Feature based methods design lexical, syntactic, and semantic features for entity pairs and their corresponding context. These features are input into relation classifiers.Many efforts are required to design features, which is a challenge.

Feature-based methods of RE

Kernel-based methods design kernel functions for support vector machines to measure the similarities between relation representations and textual instances. Many efforts are required to design kernel functions, which is a challenge. 


Kernel-based methods of RE

Graphical methods abstract the dependencies between entities, text, and relations in the form of directed acyclic graphs or DAGs and then use inference models to identify the correct relations. 

Graphical methods of RE

There are also efforts in encoding text into low-dimensional semantic spaces and extracting relations from textual embeddings. Furthermore, knowledge graph embeddings can be used for RE.

Embedding models for RE

SREs reduce human efforts and bring better coverage than pattern rules. 
The following are types of SREs:
- Feature-based methods
- Kernel-based methods
- Graphical methods
- Embedding models

Columbia University

The following approaches have been explored for relation extraction, and these methods have bridged the gap between unstructured text and structured knowledge. 
- Pattern mining
- Statistical approaches 
- Deep learning

Pioneering Explorations of Relation Extraction

Sentence analysis tools are used to identify syntactic elements in text and then automatically construct pattern rules from these elements. To extract patterns with better coverage and accuracy, larger corpora, more formats of patterns, and more efficient ways of extraction have been explored by later work in pattern extraction. Automatically extracted patterns may have mistakes so these methods require examination from human experts. 

Pattern extraction methods


Statistical Relation Extraction Models 

Most RE methods work in a simplified setting, and mainly focus on training models with large amounts of human annotations to classify two given entities within one sentence into pre-defined relations. To build an effective and robust RE system that can be deployed in the real world, more complex scenarios must be further investigated.  
- Collecting high-quality human annotation is expensive and time-consuming
- Many long-tail relations cannot provide large amounts of training examples 
- Most facts are expressed by long context consisting of multiple sentences
- Using a predefined set to cover those relations with open-ended growth is not easy

Drawbacks of existing RE methods


NRE models introduce neural networks to automatically extract semantic features from text, and can effectively capture textual information and generalize to a wider range of data. NRE studies mainly focus on designing and using different network architectures to capture relational semantics within text. NRE differs from SREs by using word embeddings and position embeddings as inputs so that semantic meanings of words can be combined with a specification of the relative distances between words and entities. 


Learn Before

Related

Learn After