system is proposed to implement CINTIA on popular distributed file systems and manage large interval data effectively in clusters of commodity machines.
IEEE TRANSACTIONS ON BIG DATA
(2021)
Article
Computer Science, Artificial Intelligence
Dingqi Yang, Bingqing Qu, Jie Yang, Liang Wang, Philippe Cudre-Mauroux
Summary: SGSketch is a highly-efficient streaming graph embedding technique that can generate high-quality node embeddings from a streaming graph by gradually forgetting outdated streaming edges, and efficiently update the generated node embeddings via an incremental embedding updating mechanism.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Editorial Material
Computer Science, Hardware & Architecture
Philippe Cudre-Mauroux
COMMUNICATIONS OF THE ACM
(2022)
Article
Computer Science, Information Systems
Alberto Lerner, Matthias Jasny, Theo Jepsen, Carsten Binnig, Philippe Cudre-Mauroux
Summary: Modern DBMS engines achieve unprecedented transaction processing speeds, but benchmarking networked clients remains challenging. This demo presents a new framework that leverages hardware-software co-design for benchmarking.
PROCEEDINGS OF THE VLDB ENDOWMENT
(2022)
Proceedings Paper
Computer Science, Cybernetics
Giuseppe Cuccu, Luca Rolshoven, Fabien Vorpe, Philippe Cudre-Mauroux, Tobias Glasmachers
Summary: DiBB is a meta-algorithm and framework that addresses the scalability issue of Black-Box Optimization (BBO). It creates a Partially Separable (PS) version of any existing black-box algorithm and distributes the computation to a set of machines while retaining the advanced features of the underlying algorithm. The performance of DiBB scales with the number of parameter-blocks defined.
PROCEEDINGS OF THE 2022 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE (GECCO'22)
(2022)
Article
Computer Science, Artificial Intelligence
Dingqi Yang, Bingqing Qu, Jie Yang, Philippe Cudre-Mauroux
Summary: In this paper, the authors propose LBSN2Vec++, a heterogeneous hypergraph embedding approach designed specifically for LBSN data. By automatically learning features, LBSN2Vec++ outperforms other methods and handcrafted features in friendship and location prediction tasks.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Laura Rettig, Shaban Shabani, Loris Sauter, Philippe Cudre-Mauroux, Maria Sokhn, Heiko Schuldt
Summary: The City-Stories system combines entity linking, multimedia retrieval, and crowdsourcing to make historical images searchable even across collections.
WEB ENGINEERING, ICWE 2021
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Vibhav Agarwal, Akansha Bhardwaj, Paolo Rosso, Philippe Cudre-Mauroux
Summary: Ad-hoc table retrieval involves finding tables relevant to a search query, with existing research utilizing various methods to address this issue, but most methods face challenges in constructing semantic representations of tabular data. Researchers propose the ConvTab method based on Convolutional Neural Networks, which achieves remarkable results in table retrieval tasks.
2021 IEEE INTERNATIONAL CONFERENCE ON BIG DATA (BIG DATA)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Michael Luggen, Julien Audiffren, Djellel Difallah, Philippe Cudre-Mauroux
Summary: Wikidata is becoming an important resource for various online tasks, and the Wiki2Prop method proposed in this work significantly outperforms competitors in predicting missing properties for entities, with the ability to incorporate multilingual and multimodal data to further improve prediction accuracy. This system provides a valuable tool for filling knowledge gaps in Wikidata and can be used as a property recommender system directly on Wikidata entity pages.
PROCEEDINGS OF THE WORLD WIDE WEB CONFERENCE 2021 (WWW 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Ines Arous, Jie Yang, Mourad Khayati, Philippe Cudre-Mauroux
Summary: The increase in paper submissions challenges the effectiveness of scientific peer review, prompting the proposal of a human-AI approach to estimate review conformity. Through a large-scale crowdsourced study, this approach has shown superior performance and easy integration into existing peer review systems, reducing the burden on meta-reviewers.
PROCEEDINGS OF THE WORLD WIDE WEB CONFERENCE 2021 (WWW 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Paolo Rosso, Dingqi Yang, Natalia Ostapuk, Philippe Cudre-Mauroux
Summary: This paper focuses on an instance completion task of suggesting relation and tail entities for a given head entity, proposing an end-to-end solution called RETA which outperforms existing techniques. RETA-Filter generates high-quality candidate pairs, while RETA-Grader significantly outperforms state-of-the-art link prediction techniques in the instance completion task.
PROCEEDINGS OF THE WORLD WIDE WEB CONFERENCE 2021 (WWW 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Ines Arous, Ljiljana Dolamic, Jie Yang, Akansha Bhardwaj, Giuseppe Cuccu, Philippe Cudre-Mauroux
Summary: The study introduces a hybrid human-AI approach that incorporates human rationales into attention-based text classification models to enhance the explainability of classification results. MARTA framework jointly learns model and worker reliability, leading to improved classification explainability and accuracy.
THIRTY-FIFTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, THIRTY-THIRD CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE AND THE ELEVENTH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE
(2021)
Article
Computer Science, Information Systems
Theo Jepsen, Alberto Lerner, Fernando Pedone, Robert Soule, Philippe Cudre-Mauroux
Summary: Transaction Triaging is a set of techniques that manipulate streams of transaction requests and responses to improve performance when reaching the database server. The algorithms do not interfere with transaction execution and can be easily ported across different database systems. Leveraging programmable networking hardware, triaging brings significant throughput improvements in high-overhead network stacks.
PROCEEDINGS OF THE VLDB ENDOWMENT
(2021)