Some methods adopt multi-instance learning by combining sentences with the same entity pairs and then selecting informative instances from them using graphical models or heuristic strategies. Attention mechanisms can also be designed to highlight informative instances.


Selecting informative instances for denoising DS

To denoise DS data, incorporating extra context information has been explored by incorporating knowledge graphs as external information to guide instance selection and adopting multilingual corpora for the information consistency and complementarity.

Incorporating extra context information for denoising DS

Sophisticated mechanisms and training strategies can be used to enhance distinctly supervised NRE models. Different architectures and training strategies can be combined to construct hybrid frameworks. A soft-label scheme can be adopted by changing unconfident labels during training. Reinforcement learning and adversarial training can also be adopted in DS.


Sophisticated mechanisms and training strategies for denoising DS

To alleviate the noise problem associated with distant supervision for relation extraction, existing methods can be grouped into three major approaches: selecting informative instances, incorporating extra context information, and utilizing sophisticated mechanisms and training strategies.

Columbia University

Google

Supervised NRE models have a lack of large-scale high-quality training data because manually labeling data is time-consuming and human intensive. To help with this, distant supervision (DS) assumption has been used to automatically label data by aligning existing knowledge graphs with plain text. For any entity pair in knowledge graphs, sentences mentioning both entities will be labeled with their corresponding relations in knowledge graphs. Large-scale training examples can be easily constructed using this training scheme. Although DS provides a feasible approach to use more data, it is inevitable for its automatic labeling mechanism to be accompanied with the wrong labeling problem, since not all sentences mentioning two given entities express their relations in knowledge graphs exactly. 

Learn Before

Related

Learn After