HHH: An Online Medical Chatbot System based on Knowledge Graph and Hierarchical Bi-Directional Attention (2002.03140v1)

Abstract: This paper proposes a chatbot framework that adopts a hybrid model which consists of a knowledge graph and a text similarity model. Based on this chatbot framework, we build HHH, an online question-and-answer (QA) Healthcare Helper system for answering complex medical questions. HHH maintains a knowledge graph constructed from medical data collected from the Internet. HHH also implements a novel text representation and similarity deep learning model, Hierarchical BiLSTM Attention Model (HBAM), to find the most similar question from a large QA dataset. We compare HBAM with other state-of-the-art LLMs such as bidirectional encoder representation from transformers (BERT) and Manhattan LSTM Model (MaLSTM). We train and test the models with a subset of the Quora duplicate questions dataset in the medical area. The experimental results show that our model is able to achieve a superior performance than these existing methods.

• The paper presents HHH, a novel system that combines a knowledge graph with a hierarchical BiLSTM attention model to address complex medical queries.
• Experimental results show HBAM achieves 81.2% accuracy on medical questions, outperforming benchmarks like BERT and MaLSTM.
• The study demonstrates that integrating structured data with deep learning can enhance patient interactions and healthcare service efficiency.

HHH: An Online Medical Chatbot System based on Knowledge Graph and Hierarchical Bi-Directional Attention

The research paper under discussion proposes an innovative chatbot framework specifically designed for addressing complex medical questions. This framework, referred to as HHH (Healthcare Helper with a Hybrid QA model), uniquely integrates a knowledge graph with a deep learning-based textual similarity model, termed the Hierarchical BiLSTM Attention Model (HBAM). This hybrid approach is tailored towards achieving superior performance in extracting and providing relevant answers from vast datasets of medical questions and answers.

Key Contributions

The primary contribution of this paper is the HHH system that efficiently combines the strengths of a knowledge graph with the adaptive and context-aware capabilities of HBAM. The knowledge graph offers structured storage and retrieval of medical data, allowing quick responses to syntactically and semantically well-defined medical inquiries. In contrast, the HBAM is adept at understanding and processing natural language, thereby addressing the limitations of rigid template-based or purely knowledge graph-centric approaches.

A standout feature of HBAM is its reliance on BiLSTM layers coupled with an attention mechanism, which effectively highlights and weighs sentence components critical for understanding medical queries. This design choice is pivotal in navigating the semantic complexities inherent in medical language.

Experimental Results

The authors conducted extensive experiments to benchmark HBAM against state-of-the-art models such as BERT and MaLSTM. Utilizing a medical subset derived from the Quora duplicate questions corpus, the HBAM demonstrated a notable accuracy of 81.2%, outperforming BERT and MaLSTM, which achieved 78.2% and 78.4% respectively. These results underscore HBAM's proficiency in pinpointing semantic equivalences between medical queries.

The paper further validates the robustness of HBAM using datasets from ehealthforumQAs, questionDoctorQAs, and webmdQAs, confirming its adaptable performance across diverse medical contexts.

Implications and Future Directions

The deployment of HHH in real-world medical settings could revolutionize patient interaction systems by significantly reducing resource wastage and time spent in primary healthcare access. By facilitating precise, context-aware responses, HHH holds the potential to enhance healthcare service delivery and patient satisfaction.

Theoretically, the integration of a knowledge graph and hierarchical BiLSTM attention forms a promising template for future AI systems tackling domain-specific question-answering tasks. This blend can be further refined to accommodate multi-turn dialogues, personalized responses utilizing user profiles, and predictive analytics for healthcare trends.

However, practical deployment warrants substantial future work, including user-centric evaluations and enhancements to handle conversational complexity beyond single-turn interactions. As the integration of AI in healthcare continues, frameworks like HHH will play vital roles in future developments, marking a step forward in the integration of digital tools in personalized medicine.