The book covers several entity and relation extraction techniques starting from the traditional feature-based techniques to the recent techniques using deep neural models. Two important focus areas of the book are – i) joint extraction techniques where the tasks of entity and relation extraction are jointly solved, and ii) extraction of complex relations where relation types can be N-ary and cross-sentence. The first part of the book introduces the entity and relation extraction tasks and explains the motivation in detail. It covers all the background machine learning concepts necessary to understand the entity and relation extraction techniques explained later. The second part of the book provides a detailed survey of the traditional entity and relation extraction problems covering several techniques proposed in the last two decades. The third part of the book focuses on joint extraction techniques which attempt to address both the tasks of entity and relation extraction jointly. Several joint extraction techniques are surveyed and summarized in the book. It also covers two joint extraction techniques in detail which are based on the authors’ work. The fourth and the last part of the book focus on complex relation extraction, where the relation types may be N-ary (having more than two entity arguments) and cross-sentence (entity arguments may span multiple sentences). The book highlights several challenges and some recent techniques developed for the extraction of such complex relations including the authors’ technique. The book also covers a few domain-specific applications where the techniques for joint extraction as well as complex relation extraction are applied.
Author(s): Sachin Sharad Pawar, Pushpak Bhattacharyya, Girish Keshav Palshikar
Series: Studies in Computational Intelligence, 1058
Publisher: Springer
Year: 2022
Language: English
Pages: 155
City: Singapore
Preface
Acknowledgements
Contents
About the Authors
1 Introduction
1.1 Entities
1.2 Relations
1.2.1 Global- versus Mention-level Relations
1.3 Motivation
1.4 Research Gaps and Objectives
1.4.1 End-to-end Relation Extraction
1.4.2 N-ary Cross-sentence Relation Extraction
1.5 Organization of the Monograph
References
2 Literature Survey
2.1 Relation Extraction
2.1.1 Feature-based Methods
2.1.2 Kernel Methods
2.1.3 Neural Approaches
2.1.4 Datasets
2.1.5 Evaluation
2.2 Joint Entity and Relation Extraction
2.2.1 Motivating Example
2.2.2 Overview of Techniques
2.2.3 Joint Inference Techniques
2.2.4 Joint Models
2.2.5 Experimental Evaluation
2.3 N-ary Cross-sentence Relation Extraction
2.3.1 Extracting Cross-sentence Relations
2.3.2 Extracting N-ary and Cross-sentence Relations
References
3 Joint Inference for End-to-end Relation Extraction
3.1 Introduction
3.1.1 Problem Definition
3.1.2 Motivation for Joint Extraction
3.2 Background: Markov Logic Networks (MLN)
3.2.1 Basics of First-order Logic
3.2.2 Basics of MLNs
3.2.3 Formal Definition
3.2.4 Inference in MLNs
3.3 Building Blocks for Our Approach
3.3.1 Identifying Entity Mention Candidates
3.3.2 Entity Type Classifier
3.3.3 Entity Type Agnostic Relation Classifier
3.3.4 Pipeline Relation Classifier
3.4 Joint Extraction using Inference in Markov Logic Networks (MLN)
3.4.1 Motivation
3.4.2 Domains and Predicates
3.4.3 Generic Rules
3.4.4 Sentence-specific Rules
3.4.5 Additional Semantic Rules
3.4.6 Joint Inference
3.5 Example
3.6 Experimental Analysis
3.6.1 Limitations of Our Approach
References
4 Joint Model for End-to-End Relation Extraction
4.1 Motivation
4.2 All Word Pairs Model (AWP-NN)
4.2.1 Features for the AWP-NN Model
4.2.2 Architecture of the AWP-NN Model
4.3 Inference Using Markov Logic Networks
4.4 Experimental Analysis
4.4.1 Datasets
4.4.2 Implementation Details
4.4.3 Results
4.4.4 Analysis of Results
4.5 Domain-Specific Entities and Relations
4.5.1 Adverse Drug Reactions
4.5.2 TAC 2017: ADR Extraction Task
References
5 N-ary Cross-Sentence Relation Extraction
5.1 Introduction
5.2 Problem Definition
5.2.1 Comparison with Relevant Past Work
5.3 Proposed Approach
5.3.1 Constructing Sequence Representations
5.3.2 Constrained Subsequence Kernel (CSK)
5.3.3 Formal Definition of CSK
5.3.4 Classifying Candidate Relation Instances
5.4 Experimental Analysis
5.4.1 Datasets
5.4.2 Implementation Details
5.4.3 Analysis of Results and Errors
5.5 Discussion on Decomposition of N-ary Relations
5.5.1 Examples of Various Relation Types
5.5.2 Generalized Theorem
References
6 Recent Advances in Entity and Relation Extraction
6.1 Joint Entity and Relation Extraction
6.1.1 Using Span-Based Representation for Entity Mentions
6.1.2 Using BERT Embeddings
6.2 N-ary Cross-Sentence Relation Extraction
6.2.1 Standard Dataset
6.2.2 Graph Neural Networks
6.2.3 Using BERT Embeddings
References
7 Conclusions
7.1 Summary of the Monograph
7.2 Future Directions
References
Appendix A Foundations
A.1 Maximum Entropy Classifier
A.2 Conditional Random Fields (CRF)
A.3 Markov Logic Networks (MLN)
A.3.1 Basics of First-order Logic
A.3.2 Basics of MLNs
A.3.3 Formal Definition
A.3.4 Example
A.3.5 Inference in MLNs
A.4 Neural Networks
A.4.1 Layers in Neural Networks
A.4.1.1 Embedding Layer (Input Layer)
A.4.1.2 Linear Layer
A.4.1.3 Softmax Layer (Output Layer)
A.4.2 Training Neural Networks
A.4.3 Long Short-Term Memory Networks (LSTM)
A.5 Support Vector Machines (SVM) with Kernels
A.5.1 Kernel Functions
A.5.2 String Subsequence Kernel
A.5.3 Generalized Subsequence Kernel
References
