Practicability of Ensemble Artificial Neural Network Models for a Classification Task: An Optimal Approach for Reproducing Classification Practices of Health Consumers Generated Text on Social Media

集成式人工神經網絡模型於分類實務之可行性：以社群媒體之健康消費者資訊分類為例

Sukjin You ； Min Sook Park ； Soohyung Joo

《圖書資訊學刊》 20卷1期 (2022/06) Pp. 1-30

https://doi.org/10.6182/jlis.202206_20(1).001

摘要

This paper reports the classification accuracy of artificial neural network (ANN) models in reproducing health consumers' classification practices in social media. Social media have driven the growth of unstructured text data across domains including health, which motivates researchers to reconsider the epistemological approach to automated classification. This study compared the performance of several types of ANN models and ensemble models based on classification results and the integration of multiple ANN structures. To train these models, two dictionaries were employed: health consumers' terms extracted from questions and answers in the health categories of Yahoo!Answers and MeSH terms. All three types of individual classifiers demonstrated accuracies of around 90%. In particular, the fully connected ANN with two layers produced relatively higher classification performances than a convolutional neural network and long short-term memory. Ensemble models based on classification results outperformed not only the ensemble models based on the integration of heterogeneous ANN structures but also individual deep-learning models. The combination of questions and best answers were found to be most effective as a training dataset to build an accurate prediction model. The findings suggest that ANN models can be an effective assistive tool in classifying online health resources generated by health consumers in natural language.

關鍵字

Automated Classification ； Deep Learning ； Artificial Neural Network ； Ensemble Classification Model ； Knowledge Organization

並列摘要

本文運用人工神經網絡（Artificial Neural Network, ANN）模型，再現社群媒體中健康資訊分類實務之準確性。本研究透過Yahoo!Answers健康類別之問答，提取健康資訊術語，並輔以醫學主題詞表（MeSH terms），訓練並比較數種類型的ANN模型和集成式模型的效能。研究顯示，ANN模型分類準確率約90%；其中，深度神經網絡（Deep Neural Network, DNN）與卷積神經網絡（Convolutional Neural Network, CNN）和長短期記憶模型（long short-term memory, LSTM）相比，分類表現更佳。基於分類結果的集成模型不僅優於以基於異質ANN結構的集成模型，也優於單一深度學習模型；本研究也發現問題和最佳答案的組合是最有效的訓練集，並可以建構準確的預測模型。研究結果顯示，ANN模型可有效輔助分類健康消費者以自然語言生成之線上健康資訊。

並列關鍵字

自動分類 ； 深度學習 ； 人工神經網絡 ； 集成分類模型 ； 知識組織

參考文獻

Agatonovic-Kustrin, S., & Beresford, R. (2000). Basic concepts of artificial neural network (ANN) modeling and its application in pharmaceutical research. Journal of Pharmaceutical & Biomedical Analysis, 22(5), 717-727. doi: 10.1016/S0731-7085(99)00272-1

Apté, C., Damerau, F., & Weiss, S. (1994). Automated learning of decision rules for text categorization. ACM Transactions on Information Systems, 12(3), 233-251. doi: 10.1145/183422.183423

Bian, S., & Wang, W. (2007). On diversity and accuracy of homogeneous and heterogeneous ensembles. International Journal of Hybrid Intelligent Systems, 4(2), 103-128. doi: 10.3233/HIS-2007-4204

Breiman, L. (1996). Bagging predictors. Machine Learning, 24(2), 123-140. doi: 10.1007/BF00058655

Calefato, F., Lanubile, F., & Novielli, N. (2016). Moving to stack overflow: Best-answer prediction in legacy developer forums. In M. Genero (Chair), Proceedings of the 10th ACM/IEEE International Symposium on Empirical Software Engineering and Measurement (pp. 1-10). New York, NY: Association for Computing Machinery. doi: 10.1145/2961111.2962585