Update README.md

README.md

@@ -581,15 +581,19 @@ size_categories:
 
 # QAPyramid
 
-This repo contains the data from our paper "QAPyramid: Fine-grained Evaluation of Content Selection
-for Text Summarization".
+This repo contains the data from our paper "QAPyramid: Fine-grained Evaluation of Content Selection for Text Summarization".
 
 Please visit [here](https://github.com/ZhangShiyue/QAPyramid) for more details of this project.
 
 QAPyramid is built on top of 500 examples from the test set of the CNNDM English news summarization dataset.
 
-Humans (crowdsourced workers) decomposed each reference summary into QA pairs. 
-
+Humans (crowdsourced workers) decomposed each reference summary into QA pairs following the QA-SRL framework. 
 
+On a 50-example subset, we get model-generated summaries from 10 summarization systems. 5 of them are models finetuned on the CNNDM training 
+set (BART, PEGASUS, BRIO, BRIO-EXT, MatchSum), and the other 5 models are 1-shot LLMs 
+(Llama-3-8b-instruct, Llama-3-70b-instruct, Mixtral-8*7b-instruct, Mixtral-8\*22b-instruct, GPT4).
 
+Humans (crowdsourced workers) labeled whether each QA pair is *present* (1) or *not present* (0) in the system summary. 
+*present* means the meaning of the QA pair is covered or can be inferred from the system summary.  
 
+System scores are the scores of each system summary for different metrics.