SOFT CLUSTERING USING SENTENCE EMBEDDING ON LEARNING OUTCOMES TEXT

• Main Author: Gianmarg H. Siahaan, Steven
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/82443
• Institution: Institut Teknologi Bandung

Analysis of learning outcomes is one of the things needed to carry out continuous improvement in the curriculum used. This problem is included in the multi-label classification task. However, there is no data that has ground truth yet, and the annotation process takes quite a long time. Therefore, in this final assignment, soft clustering of texts is carried out into competencies. In grouping text, sentence embedding shows very good results compared to embedding at word level granularity. This final assignment analyzes learning outcomes by producing a final score obtained from the product of the weight and competency representation value. The weights were obtained through experiments in developing alternative soft clustering models, namely fuzzy c-means, Gaussian mixture models and calculating semantic similarity scores. Meanwhile, the representation of the value of each competency is obtained by implementing the Weiszfeld algorithm. Experimental and testing results show that the fuzzy c-means model using a variation of the all-mpnet-base-v2 sentence embedding model shows the best results with macro average F-1 score 0.73, micro average f-1 score 0.63, and weighted average f-1 Score 0.69 compared to other variations.