检索结果-内蒙古大学图书馆

Conference on Empirical methods in natural language processing (EMNLP)

作者： Shi, Qi Zhang, Yu Yin, Qingyu Liu, Ting Harbin Inst Technol Res Ctr Social Comp & Informat Retrieval Harbin Peoples R China

ISBN: (纸本)9781955917094

Table-based fact verification task aims to verify whether the given statement is supported by the given semi-structured table. Symbolic reasoning with logical operations plays a crucial role in this task. Existing methods leverage programs that contain rich logical information to enhance the verification process. However, due to the lack of fully supervised signals in the program generation process, spurious programs can be derived and employed, which leads to the inability of the model to catch helpful logical operations. To address the aforementioned problems, in this work, we formulate the table-based fact verification task as an evidence retrieval and reasoning framework, proposing the Logic-level Evidence Retrieval and graph-based Verification network (LERGV). Specifically, we first retrieve logic-level program-like evidence from the given table and statement as supplementary evidence for the table. After that, we construct a logic-level graph to capture the logical relations between entities and functions in the retrieved evidence, and design a graph-based verification network to perform logic-level graph-based reasoning based on the constructed graph to classify the final entailment relation. Experimental results on the large-scale benchmark TABFACT show the effectiveness of the proposed approach(1).

关键词： graphic methods

来源：评论

学校读者我要写书评

暂无评论

A Survey on Extracting Knowledge graphs by Employing natural language processing

A Survey on Extracting Knowledge Graphs by Employing Natural...

引用

International Conference on Web Research (ICWR)

作者： Reza Damirchi Hadi Saboohi Amineh Amini Department of Computer Engineering Karaj Branch Islamic Azad University Karaj Iran

ISBN: (数字)9798350394986

ISBN: (纸本)9798350394993

We live in a world of information, and with the ever-increasing rate of content growth, we have no choice but to use machine-based solutions to manage, classify, and use it. However, the produced content is often unstructured, and for computer systems to be able to read and process it, there is no choice but to convert it to structured text. Knowledge graph is one of the ways to represent data in a structured way, and its roots are found in the semantic web. There are different approaches to creating a knowledge graph, but given the significant advances in artificial intelligence, the use of natural language processing is one of the most accurate and automatic ways to do this. In this paper, an attempt has been made to review the latest methods proposed for creating a knowledge graph based on natural language processing. This first allows us to have a better understanding of this research area; in addition, the approaches used by researchers in this field are reviewed and the pros and cons of these approaches are identified. Also, the existing research areas for further work and even the initial ideas for new creative solutions are provided for researchers.

关键词： Surveys Knowledge engineering Reviews Training data Knowledge graphs Production Transformers

来源：评论

学校读者我要写书评

暂无评论

Visual Question Answering Model based on CAM and GCN 5

Visual Question Answering Model Based on CAM and GCN

引用

5th International Conference on Artificial Intelligence and Pattern Recognition, AIPR 2022

作者： Wen, Ping Li, Ma Zhen, Zhang Ze, Wang Xi'an University of Posts and Telecommunications Shaanxi Provincial Key Laboratory of Network Data Analysis and Intelligent Processing China

ISBN: (纸本)9781450396899

Visual Question Answering (VQA) is a challenging problem that needs to combine concepts from computer vision and natural language processing. In recent years, researchers have proposed many methods for this typical multimodal problem. Most existing methods use a two-stream strategy, i.e., compute image and question features separately and fuse them using various techniques, rarely relying on higher-level image representations, to capture semantic and spatial relationships. based on the above problems, a visual question answering model (CAM-GCN) based on Cooperative Attention Mechanism (CAM) and graph Convolutional Network (GCN) is proposed. first, the graph learning module and the concept of graph convolution are combined to learn the problem-specific graph representation of the input image and capture the interactive image representation of the specific problem. Image region dependence, and finally, continue to optimize the fused features through feature enhancement. The test results on the VQA v2 dataset show that the CAM-GCN model achieves better classification results than the current representative models. © 2022 ACM.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

On the Relationship Between RNN Hidden-State Vectors and Semantic Structures 62

On the Relationship Between RNN Hidden-State Vectors and Sem...

引用

62nd Annual Meeting of the Association-for-Computational-Linguistics (ACL) / Student Research workshop (SRW)

作者： Muskardin, Edi Tappler, Martin Pill, Ingo Aichernig, Bernhard K. Pock, Thomas Graz Univ Technol Silicon Austria Labs SAL DES Lab Graz Austria Graz Univ Technol Inst Software Technol Graz Austria Vienna Univ Technol Inst Comp Engn Vienna Austria Graz Univ Technol Inst Comp Graph & Vis Graz Austria

ISBN: (纸本)9798891760998

We examine the assumption that hidden-state vectors of recurrent neural networks (RNNs) tend to form clusters of semantically similar vectors, which we dub the clustering hypothesis. While this hypothesis has been assumed in RNN analyses in recent years, its validity has not been studied thoroughly on modern RNN architectures. We first consider RNNs that were trained to recognize regular languages. This enables us to draw on perfect ground-truth automata in our evaluation, against which we can compare the RNN's accuracy and the distribution of the hidden-state vectors. Then, we consider context-free languages to examine if RNN states form clusters for more expressive languages. For our analysis, we fit (generalized) linear models to classify RNN states into automata states and we apply different unsupervised clustering techniques. With a new ambiguity score, derived from information entropy, we measure how well an abstraction function maps the hidden state vectors to abstract clusters. Our evaluation supports the validity of the clustering hypothesis for regular languages, especially if RNNs are well-trained, i.e., clustering techniques succeed in finding clusters of similar state vectors. However, the clustering accuracy decreases substantially for contextfree languages. This suggests that clustering is not a reliable abstraction technique for RNNs used in tasks like natural language processing.

关键词： Context free languages

来源：评论

学校读者我要写书评

暂无评论

Computational models of relations in text and knowledge graphs for logical reasoning and graph-text conversion

Computational models of relations in text and knowledge grap...

引用

作者： Martin Schmitt University of Munich

学位级别：博士

Knowledge graphs (KGs) store facts in the form of triples that contain two entities and the relation between them. This relation is usually a standardized part of the KG schema and could thus be denoted by any arbitrary unique identifier, e. g., an integer. Typically, however, each relation is assigned a short phrase, such as born in, to describe its meaning intuitively. In this way, the triple resembles a natural language statement although it often lacks the fluency of a real sentence. It is common to reason about facts from KGs and texts to infer new knowledge. An important yet underexplored setting for this is relation inference in context (RIC). Here, the validity of a triple (e 1, r, e 2 ) has to be determined based only on a single other relation r 0 that is known to hold for e 1, e 2. After disambiguating the relations r, r 0 with the context, this boils down to deciding relation inclusion, i. e., logical implication. If the relation arguments e 1, e 2 are not restricted to entities, i. e., we allow for arbitrary noun phrases, and if the relations are only sequences that lead to wellformed sentences, then RIC is also a fundamental task for many use cases in natural language processing, such as question answering, event coreference, and recognizing textual entailment. This thesis examines relations in texts and KGs in two settings: Relation inference and text-graph conversion. The latter builds the bridge between KGs and texts by improving KG interpretability (graph-to-text) and making textual facts accessible to machines (text-to-graph). In the first publication, we extract a large-scale KG from text with lemmatized dependency paths as relations and create a human-annotated RIC benchmark from it. We evaluate and analyze existing methods and propose several new methods based on argument-relation co- occurrence information. The second paper explores the use of pretrained language models (PLMs) for RIC and investigates the influence of combining them with textu

关键词：

来源：评论

学校读者我要写书评

暂无评论

Data-efficient methods for information extraction

Data-efficient methods for information extraction

引用

作者： Yaseen, Usama Ludwig-Maximilians-Universitat Munchen

学位级别：博士

The structured knowledge representation systems such as knowledge base or knowledge graph can provide insights regarding entities and relationship(s) among these entities in the real-world, such knowledge representation systems can be employed in various natural language processing applications such as semantic search, question answering and text summarization. It is infeasible and inefficient to manually populate these knowledge representation systems. In this work, we develop methods to automatically extract named entities and relationships among the entities from plain text and hence our methods can be used to either complete the existing incomplete knowledge representation systems to create a new structured knowledge representation system from scratch. Unlike mainstream supervised methods for information extraction, our methods focus on the low-data scenario and do not require a large amount of annotated data. In the first part of the thesis, we focused on the problem of named entity recognition. We participated in the shared task of Bacteria Biotope 2019, the shared task consists of recognizing and normalizing the biomedical entity mentions. Our linguistically informed named entity recognition system consists of a deep learning based model which can extract both nested and flat entities; our model employed several linguistic features and auxiliary training objectives to enable efficient learning in data-scarce scenarios. Our entity normalization system employed string match, fuzzy search and semantic search to link the extracted named entities to the biomedical databases. Our named entity recognition and entity normalization system achieved the lowest slot error rate of 0.715 and ranked first in the shared task. We also participated in two shared tasks of Adverse Drug Effect Span detection (English) and Profession Span Detection (Spanish); both of these tasks collect data from the social media platform Twitter. We developed a named entity recognition model whic

关键词：

来源：评论

学校读者我要写书评

暂无评论

Multi-label text classification based on semantic-sensitive graph convolutional network

引用

KNOWLEDGE-based SYSTEMS 2024年 284卷

作者： Zeng, Delong Zha, Enze Kuang, Jiayi Shen, Ying Sun Yat Sen Univ Gongchang Rd 66 Shenzhen 518107 Guangdong Peoples R China

Multi-Label Text Classification (MLTC) is an important but challenging task in the field of natural language processing. In this paper, we propose a novel method, Semantic-sensitive graph Convolutional Network (S-GCN), by simultaneously considering semantic and word-global associations. More specifically, we first leverage texts, words, and labels to construct a global graph, which helps mine the relevance between similar documents. Then we design and pre-train an encoder to initialize text nodes in the graph, from which the semantic features of documents are extracted. Next, we employ a graph convolutional network to classify text nodes, which can well fuse node information. Finally, we normalize the adjacency matrix and store hidden layer representations of word nodes, tackling the issue that conventional graph-based methods cannot predict texts that did not appear during training. We conduct experiments on three public datasets, AAPD, RMSC-V2, and Reuters-21578, and demonstrate the superiority of our model over the baselines on the MLTC task. Source code is available at https://***/sysu18364004/SGCN.

关键词： Multi-label Text classification graph convolutional network

来源：评论

学校读者我要写书评

暂无评论

Document-level Event Extraction via Heterogeneous graph-based Interaction Model with a Tracker 59

Document-level Event Extraction via Heterogeneous Graph-base...

引用

Joint Conference of 59th Annual Meeting of the Association-for-Computational-Linguistics (ACL) / 11th International Joint Conference on natural language processing (IJCNLP) / 6th workshop on Representation Learning for NLP (RepL4NLP)

作者： Xu, Runxin Liu, Tianyu Li, Lei Chang, Baobao Peking Univ Key Lab Computat Linguist MOE Beijing Peoples R China Peng Cheng Lab Shenzhen Peoples R China ByteDance AI Lab Beijing Peoples R China

ISBN: (纸本)9781954085527

Document-level event extraction aims to recognize event information from a whole piece of article. Existing methods are not effective due to two challenges of this task: a) the target event arguments are scattered across sentences;b) the correlation among events in a document is non-trivial to model. In this paper, we propose Heterogeneous graph-based Interaction Model with a Tracker (GIT) to solve the aforementioned two challenges. For the first challenge, GIT constructs a heterogeneous graph interaction network to capture global interactions among different sentences and entity mentions. For the second, GIT introduces a Tracker module to track the extracted events and hence capture the interdependency among the events. Experiments on a large-scale dataset (Zheng et al., 2019) show GIT outperforms the existing best methods by 2.8 F1. Further analysis reveals GIT is effective in extracting multiple correlated events and event arguments that scatter across the document.

关键词： graphic methods

来源：评论

学校读者我要写书评

暂无评论

Cross-System Data Integration based on Rule-based NLP and Node2Vec 8

Cross-System Data Integration Based on Rule-Based NLP and No...

引用

8th International Conference on Data Science in Cyberspace, DSC 2023

作者： Xiao, Jiakai Dut, Wei Xu, Zhengxiang Qian, Yang State Grid Anhui Electric Power Company Department of Internet China School of Management Hefei University of Technology China

ISBN: (纸本)9798350331035

In the era of Big data, a growing number of enterprises are paying attention to data governance and data services. Large enterprises (e.g., state grid corporations), have numerous information systems, which have caused issues such as inconsistent data standards, and different data storage methods. This makes it very difficult to share data information among different enterprises and departments. In this situation, enterprises have begin to use various methods to integrate data across different systems. In this study, we select the power industry and design a cross-system data fusion framework. Specifically, there are many types of power grid equipment, and the topology data, equipment data, and operation data associated with them are stored in different ways. For cross-system data, we first use equipment name information and design a ruled-based natural language processing (NLP) technique to match the same data between different systems. However, the equipment names in different systems do not follow a strict and unified naming standard, which leads to a set of equipment nodes that cannot be accurately matched. To solve this issue, we use network structure information about equipment and design a node representation technique based on Node2vec. based on the node representation, we further optimize the results of equipment matching. For evaluation, we use a cross-system dataset from a state grid corporation in China. Through experiments, we show that the proposed framework performs very well in cross-system data integration. © 2023 IEEE.

关键词： Artificial intelligence Computing methodologies Computing methodologies Data Integration Electricity Industry graph Embedding Information extraction Learning latent representations Machine learning Machine learning approaches naturallanguage processing NLP

来源：评论

学校读者我要写书评

暂无评论

processing Long Legal Documents with Pre-trained Transformers: Modding LegalBERT and Longformer 4

Processing Long Legal Documents with Pre-trained Transformer...

引用

4th natural Legal language processing workshop, NLLP 2022, co-located with the 2022 Conference on Empirical methods in natural language processing, EMNLP 2022

作者： Mamakas, Dimitris Tsotsi, Petros Androutsopoulos, Ion Chalkidis, Ilias Department of Informatics Athens University of Economics and Business Greece Department of Computer Science University of Copenhagen Denmark Cognitiv+ Athens Greece

ISBN: (纸本)9781959429180

Pre-trained Transformers currently dominate most NLP tasks. They impose, however, limits on the maximum input length (512 sub-words in BERT), which are too restrictive in the legal domain. Even sparse-attention models, such as Longformer and BigBird, which increase the maximum input length to 4,096 sub-words, severely truncate texts in three of the six datasets of LexGLUE. Simpler linear classifiers with TF-IDF features can handle texts of any length, require far less resources to train and deploy, but are usually outperformed by pre-trained Transformers. We explore two directions to cope with long legal texts: (i) modifying a Longformer warm-started from LegalBERT to handle even longer texts (up to 8,192 sub-words), and (ii) modifying LegalBERT to use TF-IDF representations. The first approach is the best in terms of performance, surpassing a hierarchical version of LegalBERT, which was the previous state of the art in LexGLUE. The second approach leads to computationally more efficient models at the expense of lower performance, but the resulting models still outperform overall a linear SVM with TF-IDF features in long legal document classification. © 2022 Association for Computational Linguistics.

关键词： Support vector machines

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：