Final published version
Research output: Contribution to Journal/Magazine › Journal article › peer-review
Research output: Contribution to Journal/Magazine › Journal article › peer-review
}
TY - JOUR
T1 - Classifying encyclopedia articles
T2 - Comparing machine and deep learning methods and exploring their predictions
AU - Brenon, Alice
AU - Moncla, Ludovic
AU - Mcdonough, Katherine
PY - 2022/11/30
Y1 - 2022/11/30
N2 - This article presents a comparative study of supervised classification approaches applied to the automatic classification of encyclopedia articles written in French. Our dataset is composed of 17 volumes of text from the Encyclopédie by Diderot and d'Alembert (1751-72) including about 70,000 articles. We combine text vectorization (bag-of-words and word embeddings) with machine learning methods, deep learning, and transformer architectures. In addition evaluating these approaches, we review the classification predictions using a variety of quantitative and qualitative methods. The best model obtains 86% as an average f-score for 38 classes. Using network analysis we highlight the difficulty of classifying semantically close classes. We also introduce examples of opportunities for qualitative evaluation of "misclassifications" in order to understand the relationship between content and different ways of ordering knowledge. We openly release all code and results obtained during this research.
AB - This article presents a comparative study of supervised classification approaches applied to the automatic classification of encyclopedia articles written in French. Our dataset is composed of 17 volumes of text from the Encyclopédie by Diderot and d'Alembert (1751-72) including about 70,000 articles. We combine text vectorization (bag-of-words and word embeddings) with machine learning methods, deep learning, and transformer architectures. In addition evaluating these approaches, we review the classification predictions using a variety of quantitative and qualitative methods. The best model obtains 86% as an average f-score for 38 classes. Using network analysis we highlight the difficulty of classifying semantically close classes. We also introduce examples of opportunities for qualitative evaluation of "misclassifications" in order to understand the relationship between content and different ways of ordering knowledge. We openly release all code and results obtained during this research.
UR - https://hal.science/hal-03821073
U2 - 10.1016/j.datak.2022.102098
DO - 10.1016/j.datak.2022.102098
M3 - Journal article
VL - 142
JO - Data and Knowledge Engineering
JF - Data and Knowledge Engineering
SN - 0169-023X
M1 - 102098
ER -