Publication: Exploring learning assessment in lean manufacturing training through natural languange processing
| datacite.subject.fos | oecd::Engineering and technology::Mechanical engineering | |
| dc.contributor.author | Ong, Yi Hang | |
| dc.date.accessioned | 2026-02-04T02:44:19Z | |
| dc.date.available | 2026-02-04T02:44:19Z | |
| dc.date.issued | 2023-07-01 | |
| dc.description.abstract | Using Natural Language Processing (NLP), the study investigates the learning assessment of the lean game. In the study, the 5S lean game was selected, which consists of five concepts: "Sort," "Set in order," "Shine," "Standardize," and "Sustain." Regarding research methodology, thirty students were chosen to play the game and participate in pre- and post-game interviews. Using Whisper AI, the audio recordings from both interview sessions were converted to text script in Google Collaboratory. Several text preprocessing techniques, such as tokenization, stop word, punctuation mark filtering, lemmatization, lower casing, and expansion in contraction word were used to process the text scripts. After that, a classifier was created using Fasttext to supervisedly train the keywords generated from various sources. Next, the preprocessed data were fed into the classifier to obtain the probability for each 5S concept based on the keywords identified, which associates to the learning gain. These probabilities were analyzed using three methods of aggregation: the mean, the maximum, and the threshold in probability difference between pre-game and post-game interviews. The result indicates that "Sort" and "Set in order" achieved the highest positive learning gain in the overall computation and feedback results. In contrast, the survey feedback is slightly varied compared to the results obtained from the computation methods, where the “Sustain” concept has a higher rank compared to the “Standardize” and “Shine” concepts despite obtaining the least informative and the highest vote for improvement in the survey feedback due to linguistic complexity of these concepts, the response shift between pre-game and post-game interviews, keyword overlap, and the imbalanced 5S game design. Besides, study also reveal the appropriateness of using individual probability method as the evaluation of learning compared to probability difference. The study contributes to a pioneering research in term of developing an methodology to assess learning through NLP, with a successful demonstration. Keywords: Learning assessment, Natural Language Processing, 5S, Fasttext, probability difference, pre-game interview, post-game interview. | |
| dc.identifier.uri | https://erepo.usm.my/handle/123456789/23561 | |
| dc.language.iso | en | |
| dc.title | Exploring learning assessment in lean manufacturing training through natural languange processing | |
| dc.type | Resource Types::text::report::research report | |
| dspace.entity.type | Publication | |
| oairecerif.author.affiliation | Universiti Sains Malaysia |