Can AI Tell More than the Available Abstracts?

The word limitation of abstracts often prevents them from including many important scientific discoveries and technology applications that are fully explained in the corresponding full-text articles. This paper investigates whether some information embedded in the full-text articles can be deduced from the abstracts. Specifically, we explore the use of artificial intelligence (AI) to predict if certain experimental procedures or technologies, described in the Materials/Methods section of full-text articles, can be inferred based solely on the available abstracts. Using Pubmed Central (PMC) full-text articles as the data sample, we established a natural language processing (NLP) model and employed a pretrained distilled BERT model (DistilBERT). While the NLP model failed due to overfitting, the DistilBERT model achieved approximately 80% accuracy in predicting both xenograft procedures and lentiviral technology. Further work is underway to improve the DistilBERT model performance further.