EduRABSA_SLM_v1_SLERP_phi4mini: Education-Domain Opinion Mining

The EduRABSA_SLM_v1_SLERP_phi4mini model is part of the EduRABSA_SLM family, developed by Yan Cathy Hua and colleagues. This model is a fine-tuned multi-task small language model (SLM) specifically engineered for resource-efficient opinion mining within education-domain reviews, such as student course or teaching evaluations and open-ended survey responses. It is built upon the Phi4-mini-instruct base model.

Key Capabilities

This model is adept at various fine-grained Aspect-based Sentiment Analysis (ABSA) tasks, extracting detailed opinion information from review entries. Its capabilities include:

• Opinion Extraction (OE)
• Aspect-Opinion Pair-Extraction (AOPE)
• Aspect-opinion Categorisation (AOC) (also known as ASC with opinion term)
• Aspect-(opinion)-Sentiment Triplet Extraction (ASTE)
• Aspect-(opinion-category)-Sentiment Quadruplet Extraction (ASQE)

Model Development

The EduRABSA_SLM_v1_SLERP_phi4mini model was created by weight-merging two LoRA fine-tuned models using the SLERP algorithm on the Phi4-mini-instruct base. The LoRA adaptors were fine-tuned using the specialized EduRABSA dataset. Full details on its development and performance are available in the paper "Data-Efficient Adaptation and a Novel Evaluation Method for Aspect-based Sentiment Analysis".

Good For

• Analyzing student feedback and evaluations.
• Extracting specific opinions and sentiments related to courses, teaching staff, and universities.
• Researchers and developers working on sentiment analysis in educational contexts.
• Applications requiring fine-grained ABSA on domain-specific text with resource efficiency.