Intelligent HIV/AIDS FAQ information retrieval system using neural networks

Abstract

HIV/AIDS has no cure to this date and it has been noted that the most effective means of mitigating the disease infectious rate is information sharing. HIV/AIDS Frequently Asked Questions (HIV/AIDS FAQ) is another approach for sharing information. The research proposes an automated FAQ information retrieval system for sharing HIV/AIDS FAQs. One of the challenges in FAQ retrieval is mapping of a user query in an FAQ retrieval system to appropriate FAQ question in the FAQ retrieval system repository. To address this challenge a number of approaches have been proposed, most of which are based on traditional information retrieval techniques. The goal of this research is to design and implement an artificial neural network retrieval system to experiment mapping of arbitrary HIV/AIDS FAQ user question to similar in meaning or equivalent HIV/AIDS FAQ question stored in the FAQ retrieval system repository. Question to question similarity matching technique shall be used. System performance is benchmarked with traditional information retrieval (key word based) HIV/AIDS FAQ retrieval system. Golden standard approach is used to judge system efficiency using rejection rate and recall rate metrics. The study compiled an HIV/AIDS FAQ corpus. The Intelligent HIV/AIDS FAQ retrieval system (IHAFR) operational parameters were designed based on heuristics rules and experimental determinants. A portion of the HIV/AIDS FAQ corpus was used to train the IHAFR using MATLAB. Unknown HIV/AIDS FAQ questions were posed to the systems and performance benchmarked with traditional keyword based HIV/AIDS FAQ system. HIV/AIDS counselors, students participating in HIV/AIDS organized activities evaluated the performance of these systems. The analysis results revealed that IHAFR had recall rate 79.17% and traditional keyword based FAQ system 55.83% for equivalent or similar HIV/AIDS FAQs. Traditional keyword based FAQ retrieval system attained a rejection rate of 82.50%, compared to 61.67 % for neural network system. Based on these general results, the research concludes that neural network systems have a better ability to provide alternative FAQ questions which are semantically similar because of the neural network generalization trait. In contrast, key word based retrieval systems recall rate are poor because they do syntactical similarity matching, however this same trait gives them a better rejection rate. Due to the generalization ability of the neural network approach, it could be an ideal technique for implementing HIV/AIDS FAQ retrieval as it semantically provides related FAQ question and therefore an answer.