Abstract

In modern natural language processing, it is still difficult to extract entity and semantic links from biomedical literature, such as drug-drug, drug-gene, drug-test, drug-disease, drug-herb, drug-food and drug-lab range interactions. In this work XLNet, a large language model based on transformers, is finetuned with Bayesian network that have been improved by the Quantum Approximate Optimization Algorithm (QAOA) by using directed acyclic graphs (DAGs) and conditional probability tables (CPTs) to model complicated biomedical interactions. This work combines the ability of XLNet to capture two-way context with the ability of Bayesian networks to use probabilistic reasoning. QAOA improves computing performance by allowing scalable inference on big datasets. Ranked on a benchmark biomedical dataset, our strategy exceeded current techniques with a 94% accuracy in relationship extraction. With consistency, practical data from unstructured texts, this development improves the accuracy and interpretability of scientific findings, hence enabling drug discovery and personalized treatment.

Statistics