The HoPE Model Architecture: a Novel Approach to Pregnancy Information Retrieval Based on Conversational Agents

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Healthcare Informatics

Background:

• Conversational agents require assertive and fast responses for effective human interaction.
• Transformer architectures, while powerful, may present speed limitations in conversational agent applications.
• Developing specialized models for specific domains like pregnancy care is crucial for improving performance.

Purpose of the Study:

• To evaluate and propose the HoPE (Healthcare Obstetric in PrEgnancy) model tailored for pregnancy-related data.
• To identify the optimal pre-trained Portuguese language model for a conversational agent architecture in obstetrics.
• To enhance information retrieval speed and accuracy for pregnancy health queries.

Main Methods:

• Dataset extraction and construction from diverse health documents (breastfeeding, childcare, nutrition, etc.).
• Evaluation of two pre-trained Portuguese language models for conversational agent architecture.
• Selection and fine-tuning of the BERTimbau model using data augmentation strategies.
• Development of the HoPE model architecture.

Main Results:

• The BERTimbau model demonstrated superior speed and accuracy in information retrieval compared to alternatives.
• Fine-tuning BERTimbau with 1,500 augmented pairs achieved a 95.55 Spearman correlation.
• The HoPE model architecture attained an F1-Score of 0.89, surpassing other tested combinations.

Conclusions:

• The HoPE model, leveraging BERTimbau, offers a high-performance solution for pregnancy-focused conversational agents.
• The model shows significant improvements in accuracy and speed for retrieving obstetric health information.
• Future clinical studies are planned to evaluate the HoPE model's real-world applicability.