Academic

Big Data Analytics for Reputational Reliability Assessment Using Customer Review Data, (2021)
ESREL2021 Conference

Authors : Jean Meunier-Pion, Zhiguo Zeng, Jie Liu

Access link : https://cmswebonline.com/esrel2021-epro/html/434.xml

Short description : This paper is a first step in order to assess reliability based on online reviews written by customers. In this paper, we focus on logistic regression models which use the scores given by customers and the reviews that they wrote, so as to determine whether the products they bought failed, and in such a case, we try to assess if the failure was severe. The major contribution in this paper is the use of an ensemble learning technique to obtain better results on our dataset. The main issue that we faced, however, was the size of our dataset being small compared to the number of textual features, which led to strong overfitting when training models with textual data.

GitHub : not public yet. ⏳

DCASE CHALLENGE 2022 : SELF-SUPERVISED LEARNING PRE-TRAINING, TRAINING FOR UNSUPERVISED ANOMALOUS SOUND DETECTION, (2022)
DCASE2022 Challenge Task2 Technical Report

Authors : Ismail Nejjar, Jean Meunier-Pion, Gaetan Frusque, Olga Fink

Access link : https://dcase.community/documents/challenge2022/technical_reports/DCASE2022_Nejjar_66_t2.pdf

Short description : Technical report describing the submission for the DCASE2022 Challenge Task2. The main contribution is the introduction of DG-Mix, a VICReg-inspired framework, for pre-training frameworks in scenarios where data are subject to highly diverse domain shifts. The method, without heavy fine-tuning, nor ensemble methods, outperformed the challenge baselines on both the development and evaluation datasets.