Audio-driven talking head generation requires precise synchronization between facial animations and audio signals. This paper introduces ATL-Diff, a novel approach addressing synchronization limitations while reducing noise and computational costs. Our framework features three key components: a Landmark Generation Module converting audio to facial landmarks, a Landmarks-Guide Noise approach that decouples audio by distributing noise according to landmarks, and a 3D Identity Diffusion network preserving identity characteristics. Experiments on MEAD and CREMA-D datasets demonstrate that ATL-Diff outperforms state-of-the-art methods across all metrics. Our approach achieves near real-time processing with high-quality animations, computational efficiency, and exceptional preservation of facial nuances. This advancement offers promising applications for virtual assistants, education, medical communication, and digital platforms. The source code is available at: https://github.com/sonvth/ATL-Diff

Abstract Due to the exponential growth of online information, users are often welcomed with a huge range of products and services along with descriptions, reviews, and comments. Although this information available to users is valuable, at the same time, massive data sources confuse them to retrieve desired content, which is known as information overload. Recommender systems are examined as effective tools that play a vital role in filtering information and ultimately addressing the information overload problem. Various online platforms use recommendation systems to provide users with more relevant and personalized content. With the remarkable success of deep learning in the field of artificial intelligence, it procures much attention in the recommendation research area in recent years. The exiting literature on recommender system research commonly distinguishes between two main directions: rating prediction and top-N ranking. In this survey paper, we examine deep learning methodologies in the context of both rating prediction and top-N ranking recommendation approaches. Additionally, we investigate pre- and post-modeling critiques of recommender systems and provide insights into exiting benchmark datasets, feature learning analysis, and evaluation measuring techniques. In the end, we highlight the new generation recommender system trend with respective future research directions.

Textual reviews contain fine-grained information that can effectively infer user preferences over the items. Accordingly, the latest studies in the field of recommender systems exploit content-rich review texts to complement user and item representations and improve the ability to make personalized recommendations. Furthermore, the interactive deep learning mechanism can better model the user-item interaction from fine-grained textual reviews compared to traditional recommendation approaches improving the predictive performances. Therefore, it becomes important to investigate the design of existing deep learning methods for review-based recommender systems and innovate to make them capable of meeting desired recommendation schemes. The purpose of this research is to explore the performance of deep learning networks and deep transformer models in review-based recommender systems. In this paper, we conduct a compendious survey of the latest deep learning techniques in review-based recommender systems. Then investigation calls to employ and analyze deep transformer models for the review-based recommender systems. The wide range of experiments shows that deep transformer models can extract interpretable and relevant user/item representations than traditional deep learning networks. The findings indicate that the best deep transformer performance gains the maximum relative improvement (RMSE=4.6%, MAE=7.4%) with Amazon electronics, compared to the best outcome from traditional deep learning networks. In the end, this article highlights research gaps and outlines research opportunities for future research in this field.

Abstract Cross-domain recommendations (CDR) present a viable solution and are increasingly used to address the cold-start problem. Recently, CDR methods are utilizing deep models to generate latent preferences from context vectors or rating matrices and transfer these preferences between domains. However, many of these models focus on learning latent preferences using domain-related information and often disregard preference patterns from the contrary domain. Incorporating the contrary domain preference patterns into deep models can improve the generation of more effective latent representations. Moreover, existing CDR models face challenges in effectively transferring mapped preferences between domains due to the large features disparity between them. In this study, we tackle these problems and present a novel Deep Shared Learning and Attentive Domain Mapping (DSAM) approach for CDR. Specifically, we propose a variant of Long Short-Term Memory (LSTM) called shared learning LSTM, which incorporates the learning of cross-domain preference patterns alongside domain-specific user/item embeddings derived from textual reviews to dynamically generate shared contextual representations in each domain. We further exploit a multi-head self-attentive network to match item-specific knowledge from the source and target domains into different subspaces. We aggregate this learned knowledge to predict rating scores for cold-start users in the target domain. We efficiently optimize this framework in an end-to-end fashion. Experimental results on five real-world datasets demonstrate the effectiveness of our proposed approach against various groups of recommendation models. Additionally, we provide insights to help understand the model architecture and its robustness in handling cold-start users.

Abstract Learning can generally be categorised into three domains, which include cognitive (thinking), affective (emotions or feeling) and psychomotor (physical or kinesthetic). In the learner model, acknowledging the affective aspects of learning is important for a range of learner outcomes, including motivation, persistence, and engagement. Learners’ affective states can be detected using physical (e.g. cameras) and physiological sensors (e.g., EEG) in online learning. Although these detectors demonstrate high accuracy, they raise privacy concerns for learners and present challenges in deploying them on a large scale to larger groups of students or in classroom settings. Consequently, researchers have designed an alternative method that can recognise students’ affective states at any point during online learning from their interaction with a computer-based learning platform (i.e. intelligent tutoring systems) without using any sensors. Existing sensor-free affect detectors however, are less accurate and not directly generalisable to other domains and systems. This research focuses on developing generalisable sensor-free affect detectors to identify students’ frustration during online learning using machine learning classifiers. The detectors were built by identifying minimal optimal features associated with frustration from the high-dimensional feature space through a series of experiments on a real-world students’ affective dataset, which are generalisable across various learning platforms and domains. To evaluate their accuracy and generalisability, the detectors’ performance was validated on two independent datasets collected from different educational institutions. The experimental results show that cost-sensitive Bayesian classifiers can achieve higher affect detection accuracies with a small number of generalisable features compared to other classifiers.

Affective Tutoring Systems (ATS) is a next-generation Intelligent Tutoring System (ITS) that can detect learners' affective states and provide affective interventions to encourage learners and improve their motivation to learn using artificial intelligence techniques. However, little research has been done on creating an educational affective lexicon in English that those systems can use for providing affective intervention. There needs to be more evidence as to which phrases learners may like to receive from an ATS. This study investigates popular congratulating and encouraging phrases to build a comprehensive affective educational lexicon in English for ATS. Firstly, we examined and collected phrases by reviewing existing learning support systems from the literature that provide affective feedback to learners for constructing an affective lexicon. Secondly, 84 students from various qualifications and backgrounds evaluated the collected phrases by choosing the most popular ones in the congratulating and encouraging dimensions. The lexicon currently consists of 43 encouraging and 32 congratulating phrases categorised by popularity. The findings from this study will augment the ATS to provide better affective tutoring support for students by choosing more constructive and popular affective phrases.

The rapid development of VR technology in training and learning is based on the assumption that it is beneficial for skill training within an immersive environment. However, extra cognitive load may be induced due to the additional sensory information and hence learning ability might be affected. In this study, we examined and compared the impact of cognitive load and task performance in real-world and VR environments through an empirical quadrant model. Forty-six participants completed the tasks with and without the secondary task in realworld and VR environments. The detection response task (DRT), as the secondary task, was adopted to estimate cognitive load based on response time and omission rate. No statistically significant differences were found in cognitive load and task performance in the comparison of VR and non-VR environment settings. There was an encouraging trend observed that VR environments have some advantages over the real-world, such as a higher level of immersion, which suggests that VR can benefit trainees with improved concentration levels and task performance. As evidenced by the variation in performance between females and males in our study, it appears that females tend to perform less well in VR environments, with a slightly higher cognitive load.

The objective of this study was to investigate if the Project-Based Collaborative Learning (PBCL) approach could positively impact on a student's performance within an introductory programming subject. PBCL is a student-centred approach that allows students to collaboratively engage in an authentic complex project that facilitates students developing knowledge and skills while systematically completing learning tasks that combine to achieve the project. In this study, PBCL was implemented within a postgraduate web-based programming subject for students who mostly did not have a prior history with programming. This study used inferential descriptive statistical analysis to compare the students' performance with a prior offering where students completed all tasks as individuals. The size of the study was 799 students over three years (n=338 in 2019, 238 in 2020 and 223 in 2021). The results indicate a statistically significant improvement, with failure rates declining by 7% and average overall performance improving by 5.8%. As there were a substantial number of female students an analysis based on gender was also possible (38% female students). This approach indicated a statistically significant improvement for female students, with failure rates declining by 10% and average overall performance improving by 8.3%. The results of this study provide evidence of the effectiveness of the PBCL approach to engage and improve a student's, especially female students' learning within an introductory programming subject.

Challenges arise when extracurricular programs aren't aligned to the local curriculum and when courses are inflexible regarding the students learning pace. Modification of pre-existing programs can be achieved with enough domain knowledge, which can solve relevant gaps for digital technology Curriculums. The process of modifying such programs can bring to light other challenges like mixing very different technology, encouraging self-efficacy and engagement, and pacing learning to individual needs. This pilot-study highlights some of the challenges faced in using a program not designed for the local audience; the steps taken to change the program for the Australian digital technology curriculum; and identifies some areas of further research in extracurricular activity development for a digital technology curriculum.

Emotion recognition plays an indispensable role in human–machine interaction system. The process includes finding interesting facial regions in images and classifying them into one of seven classes: angry, disgust, fear, happy, neutral, sad, and surprise. Although many breakthroughs have been made in image classification, especially in facial expression recognition, this research area is still challenging in terms of wild sampling environment. In this paper, we used multi-level features in a convolutional neural network for facial expression recognition. Based on our observations, we introduced various network connections to improve the classification task. By combining the proposed network connections, our method achieved competitive results compared to state-of-the-art methods on the FER2013 dataset.

A virtual class delivers most of its content through multimedia learning objects. To support such multimedia learning objects, its multimedia data service system should have a capacity to serve the growing number of clients and new data. A streaming server transfers multimedia files to clients from a repository of files in real time. The server must guarantee concurrent and uninterrupted delivery of each video stream requested by clients. To provide efficient services, many stream servers adopt multi-processors, sufficient memory, and RAID or SAN in their systems. In this paper, we propose a Linux-based parallel media streaming server. This system uses a unique striping policy to distribute multimedia files into the parallel storage nodes. If a service request occurs, each storage node transmits striped files concurrently to the client. Its performance is better than the existing single media streaming service.