Patekote į pirmąjį lietuvišką domeną. 
2026 m. kovo 9 d., 13 val.
Vilnius, Akademijos g. 4, 203 kab.
Nuotoliniu būdu „MS Teams“ aplinkoje (https://bit.ly/DMSTI_2026-03-09)
Vaida Masiulionytė-Dagienė
(vadovė doc. dr. Tatjana Jevsikova)
„Informatinio mąstymo automatinio vertinimo sistemos modeliavimas“
Anotacija: This dissertation proposes a process-based model for automated assessment of computational thinking (CT) using behavioural data from interactive Bebras Challenge tasks. Instead of evaluating only final correctness, the model analyses solution processes based on interaction logs.
Solution attempts are transformed into structured behavioural representations including total actions, solution duration, object-level interactions, and click order. All features are encoded into a unified representation and normalised to enable comparability across tasks with the same interaction structure.
A pilot study applied k-means clustering to explore behavioural grouping of solutions. In the main experiments, Affinity Propagation was used to automatically determine cluster structure. FALSE solutions were grouped into 7 behavioural classes and TRUE solutions into 4 classes.
Solution types were classified using a dual-branch neural network. One branch analyses click sequences with an LSTM layer, while the second branch analyses static features such as solution duration and total actions. The two outputs are merged to produce the final prediction. The unweighted model achieved 85.9% accuracy for FALSE solutions and 93.0% for TRUE solutions. Minority FALSE classes showed recall values of 0.464 and 0.321. Applying class weighting increased recall to 0.844 and 0.728, while overall accuracy decreased slightly to 83.4% and 90.6%.
Generalisation experiments with unseen tasks of the same interaction type confirmed transferability of behavioural classes. The results show that encoded and normalised behavioural data can support scalable, automated, and more informative CT assessment beyond binary scoring. An additional expert validation confirmed that the identified behavioural solution groups are interpretable and meaningful for assessment purposes.