Efficient Information Extraction in Few-Shot Relation Classification through Contrastive Representation Learning

Anonymous

Abstract: Differentiating relationships between entity pairs with limited labeled instances poses a significant challenge in few-shot relation classification. Representations of textual data extract rich information spanning the domain, entities, and relations. In this paper, we introduce a novel approach to enhance information extraction using multiple noisy representations and contrastive learning. While sentence representations in relation classification commonly combine information from entity marker tokens, we argue that substantial information within the internal model representations remains untapped. To address this, we propose aligning multiple noisy sentence representations, such as the *[CLS]* token, the *[MASK]* token used in prompting, and entity marker tokens. We employ contrastive learning to reduce the noise contained in the individual representations. We demonstrate the adaptability of our representation contrastive learning approach, showcasing its effectiveness for both sentence representations and additional data sources, such as relation description representations. Our evaluation underscores the efficacy of incorporating multiple noisy representations through contrastive learning, enhancing information extraction in settings where available data is limited. Our model is available at https://anonymous.4open.science/r/MultiRep-6E39.

Paper Type: short

Research Area: Information Extraction

Contribution Types: Approaches to low-resource settings

Languages Studied: English

Consent To Share Submission Details: On behalf of all authors, we agree to the terms above to share our submission details.