Interpersonal communication between humans is driven by different emotions expressed by various communication channels, including touch. Touch is a fundamental aspect of social interaction that has connected people of all ages to different concepts and mechanisms in the physical world and directly influences emotional expression and recognition through intimacy. The use of digital communication, however, has reduced physical interactions and has shifted the recognition of emotions via other modalities, such as face and voice, collected using cameras. However, these modalities can only sometimes support the prevalence of smartphone use. The face may not always be captured, nor the excellent quality audio collected for mobile applications to recognize emotions. This work introduces an alternative modality - haptic touch for emotion recognition. This paper presents haptic touch data collected using a mobile application called GetEmotion. It collects the responses of twenty (20) participants as they view clips from the LIRIS-ACCEDE dataset. Following the theory of James Russell's Circumplex Model of Affect, features from the collected data were pre-processed and analyzed for correlation and significance to the valance and arousal values of the video clips. Results show that three features have a significant relation to the valence values, namely, pressure, start, and end coordinates along the$\boldsymbol{x}-\boldsymbol{axis}$. These findings suggest that the location of touch interaction on the screen and the intensity of touch indicate emotional states. More so, arousal levels have been found to significantly correlate with diverse features such as duration, touch count, distance, and various velocity and acceleration values. The findings indicate that both the nature and intensity of touch activities can reflect users' emotional responses to video stimuli.

Recognizing student emotions can significantly enhance the learning process. This study investigates the effectiveness of transfer learning with VGG-16 and ResNet-18 models for classifying student emotions based on facial expressions. Leveraging pre-trained models and employing cross-validation, we achieved a robust valence classification accuracy of 92% on FER2013 dataset. However, when applied to MAHNOB-HCI and ACADEMO datasets characterized by limited and subtle emotional cues, performance declined to approximately 82% with overfitting. To enhance model generalization and mitigate overfitting, strategies such as data augmentation, regularization techniques, and hyperparameter optimization are proposed. Our findings demonstrate the effectiveness of transfer learning in recognizing student emotions, which may significantly impact education through personalized learning, improved student engagement, and early intervention.

This is an introduction to the Second International Workshop on Context Based Affect Recognition CBAR 2013 Held in conjunction with Affective Computing and Intelligent Interaction 2-5 September 2013, Geneva, Switzerland.

Academic emotions such as confidence, excitement, frustration and interest may be predicted based on brainwaves signals. It is shown that the prediction rate can be improved further when the data from brainwaves signals are complemented by data based on mouse click behavior. Twenty -five (25) undergraduate students were asked to use a math tutoring software while an EEG se nsor was attached to their head to capture their brainwaves sign als throughout the learning session. At the same time, mouse -click features such as the number of clicks, the duration of each click and the distance traveled by the mouse were automatically captured. Using a Multi -Layered Perceptron classifier, classifica tion using brainwaves data alone had accuracy rates of 54 to 88%. Prediction rates based purely on mouse features had accuracy rates of only 32 to 48%. When the two input modalities are combined, accuracy rates increased to up to 92%. Furthermore, the expe riments confirmed that the predication accuracy rate increases as the number of feature values that deviate significantly from the mean increases. In particular, the prediction rates exceed 80% when at least 33% of the features have values that deviate from the mean by more than 1 standard deviation.

Students who performed well in their college mathematics subjects, referred to here as academic achievers, were divided into two groups according to the self-reported level of difficulty faced by them while performing several programming tasks in LOGO - a programming language using turtle-graphics. It is shown that, to some extent, the level of difficulty of tasks faced by academic achievers can be predicted, based on their measured affective levels of excitement, frustration and engagement. These affective states are measured using brainwaves sensors that are attached to the head of the student. Those who assessed the learning experience as easy tend to have higher levels of excitement than those who reported to have experienced difficulty in learning the language. On the other hand, the level of frustration among those having difficulty with the tasks registered slightly higher frustration levels. Three machine learning algorithms were used to predict whether or not a learner finds the tasks to be easy. The average predictive accuracy is 70%.

Students have different ways of learning and have varied reactions to feedback. Thus, allowing a system to predict how students would appraise certain feedback gives it the capability to adapt to what would help a student learn better. This research focuses on the prediction of a student’s appraisal of feedback provided in an intelligent tutoring system (ITS). A regression model for frustration and excitement is created to perform prediction. The frustration model was able to achieve a 0.724 correlation with a 0.164 RMSE and the excitement model was able to achieve 0.6 a correlation with a 0.189 RMSE. These results indicate the potential of using these models for allowing systems to adjust feedback automatically based on student’s reactions while using an ITS.

Data-centric affect modeling may render itself restrictive in practical applications for three reasons, namely, it falls short of feature optimization, infers discrete affect classes, and deals with relatively small to average sized datasets. Though it seems practical to use the feature combinations already associated to commonly investigated sensors, there may be other potentially optimal features that can lead to new relations. Secondly, although it seems more realistic to view affect as continuous, it requires using continuous labels that will increase the difficulty of modeling. Lastly, although a large scale dataset reflects a more precise range of values for any given feature, it severely hinders computational efficiency. We address these problems when inferring physiology-affect relations from datasets that contain 2-3 million feature vectors, each with 49 features and labelled with continuous affect values. We employ automatic feature selection to acquire near optimal feature subsets and a fast approximate kNN algorithm to solve the regression problem and cope with the challenge of a large scale dataset. Our results show that high estimation accuracy may be achieved even when the selected feature subset is only about 7% of the original features. May the results here motivate the HCI community to pursue affect modeling without being deterred by large datasets and further the discussions on acquiring optimal features for accurate continuous affect approximation.