Context-Aware Symptom Checking for Disease Diagnosis Using Hierarchical Reinforcement Learning

Authors: Hao-Cheng Kao, Kai-Fu Tang, Edward Chang

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical studies on a simulated dataset show that our proposed model drastically improves disease prediction accuracy by a significant margin (for top-1 prediction, the improvement margin is 10% for 50 common diseases1 and 5% when expanding to 100 diseases).
Researcher Affiliation Industry Hao-Cheng Kao HTC Research & Healthcare haocheng kao@htc.com Kai-Fu Tang HTC Research & Healthcare tkevin tang@htc.com Edward Y. Chang HTC Research & Healthcare edward chang@htc.com
Pseudocode Yes Algorithm 1: Training Master Model
Open Source Code No The paper refers to the use of 'Deep Q Open AI Platform (Zou et al. 2017)' and 'Deep Q Tricorder (Chang et al. 2017)' but does not provide any explicit statement or link to the source code for the methodology described in this paper.
Open Datasets Yes To evaluate our algorithm, we generated simulated data based on Sym CAT s symptom-disease database (AHEAD Research Inc 2017) composed of 801 diseases.
Dataset Splits No The paper describes generating training data and sampling test data ('We sampled 10,000 simulated patients for each disease in the testing dataset for each task') but does not specify a separate validation dataset split (e.g., by percentage or count).
Hardware Specification No The paper mentions using 'Deep Q Open AI Platform' for training management but does not specify any hardware details like GPU/CPU models, memory, or specific computing environments used for the experiments.
Software Dependencies No The paper mentions using 'Deep Q Open AI Platform' and concepts like 'DQN', but does not provide specific version numbers for any software dependencies, libraries, or frameworks used in the experiments.
Experiment Setup Yes In all tasks, we used ten million mini-batches, each consisting of 128 samples for training.