An Inception Convolutional Autoencoder Model for Chinese Healthcare Question Clustering.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Machine Learning

Background:

• Healthcare question answering (HQA) systems are crucial for patient engagement but face challenges with complex, high-dimensional, and noisy question data.
• Existing methods struggle to effectively represent and cluster diverse healthcare queries, hindering the development of advanced HQA systems.

Purpose of the Study:

• To propose an Inception Convolutional Autoencoder Model for Chinese Healthcare question Clustering (ICAHC).
• To enhance the representation learning and clustering of Chinese healthcare questions by addressing issues like high dimensionality, sparseness, and nonprofessional language.

Main Methods:

• Utilized convolutional autoencoder networks with multiple kernel sizes to capture diverse data representations.
• Developed four ensemble operators to merge representations, incorporating skip connections for improved feature integration.
• Mapped encoder features into a lower-dimensional space for efficient clustering.

Main Results:

• The proposed ICAHC model demonstrated superior performance in discovering effective clustering solutions compared to existing algorithms.
• Experimental results validated the model's effectiveness on a Chinese healthcare dataset.
• The clustering approach successfully captured diverse and high-quality representations of healthcare questions.

Conclusions:

• ICAHC offers a robust solution for clustering challenging healthcare question datasets.
• The findings support the prediction of patient conditions and the advancement of automated HQA systems.
• This research contributes to improving the accessibility and utility of healthcare information through advanced AI techniques.