Product data can be useful to perform environmental impact assessments of product life-cycles. In order to automatize such assessments, this research is examining methodologies that encounter the...Show moreProduct data can be useful to perform environmental impact assessments of product life-cycles. In order to automatize such assessments, this research is examining methodologies that encounter the challenges with the processing and classification of product data. We consider a large imbalanced and multilingual dataset with short and noisy product descriptions that have been labeled by human annotators. The product classes are hierarchically ordered and characterized by two levels. To treat the class imbalance we proposed two data enrichment methods on the training data. Oversampling and a web scraping method with a prior-filtering. The web scraper was parsing web data using a search engine and used Sentence-BERT with cosine-similarity to assess semantic relevant information. In addition, we proposed two classification methods, Support Vector Machines (SVM) and BERT. Both models were evaluated according to several experiments considering a flattened- and hierarchical classification approach of the products. In addition, we perform an extensive error analysis on the model results considering the SVM feature importance and the BERT attention weights. The results showed that both models show similar flattened classification performance using the normal data, i.e. no data enrichment. SVM show better flattened classification performance after treated the class imbalance with data enrichment. BERT show poor performance using data enrichment and is overfitting the training data. Hierarchical classification improved the classification performance of BERT using oversampling. SVM did not benefit from the hierarchical classification approach and show better classification performance using flattened classification. As last, the error analysis have showed that the data consist of incorrect or subjective manual labeling. The SVM feature importance and BERT attention weights results suggest that nonrepresentative tokens or out-of-vocabulary tokens have the tendency to decrease the classification performance.Show less
Product data can be useful to perform environmental impact assessments of prod-uct life-cycles. In order to automatize such assessments, this research is examiningmethodologies that encounter the...Show moreProduct data can be useful to perform environmental impact assessments of prod-uct life-cycles. In order to automatize such assessments, this research is examiningmethodologies that encounter the challenges with the processing and classification ofproduct data. We consider a large imbalanced and multilingual dataset with short andnoisy product descriptions that have been labeled by human annotators. The productclasses are hierarchically ordered and characterized by two levels. To treat the class im-balance we proposed two data enrichment methods on the training data. Oversamplingand a web scraping method with a prior-filtering. The web scraper was parsing webdata using a search engine and used Sentence-BERT with cosine-similarity to assesssemantic relevant information. In addition, we proposed two classification methods,Support Vector Machines (SVM) and BERT. Both models were evaluated according toseveral experiments considering a flattened- and hierarchical classification approach ofthe products. In addition, we perform an extensive error analysis on the model resultsconsidering the SVM feature importance and the BERT attention weights.The results showed that both models show similar flattened classification performanceusing the normal data, i.e. no data enrichment. SVM show better flattened classi-fication performance after treated the class imbalance with data enrichment. BERTshow poor performance using data enrichment and is overfitting the training data.Hierarchical classification improved the classification performance of BERT using over-sampling. SVM did not benefit from the hierarchical classification approach and showbetter classification performance using flattened classification. As last, the error anal-ysis have showed that the data consist of incorrect or subjective manual labeling.The SVM feature importance and BERT attention weights results suggest that non-representative tokens or out-of-vocabulary tokens have the tendency to decrease theclassification performance.Show less
The two-sample independent variables t-test for unequal variances or Welch’s t-test is known to lose accuracy when sample sizes are small or when there is a high level of skewness in the sample....Show moreThe two-sample independent variables t-test for unequal variances or Welch’s t-test is known to lose accuracy when sample sizes are small or when there is a high level of skewness in the sample. This study compares the bootstrap t-test to the Welch’s t-test in a multifactor simulation study. Bootstrapping is a different approach to the t-test, where the test makes use of resampling. The data was generated for different sample size ratios, levels of skew, variance, and multiple effect sizes. In the literature which the bootstrap method was often found to have higher power for small sample sizes and high skewness levels (Hayes & Cai, 2007; Tsagris et al., 2020). However, the findings of this study were only partially in line with this. For the Type I error the bootstrap t-test was found to perform similarly to the Welch’s t-test with ranges staying within or just outside the specified alpha level of 5%. On the Power comparison, the bootstrap t-test underperformed. With the Welch’s t-test having equal or higher power in most cases, especially as skewness was introduced in the dataset. In total, there were 13 out of 192 conditions where the bootstrap had a significantly higher power. All of these were conditions with very small sample sizes, usually combined large sample size ratios. Which was expected according to literature (de Winter, 2013), but this was not enough to confirm my hypothesis. Based on these findings I suggest that further research is needed into the potential use cases of the bootstrap t-test.Show less
