Human activity recognition has become an active research field over the past few years due to its wide application in various fields such as health-care, smart home monitoring, and surveillance. Existing approaches for activity recognition in smart homes have achieved promising results. Most of these approaches evaluate real-time recognition of activities using only sensor activations that precede the evaluation time (where the decision is made). However, in several critical situations, such as diagnosing people with dementia, “preceding sensor activations” are not always sufficient to accurately recognize the inhabitant's daily activities in each evaluated time. To improve performance, we propose a method that delays the recognition process in order to include some sensor activations that occur after the point in time where the decision needs to be made. For this, the proposed method uses multiple incremental fuzzy temporal windows to extract features from both preceding and some oncoming sensor activations. The proposed method is evaluated with two temporal deep learning models (Convolutional Neural Network and Long Short-Term Memory), on a binary sensor dataset of real daily living activities. The experimental evaluation shows that the proposed method achieves significantly better results than the real-time approach, and that the representation with fuzzy temporal windows enhances performance within Deep Learning models.

Feature construction and selection are two key factors in the field of machine learning (ML). Usually, these are very time-consuming and complex tasks because the features have to be manually crafted. The features are aggregated, combined or split to create features from raw data. In this paper, we propose a methodology that makes use of ontologies to automatically generate features for the ML algorithms. The features are generated by combining the concepts and relationships that are already in the knowledge base, expressed in form of ontology. The proposed methodology has been evaluated with three different activities of a popular dataset, showing its effectiveness in the recognition of activities of daily living (ADL). The obtained successful results indicate that the use of extended feature vectors provided by the use of ontologies offers a better accuracy, regarding the original feature vectors of the classic approach, where each feature corresponds to the activation of a sensor. Although the classic approach produces classifiers with accuracies above 92%, the proposed methodology improves those results by 1.9%, on average, without adding more information to the dataset.

Suspense is a key narrative issue in terms of emotional gratifications. Reactions in response to this type of entertainment are positively related to enjoyment, having a significant impact on the audience's immersion and suspension of disbelief. Related to computational modelling of this feature, some automatic storytelling systems include limited implementations of suspense management system in their core. In this way, the interest of this subject in the area of creativity has resorted to different definitions from fields as narratology and the film industry, as much as several proposals of its constituent features. Among their characteristics, uncertainty is one of the most discussed in terms of impact and need: while many authors affirm that uncertainty is essential to evoke suspense, others limit or reject its influence. Furthermore, the paradox of suspense reflects the problem of including uncertainty as a component required in suspense creation systems. Due to this need to contrast the effects of the uncertainty in order to compute a general model for automatic storytelling systems, we conducted an experiment measuring suspense experienced by a group of subjects that read a story. While a group of them were told the ending of the story in advance, the members of the other group experimented the same story in chronological order. Both the subjects' reported suspense and their physiological responses are gathered and analyzed. Results provide evidence to conclude that uncertainty affects the emotional response of readers, but independently and in a different form than suspense does. It will help to propose a model in which uncertainty is processed separately as management of the amount of knowledge about the outcome available to the spectator, which acts as a control signal to modulate the input features, but not directly in suspense computing.

Data driven approaches for human activity recognition learn from pre-existent large-scale datasets to generate a classification algorithm that can recognize target activities. Typically, several activities are represented within such datasets, characterized by multiple features that are computed from sensor devices. Often, some features are found to be more relevant to particular activities, which can lead to the classification algorithm providing less accuracy in detecting the activity where such features are not so relevant. This work presents an experimentation for human activity recognition with features derived from the acceleration data of a wearable device. Specifically, this work analyzes which features are most relevant for each activity and furthermore investigates which classifier provides the best accuracy with those features. The results obtained indicate that the best classifier is the k-nearest neighbor and furthermore, confirms that there do exist redundant features that generally introduce noise into the classification, leading to decreased accuracy.

The recognition of activities of daily living is an important research area of interest in recent years. The process of activity recognition aims to recognize the actions of one or more people in a smart environment, in which a set of sensors has been deployed. Usually, all the events produced during each activity are taken into account to develop the classification models. However, the instant in which an activity started is unknown in a real environment. Therefore, only the most recent events are usually used. In this paper, we use statistics to determine the most appropriate length of that interval for each type of activity. In addition, we use ontologies to automatically generate features that serve as the input for the supervised learning algorithms that produce the classification model. The features are formed by combining the entities in the ontology, such as concepts and properties. The results obtained show a significant increase in the accuracy of the classification models generated with respect to the classical approach, in which only the state of the sensors is taken into account. Moreover, the results obtained in a simulation of a real environment under an event-based segmentation also show an improvement in most activities.