Data augmentation is a promising technique in improving exploration and convergence speed in deep reinforcement learning methodologies. In this work, we propose a data augmentation framework based on generative models for creating completely novel states and increasing diversity. For this purpose, a diffusion model is used to generate artificial states (learning the distribution of original, collected states), while an additional model is trained to predict the action executed between two consecutive states. These models are combined to create synthetic data for cases of high and low immediate rewards, which are encountered less frequently during the agent’s interaction with the environment. During the training process, the synthetic samples are mixed with the actually observed data in order to speed up agent learning. The proposed methodology is tested on the Atari 2600 framework, producing realistic and diverse synthetic data which improve training in most cases. Specifically, the agent is evaluated on three heterogeneous games, achieving a reward increase of up to 31%, although the results indicate performance variance among the different environments. The augmentation models are independent of the learning process and can be integrated to different algorithms, as well as different environments, with slight adaptations.

This paper proposes the utilization of large language models as recommendation systems for museum visitors. Since the aforementioned models lack the notion of context, they cannot work with temporal information that is often present in recommendations for cultural environments (e.g., special exhibitions or events). In this respect, the current work aims to enhance the capabilities of large language models through a fine-tuning process that incorporates contextual information and user instructions. The resulting models are expected to be capable of providing personalized recommendations that are aligned with user preferences and desires. More specifically, Generative Pre-trained Transformer 4, a knowledge-based large language model is fine-tuned and turned into a context-aware recommendation system, adapting its suggestions based on user input and specific contextual factors such as location, time of visit, and other relevant parameters. The effectiveness of the proposed approach is evaluated through certain user studies, which ensure an improved user experience and engagement within the museum environment.

Parkinson’s Disease is the second most prevalent neurodegenerative disorder, currently affecting as high as 3% of the global population. Research suggests that up to 80% of patients manifest phonatory symptoms as early signs of the disease. In this respect, various systems have been developed that identify high risk patients by analyzing their speech using recordings obtained from natural dialogues and reading tasks conducted in clinical settings. However, most of them are centralized models, where training and inference take place on a single machine, raising concerns about data privacy and scalability. To address these issues, the current study migrates an existing, state-of-the-art centralized approach to the concept of federated learning, where the model is trained in multiple independent sessions on different machines, each with its own dataset. Therefore, the main objective is to establish a proof of concept for federated learning in this domain, demonstrating its effectiveness and viability. Moreover, the study aims to overcome challenges associated with centralized machine learning models while promoting collaborative and privacy-preserving model training.

Medical imaging modalities, like magnetic resonance imaging (MRI), have enabled efficient diagnosis of various conditions, including cancer, lung disease, and brain tumors. With the advancements in machine learning, AI-based medical image segmentation and classification systems have emerged, potentially replacing human diagnosis. However, the security and robustness of these systems are crucial, as they are vulnerable to adversarial attacks, as demonstrated in previous studies. In this respect, the current work explores the one-pixel attack’s impact on the reliable VGG16 model, the effectiveness of combining the one-pixel attack with the FGSM attack, the potential of using the squeezing color bits detector to counter the one-pixel attack, and the possibility of using a combination of the squeezing color bits and PCA whitening detectors to protect against the aforementioned attacks.

Tourism is an important economic activity for many countries and the ability of understanding visitor needs as they evolve over time is a priority for all involved stakeholders. The analysis of textual reviews written by travelers on various online platforms may be a valuable tool in this direction. In this work, we showcase the potential of this idea by examining 8 well-known attractions in the City of Athens, Greece. After retrieving the relevant data from two popular online services, we employ a state-of-the-art transformer-based language model for two tasks; the extraction of distinctive keywords and phrases out of the free-text reviews and the assignment of a sentiment score to each review. Based on this information, we can associate certain keywords and phrases with specific sentiment values and monitor their evolution over time, in the context of specific touristic {&} cultural places. The analysis that follows explores the potential of this idea in more detail.

Open laboratories (OpenLabs) in Cultural Heritage institutions are an effective way to provide visibility into the behind-the-scenes processes and promote documentation data collected and produced by domain specialists. However, presenting these processes without proper explanation or communication with specialists may cause issues in terms of visitors’ understanding. To support OpenLabs and disseminate information, digital media and efficient data management can be utilized. The CAnTi (Conservation of Ancient Tiryns) project seeks to design and implement virtual and mixed reality applications that visualize conservation and restoration data, supporting OpenLab operations at the Acropolis of Ancient Tiryns. Semantic Web technologies will be used to model the digital content, facilitating organization and interoperability with external sources in the future. These applications will be part of the OpenLab activities on the site, enhancing visitors’ experiences and understanding of current and past conservation and restoration practices.

The research field of digital storytelling is cross-disciplinary and extremely wide. In this paper, methods, frameworks, and tools that have been created for authoring and presenting digital narratives, are selected and examined among hundreds of works. The basic criterion for selecting these works has been their ability to create content by computational, emergent methods. By delving into the work of many researchers, the objective is to study current trends in this research field and discuss possible future directions. Most of the relevant tools and methods have been designed with a specific purpose in mind, but their use could be expanded to other areas of interest or could at least be the steppingstone for other ideas. Therefore, the following works show elements of computational and emergent narrative creation and a classification is proposed according to their purpose of existence. Finally, new potential research directions in the field are identified and possible future research steps are discussed.

Importance sampling, a variant of online sampling, is often used in neural network training to improve the learning process, and, in particular, the convergence speed of the model. We study, here, the performance of a set of batch selection algorithms, namely, online sampling algorithms that process small parts of the dataset at each iteration. Convergence is accelerated through the creation of a bias towards the learning of hard samples. We first consider the baseline algorithm and investigate its performance in terms of convergence speed and generalization efficiency. The latter, however, is limited in case of poor balancing of data sets. To alleviate this shortcoming, we propose two variations of the algorithm that achieve better generalization and also manage to not undermine the convergence speed boost offered by the original algorithm. Various data transformation techniques were tested in conjunction with the proposed scheme to develop an overall training method of the model and to ensure robustness in different training environments. An experimental framework was constructed using three naturally imbalanced datasets and one artificially imbalanced one. The results assess the advantage in convergence of the extended algorithm over the vanilla one, but, mostly, show better generalization performance in imbalanced data environments.

Aspect-based sentiment prediction is a specific area of sentiment analysis that models the sentiment of a text excerpt as a multi-dimensional quantity pertaining to various interpretations, rather than a scalar one, that admits a single explanation. Extending earlier work, the said task is examined as a part of a unified architecture that collects, analyzes and stores documents from various online sources, including blogs & social network posts. The obtained data are processed at various levels; initially, a hybrid, attention-based bi-directional long short-term memory network, coupled with convolutional layers, is used to extract the textual features of the document. Following, an additional number of document metadata are also examined, such as the number of repetitions, the existence, type and frequency of emoji ideograms and, especially, the presence of keywords, assigned either manually (e.g. in the form of hashtags) or automatically. All of the aforementioned features are subsequently provided as input to a fully-connected, multi-layered, feed-forward artificial neural network that performs the final prediction task. The overall approach is tested on a large corpus of documents, with encouraging results.

Package recommendation systems have gained in popularity especially in the tourism domain, where they propose combinations of different types of attractions that can be visited by someone during a city tour. These systems can also be applied in suggesting home entertainment, proper nutrition or academic courses. Such systems must optimize multiple user criteria in tandem, such as preference score, package cost or duration. This work proposes a flexible framework for recommending packages that best fit users' preferences while satisfying several constraints on the set of the valid packages. This is achieved by modeling the relation between the items and the categories these items belong to, aiming at recommending to each user the top-k packages that cover their preferred categories and the restriction of a maximum package cost. Our contribution includes an optimal and a greedy algorithm, that both outperform a state-of-the-art system and a popularity-based baseline solution. The novelty of the optimal algorithm is that it combines the collaborative filtering predictions with a graph-based model to produce package recommendations. The problem is expressed through a minimum cost flow network and is solved by integer linear programming. The greedy algorithm has a low computational complexity and provides recommendations which are close to the optimal one. An extensive evaluation of the proposed framework has been carried out on six popular recommendation datasets. The results obtained using a set of widely accepted metrics show promising performance. Finally, the formulation of the problem for specific domains has also been addressed.

Recent developments in digital technologies regarding the cultural heritage domain have driven technological trends in comfortable and convenient traveling, by offering interactive and personalized user experiences. The emergence of big data analytics, recommendation systems and personalization techniques have created a smart research field, augmenting cultural heritage visitor’s experience. In this work, a novel, hybrid recommender system for cultural places is proposed, that combines user preference with cultural tourist typologies. Starting with the McKercher typology as a user classification research base, which extracts five categories of heritage tourists out of two variables (cultural centrality and depth of user experience) and using a questionnaire, an enriched cultural tourist typology is developed, where three additional variables governing cultural visitor types are also proposed (frequency of visits, visiting knowledge and duration of the visit). The extracted categories per user are fused in a robust collaborative filtering, matrix factorization-based recommendation algorithm as extra user features. The obtained results on reference data collected from eight cities exhibit an improvement in system performance, thereby indicating the robustness of the presented approach.

Sentiment analysis is a vigorous research area, with many application domains. In this work, aspect-based sentiment prediction is examined as a component of a larger architecture that crawls, indexes and stores documents from a wide variety of online sources, including the most popular social networks. The textual part of the collected information is processed by a hybrid bi-directional long short-term memory architecture, coupled with convolutional layers along with an attention mechanism. The extracted textual features are then combined with other characteristics, such as the number of repetitions, the type and frequency of emoji ideograms in a fully-connected, feed-forward artificial neural network that performs the final prediction task. The obtained results, especially for the negative sentiment class, which is of particular importance in certain cases, are encouraging, underlying the robustness of the proposed approach.

Short-term property rentals are perhaps one of the most common traits of present day shared economy. Moreover, they are acknowledged as a major driving force behind changes in urban landscapes, ranging from established metropolises to developing townships, as well as a facilitator of geographical mobility. A geolocation ontology is a high level inference tool, typically represented as a labeled graph, for discovering latent patterns from a plethora of unstructured and multimodal data. In this work, a two-step methodological framework is proposed, where the results of various geolocation analyses, important in their own respect, such as ghost hotel discovery, form intermediate building blocks towards an enriched knowledge graph. The outlined methodology is validated upon data crawled from the Airbnb website and more specifically, on keywords extracted from comments made by users of the said platform. A rather solid case-study, based on the aforementioned type of data regarding Athens, Greece, is addressed in detail, studying the different degrees of expansion & prevalence of the phenomenon among the city’s various neighborhoods.

A critical issue for the safe operation of nuclear power plants is to quickly and accurately detect possible anomalies and perturbations in the reactor. Defects in operation are principally identified through changes in the neutron flux, as captured by detectors placed at various points inside and outside of the core. While wavelet-based analysis of the measured signals has been thoroughly used for anomaly detection, it has yet to be coupled with deep learning approaches. To this end, this work presents a novel technique for anomaly detection on nuclear reactor signals through the combined use of wavelet-based analysis and convolutional neural networks. In essence, the wavelet transform is applied to the signals and the corresponding scaleograms are produced, which are subsequently used to train a convolutional neural network that detects possible perturbations in the reactor core. The overall methodology is experimentally validated on a set of simulated nuclear reactor signals generated by a well-established relevant tool. The obtained results indicate that the trained network achieves high levels of accuracy in failure detection, while at the same time being robust to noise.

This work proposes a novel framework for enhancing abstractive text summarization based on the combination of deep learning techniques along with semantic data transformations. Initially, a theoretical model for semantic-based text generalization is introduced and used in conjunction with a deep encoder-decoder architecture in order to produce a summary in generalized form. Subsequently, a methodology is proposed which transforms the aforementioned generalized summary into human-readable form, retaining at the same time important informational aspects of the original text and addressing the problem of out-of-vocabulary or rare words. The overall approach is evaluated on two popular datasets with encouraging results.

Recent theoretical and practical advances have led to the emergence of review-based recommender systems, where user preference data is encoded in at least two dimensions; the traditional rating scores in a predefined discrete scale and the user-generated reviews in the form of free-text. The main contribution of this work is the presentation of a new technique of incorporating those reviews into collaborative filtering matrix factorization algorithms. The text of each review, of arbitrary length, is mapped to a continuous feature space of fixed length, using neural language models and more specifically, the Paragraph Vector model. Subsequently, the resulting feature vectors (the neural embeddings) are used in combination with the rating scores in a hybrid probabilistic matrix factorization algorithm, based on maximum a-posteriori estimation. The proposed methodology is then compared to three other similar approaches on six datasets in order to assess its performance. The obtained results demonstrate that the new formulation outperforms the other systems on a set of two metrics, thereby indicating the robustness of the idea.

Although abundant research work has been published in the area of path recommendation and its applications on travel and routing topics, scarce work has been reported on context-aware route recommendation systems aimed to stimulate optimal cultural heritage experiences. This paper tries to address this issue, by proposing a personalized and content adaptive cultural heritage path recommendation system, where location is modeled using mean-shift clustering trained with actual user movement patters. Additionally, topic modeling is incorporated to formalize the implicit cultural heritage content, while first order Markov models address the movement as a temporal transition aspect of the problem. The overall architecture is applied on data collected from actual visits to the archaeological sites of Gournia and {\c{C}}atalhöyük and extensive analysis on visitor movement patterns follows, especially in comparison to the curated paths in the aforementioned sites. Finally, the offline evaluation results of the proposed recommendation scheme are encouraging, validating its efficiency and setting a positive paradigm for cultural heritage route recommendations.

This work constitutes a theoretically-informed empirical analysis of the spatial characteristics of the short-term rentals' market and explores their linkage with shifts in the wider housing market within the context of a south-eastern EU metropolis. The same research objective has been pursued for a variety of international paradigms; however, to the best of our knowledge, there has not been a thorough and systematic study for Athens and its neighborhoods. With a theoretical framework that draws insight from the political-economic views of Critical Geography, this work departs from an assessment of Airbnb listings, and proceeds inquiring the expansion of the phenomenon with respect to the rates of long-term rent levels in the neighborhoods of Central Athens, utilizing relevant data. The geographical framework covers the City of Athens as a whole, an area undergoing profound transformations in recent years, stemming from diverse factors that render the city one of the most dynamic destinations of urban tourism and speculative land investment. The analysis reveals a prominent expansion of the short-term rental phenomenon across the urban fabric, especially taking ground in hitherto underexploited areas. This expansion is multifactorial, asynchronous and exhibits signs of positive relation with the long-term rentals shifts; Airbnb not only affects already gentrifying neighborhoods, but contributes to a housing market disruption in non-dynamic residential areas.

Genetic Algorithms (GAs) have been predominantly used in video games for finding the best possible sequence of actions that leads to a win condition. This work sets out to investigate an alternative application of GAs on action-adventure type video games. The main intuition is to encode actions depending on the state of the world of the game instead of the sequence of actions, like most of the other GA approaches do. Additionally, a methodology is being introduced which modifies a part of the agent's logic and reuses it in another game. The proposed algorithm has been implemented in the GVG-AI competition's framework and more specifically for the Zelda and Portals games. The obtained results, in terms of average score and win percentage, seem quite satisfactory and highlight the advantages of the suggested technique, especially when compared to a rolling horizon GA implementation of the aforementioned framework; firstly, the agent is efficient at various levels (different world topologies) after being trained in only one of them and secondly, the agent may be generalized to play more games of the same category.

Social collaborative filtering recommender systems extend the traditional user-to-item interaction with explicit user-to-user relationships, thereby allowing for a wider exploration of correlations among users and items, that potentially lead to better recommendations. A number of methods have been proposed in the direction of exploring the social network, either locally (i.e. the vicinity of each user) or globally. In this paper, we propose a novel methodology for collaborative filtering social recommendation that tries to combine the merits of both the aforementioned approaches, based on the soft-clustering of the Friend-of-a-Friend (FoaF) network of each user. This task is accomplished by the non-negative factorization of the adjacency matrix of the FoaF graph, while the edge-centric logic of the factorization algorithm is ameliorated by incorporating more general structural properties of the graph, such as the number of edges and stars, through the introduction of the exponential random graph models. The preliminary results obtained reveal the potential of this idea.

The popularity of recommendation systems has made them a substantial component of many applications and projects. This work proposes a framework for package recommendations that try to meet users' preferences as much as possible through the satisfaction of several criteria. This is achieved by modeling the relation between the items and the categories these items belong to aiming to recommend to each user the top-k packages which cover their preferred categories and the restriction of a maximum package cost. Our contribution includes an optimal and a greedy solution. The novelty of the optimal solution is that it combines the collaborative filtering predictions with a graph based model to produce recommendations. The problem is expressed through a minimum cost flow network and is solved by integer linear programming. The greedy solution performs with a low computational complexity and provides recommendations which are close to the optimal solution. We have evaluated and compared our framework with a baseline method by using two popular recommendation datasets and we have obtained promising results on a set of widely accepted evaluation metrics.

In this chapter, we focus on recommender systems that are enhanced with social information in the form of trust statements between their users. The trust information may be processed in a number of ways, including the random walks in the social graph, where every step in the walk is chosen almost uniformly at random from the available choices. Although this strategy yields satisfactory results in terms of the novelty and the diversity of the produced recommendations, it exhibits poor accuracy because it does not fully exploit the similarity information among users and items. Our work tries to model user-to-user and user-to-item relation as a probability distribution using a novel approach based on Rejection Sampling in order to decide its next step (biased random walk). Some initial results on reference datasets indicate that a satisfying trade-off among accuracy, novelty, and diversity is achieved.

Most recommender systems have too many items to propose to too many users based on limited information. This problem is formally known as the sparsity of the ratings' matrix, because this is the structure that holds user preferences. This paper outlines a Collaborative Filtering Recommender System that tries to amend this situation. After applying Singular Value Decomposition to reduce the dimensionality of the data, our system makes use of a dynamic Artificial Neural Network architecture with boosted learning to predict user ratings. Furthermore we use the concept of k-separability to deal with the resulting noisy data, a methodology not yet tested in Recommender Systems. The combination of these techniques applied to the MovieLens datasets seems to yield promising results.

Most recommender systems usually have too many items to recommend to too many users using limited information. This problem is formally known as the sparsity of the ratings' matrix, because this is the structure that holds user preferences. This article outlines a collaborative recommender system, that tries to amend this situation. The system is built around the notion of k-separability combined with a constructive neural network algorithm.

PDAs and other handheld devices are commonly used for processing private or otherwise secret information. Their increased usage along with their networking capabilities raises security considerations for the protection of the sensitive information they contain and their communications.