Table 1.
The 11 CBT quality codes defined by CTRS.
Fig 1.
Distribution of the 11 CTRS codes (and the total CTRS) across the Likert scale.
A total CTRS score above or equal to 40 indicates competent delivery of CBT.
Table 2.
Examples of utterances within sessions with high/low scores for the CTRS dimensions of agenda and homework.
Table 3.
Size of the datasets used to train and evaluate the proposed models.
Fig 2.
Proposed architecture for total CTRS score classification following a single-task approach.
Here, only the utterances attributed to one of the interlocutors (i.e., therapist) are used.
Fig 3.
Proposed architecture for total CTRS score classification following a multi-task approach modeling each CTRS code.
Here, only the utterances attributed to one of the interlocutors (i.e., therapist) are used.
Table 4.
Accuracy (%) for the next sentence prediction task before and after BERT adaptation when the models are evaluated on the CBT dataset.
Table 5.
F1 score (%) based on 10-fold cross validation.
Table 6.
Contribution of each proposed technique to the performance of the system.
Fig 4.
Mean attention weights across all the sessions, combining the results from all the testing folds of the cross validation.
(left) Weights for the codes agenda, homework and feedback from the multi-task architecture. (right) Weights for the total CTRS from the single-task architecture and mean weights of all the 11 individual codes from the multi-task architecture.