A Benchmark dataset on Semantic Change in Scholarly Publications on Disability

This is a benchmark dataset for semantic shift detection in disability-related corpora, including collected title and abstract text from PubMed and ArXiv, annotation sets based on domain experts and LLMs, and extracted KGs (Wikidata entity claims). The corpus from PubMed covers the period from the 1900s to 2023, while the corpus from ArXiv covers the period from the 1990s to 2023. The corpus was filtered based on 16 disability-related target words. In the annotation sets, '1' indicates that a semantic shift occurred for a target word, while '0' indicates the opposite. In particular, the LLM-based annotation sets include their generated text, and we used the Llama2 and GPT-4 models. '7b' refers to the parameter size of the Llama2 model. Graph_data.zip contains Wikidata entity claims.