Leveraging ChatGPT and Long Short-Term Memory in Recommender Algorithm for Self-Management of Cardiovascular Risk Factors

Abstract

One of the new trends in the development of recommendation algorithms is the dissemination of their capabilities to support the population in managing their health, in particular cardiovascular health. Cardiovascular diseases (CVDs) affect people in their prime years and remain the main cause of morbidity and mortality worldwide, and their clinical treatment is expensive and time consuming. At the same time, about 80% of them can be prevented, according to the World Federation of Cardiology. The aim of this study is to develop and investigate a knowledge-based recommender algorithm for the self-management of CVD risk factors in adults at home. The proposed algorithm is based on the original user profile, which includes a predictive assessment of the presence of CVD. To obtain a predictive score for CVD presence, AutoML and LSTM models were studied on the Kaggle dataset, and it was shown that the LSTM model, with an accuracy of 0.88, outperformed the AutoML model. The algorithm recommendations generated contain items of three types: targeted, informational, and explanatory. For the first time, large language models, namely ChatGPT-3.5, ChatGPT-4, and ChatGPT-4.o, were leveraged and studied in creating explanations of the recommendations. The experiments show the following: (1) In explaining recommendations, ChatGPT-3.5, ChatGPT-4, and ChatGPT-4.o demonstrate a high accuracy of 71% to 91% and coherence with modern official guidelines of 84% to 92%. (2) The safety properties of ChatGPT-generated explanations estimated by doctors received the highest score of almost 100%. (3) On average, the stability and correctness of the GPT-4.o responses were more acceptable than those of other models for creating explanations. (4) The degree of user satisfaction with the recommendations obtained using the proposed algorithm was 88%, and the rating of the usefulness of the recommendations was 92%.

Keywords

• large language model
• multidimensional recommendation
• predictive assessment
• deep machine learning
• official guidelines