AI Memory Assistant for Alzheimer's Patients with LLM and Graph-Based Approach vs LSTM for Recall Accuracy

Aim: In this research, an AI Memory Assistant is being designed for the purpose of helping Alzheimer's patients through a Large Language Model (LLM) in conjunction with a graph-based approach, and will additionally be compared to a previously developed Long Short-Term Memory (LSTM) based approach. Materials and Methods: The outcomes of this research will involve two groups: Group 1 will be an experimental group consisting of individuals who will utilize an existing LSTM memory assistant which has been evaluated using 15 subjects and has a mean recall accuracy of 0.71, and Group 2 will involve utilizing the proposed LLM + Graph based AI Memory Assistant that will be assessed against a set of four metrics (accuracy, precision, recall and F1-Score). The sample size of both groups was selected using information from prior research and has an expected amount of 80% test power, 0.05 significance level and 95% confidence interval. Statistical methods were applied using the SPSS 26.0 software program and the independent samples t-test was utilized for the comparison of both groups. Result: The results indicate that the proposed LLM + Graph based AI Memory Assistant outperforms the LSTM, CNN-LSTM and Transformer memory assistants (92.8% accuracy, 92.4% recall, 93.1% precision, 92.7% F1-Score,$\mathrm{p}<0.001$). Conclusion: These results indicate that the proposed LLM + Graph based Memory Assistant provides a superior amount of recall accuracy and improved levels of personalized support and contextual awareness as compared to existing LSTM based systems and therefore provides a more effective means of assisting Alzheimer's patients and their caregivers.