Reproduciblity Report on "Multi Emotion Detection in User Generated Reviews" by Buitinck et al. (ECIR 2015)
| Main Authors: | Eggerth Cordula, Kiesel Markus, Niedermayer Kolos |
|---|---|
| Format: | Report Journal |
| Bahasa: | eng |
| Terbitan: |
, 2021
|
| Subjects: | |
| Online Access: |
https://zenodo.org/record/4459925 |
Daftar Isi:
- This paper aims to reproduce the supervised multi-emotion detection experiment carried out by Buitinck et al. 2015 based on textual data stemming from 44 Hollywood movie reviews. The training and test dataset provided by Buitinck et al. are used as a basis for recreating the experiment, which compares the One-vs.-Rest (OvR) classifier and the Random k-labelsets (RAKEL) algorithm according to the F1 score as the main performance metric. Reproducibility of the project was not given out of the box based on the available paper and train/test split. Indeed, a number of decisions had to be made with regard to the choice of algorithm implementations and versions (in Python’s scikit-learn library)), the related tuning parameters and methods as well as the implementations of the significance tests. All in all, this paper achieves to reproduce the structure and workflows of the experiment. However, differences in the actual outcome can be observed given the series of assumptions that had to be made due to the lack of clear specification by the authors of the original paper.