Filtered by themeDeep learning12 papersclear ✕

Improving Language Understanding by Generative Pre-TrainingAlec RadforNatural language understanding comprises a wide range of diverse tasks such as textual entailment, question answering, semantic similarity assessment, and document classification. Although large unlabeled text corpora are abundant, labeled data for learning these specific tasks is scarce, making it challenging for discriminatively trained models to perform adequately. We demonstrate that large gains on these tasks can be realized by generative pre-training of a language model on a diverse corpus of unlabeled text, followed by discriminative fine-tuning on each specific task. In contrast to previous approaches, we make use of task-aware input transformations during fine-tuning to achieve effective transfer while requiring minimal changes to the model architecture. We demonstrate the effectiveness of our approach on a wide range of benchmarks for natural language understanding. Our general task-agnostic model outperforms discriminatively trained models that use architectures specifically crafted for each task, significantly improving upon the state of the art in 9 out of the 12 tasks studied. For instance, we achieve absolute improvements of 8.9% on commonsense reasoning (Stories Cloze Test), 5.7% on question answering (RACE), and 1.5% on textual entailment (MultiNLI).

UX of AI & PersonalizationResearch Papers

Harnessing AI potential in E-Commerce: improving user engagement and sales through deep learning-based product recommendationsZhang · 2024Artificial intelligence (AI) has become a game-changing influence in e-commerce, reshaping the way businesses inter­ act with customers and increasing operational efficiency. In today’s digital era, AI technologies are being progressively embedded into e-commerce platforms to enhance the user experience, improve processes, and drive business growth. One of the primary areas where AI demonstrates its potential is in personalized product recommendations. This study explores the development of AI-driven recommendations regarding product design, sales, and customer experience in e-commerce environments. The research employs a quantitative approach to explore how AI technologies can enhance various facets of online retail. Using purposive sampling, 439 consumers and 356 sellers from diverse regions of China participated in an online survey designed to gather insights into their interactions with AI-driven recommendations. Data analysis was rigorously conducted using Lisrel 10.20 and Smart PLS 3 software, focusing on validating hypotheses related to AI’s influence on consumer behavior and business outcomes. Key findings include AI’s significant role in boosting productiv­ ity, enhancing sales figures, and enriching consumer experiences through personalized recommendations. This research provides a deeper insight into AI’s potential in e-commerce and provides recommendations for businesses focused on using AI technology to increase user engagement and increase sales in digital marketing.

UX of AI & PersonalizationResearch Papers

Next-Gen Human-Computer Interaction: A HybridRavindra Changal · 2024In the rapidly evolving landscape of human- computer interaction (HCI), the demand for personalized and adaptive user experiences has grown exponentially. To meet this demand, Research propose a groundbreaking approach leveraging the fusion of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) architectures. This hybrid LSTM-CNN model is designed to enhance adaptability, responsiveness, and user engagement across various interactive platforms.Traditional HCI models often struggle to effectively capture the dynamic nature of user behavior and preferences. By integrating LSTM and CNN, This model achieves a synergistic blend of temporal and spatial feature extraction capabilities. The LSTM component excels in capturing sequential dependencies and long-term patterns, enabling the system to learn from past interactions and anticipate future user actions. Meanwhile, the CNN component efficiently processes spatial information, extracting meaningful features from multimedia inputs such as images, videos, and text.One of the key strengths of proposed model is its adaptability to diverse user contexts and preferences. Through continuous learning and adaptation, it dynamically adjusts its interface, content, and interaction patterns to match the evolving needs and preferences of individual users. Moreover, the hybrid architecture enables real-time processing of multimodal inputs, facilitating seamless interaction across a wide range of devices and platforms.In experimental evaluations, Hybrid LSTM-CNN model demonstrated superior performance compared to baseline methods in terms of user satisfaction, engagement, and task completion rates. Furthermore, it exhibited robustness and scalability, making it suitable for deployment in real-world applications across domains such as e-commerce, entertainment, education, and healthcare.In summary, The proposed approach represents a significant advancement in HCI research, paving the way for next-generation interactive systems that deliver highly adaptive and personalized user experiences.

UX of AI & PersonalizationResearch Papers

An empirical study of AI techniques in mobile applications ✩Haoye Tian b, Zhijie Wang c et al. · 2024The integration of artificial intelligence (AI) into mobile applications has significantly transformed various domains, enhancing user experiences and providing personalized services through advanced machine learning (ML) and deep learning (DL) technologies. AI-driven mobile apps typically refer to applications that leverage ML/DL technologies to perform key tasks such as image recognition and natural language processing. Despite existing research exploring how mobile apps exploit AI techniques, they have the following main limitations: (1) Most existing studies focus on DL-based apps, with limited research on ML-based apps. (2) Existing research typically focuses on investigating the apps and the technologies utilized in the apps, lacking user-level analysis. (3) The number of apps studied is limited, with only 1,000 to 2,000 ML/DL apps identified after filtering. To fill the gap, in this paper, we conducted the most extensive empirical study on AI applications, exploring on-device ML apps, on-device DL apps, and AI service-supported (cloud-based) apps. Our study encompasses 56,682 real- world AI applications, focusing on three crucial perspectives: (1) Application analysis, where we analyze the popularity of AI apps and investigate the update states of AI apps; (2) Framework and model analysis, where we analyze AI framework usage and AI model protection; (3) User analysis, where we examine user privacy protection and user review attitudes. Our study has strong implications for AI app developers, users, and AI R&D. On one hand, our findings highlight the growing trend of AI integration in mobile applications, demonstrating the widespread adoption of various AI frameworks and models. On the other hand, our findings emphasize the need for robust model protection to enhance app security. Additionally, our study highlights the importance of user privacy and presents user attitudes towards the AI technologies utilized in current AI apps. We provide our AI app dataset (currently the most extensive AI app dataset) as an open-source resource for future research on AI technologies utilized in mobile applications.

UX of AI & PersonalizationResearch Papers

Toward an Interactive Reading Experience: Deep Learning Insights and Visual Narratives of Engagement and EmotionJayasankar Santhosh, Akshay Palimar Pai et al. · 2024Engagement and emotion are critical components that significantly influence a reader’s experience during a reading task. Despite the crucial role of engagement and emotions in shaping our reading experience, accurately tracking these dynamic states during actual reading remains a significant challenge. This study bridges this gap by detecting engagement and emotion levels during a reading task by leveraging the power of state-of-the-art deep learning models and investigating the correlations between the engagement levels and emotions. An experiment was conducted involving 18 university students reading 14 documents followed by a questionnaire to rate their levels of engagement, valence, and arousal after reading each document. A Tobii 4C eye-tracker with a pro license along with an Empatica E4 wristband were utilized to record behavioral and physiological data from the participants. A range of deep learning models were utilized for computing the engagement, valence, and arousal values, employing both user-independent and user- dependent methods. Our investigation revealed distinct yet complementary strengths in two deep learning models: Transformer excelled in user-independent detection of engagement and emotion with an accuracy of 80.38% (engagement), 71.28% (arousal) and 73.98% (valence) while ResNet shined in the user-dependent setting with an accuracy of 93.56% (engagement), 90.62% (arousal) and 88.70% (valence) which highlights the interplay between individual differences and reading dynamics. Intriguingly, we observed strong, document-specific correlations between engagement and emotion states, suggesting that different texts evoke unique affective responses. We developed an interactive dashboard visualizing predicted engagement and emotions, offering real-time feedback and personalized learning possibilities. The dashboard features an engagement gauge that displays the reader’s level of engagement based on predicted class probabilities, and an emotion emoji serving as a visual cue that illustrates the predicted emotional state of the reader. This technology can inform the design of dynamic interfaces that adjust to individual reading styles and emotional responses, potentially enhancing comprehension and involvement. INDEX TERMS Digital reading, physiological sensing, eye tracking, deep learning, affective state.

UX of AI & PersonalizationResearch Papers

Recommendation Engines: Traditional vs Deep Learning ApproachesAniket Dhawa, Suranjana Sarkar et al. · 2021With digitization growing by leap and bounds, websites are now overloaded with products and information density, making it challenging for customers to choose between a variety of products. The Recommendation engine enables companies to deliver their customers more personalized and customized products or information efficiently. As a result, the recommendation engine eventually helps in boosting different business performance metrics such as revenues, Click-Through Rates (CTRs), conversions, etc., for these companies significantly. And in the long run, it also optimizes user experience, thus translating to higher customer satisfaction and retention. In this paper, we will highlight some of the contemporary algorithms used in building high-performance recommendation engines, illustrate use cases from Tech Giants that can be applied in the context of the banking and financial services domain and identify major challenges that are open for further research work. Also, in this paper, we will be using recommendation engine and recommender system alternatively as both are synonymous. Index Terms—Recommendation engine, recommender system, Machine Learning, Deep Learning, content based recommenda- tion, collaborative filtering, State of Art Technique (SOAT)

UX of AI & PersonalizationResearch Papers

Deep Learning for Emotion Driven User ExperiencesCarmen Bisogni, Lucia Cascone et al. · 2021Deep knowledge about user characteristics and behaviors opens new and promising landscapes to the User Experience. Thanks to emerging machine learning techniques, a new form of communication chan- nel among users and applications may be exploited for customizing and finely tuning the dynamic behav- ior of applications to the peculiarities of their users. This work investigates the empathic improvement of the User Experiences and exploits inferences on user expressions for activating gaming and entertain- ment events, that are adopted in a cinematic way to create dynamic application behaviors. The presented approach is applied to a third/first-person horror adventure and a classic table game. As already verified in a preliminary phase of the research, user impressions, collected as subjective evaluation in a controlled experiment, are positive and encourage further steps, like the evaluation of other inferences on users. © 2021 Published by Elsevier B.V.

UX of AI & PersonalizationResearch Papers

Deep Learning UI Design Patterns of Mobile AppsPhong Minh Vu, Hung Viet Pham et al. · 2018User interface (UI) is one of the most important components of a mobile app and strongly influences users’ perception of the app. However, UI design tasks are typically manual and time-consuming. This paper proposes a novel approach to (semi)-automate those tasks. Our key idea is to develop and deploy advanced deep learning models based on recurrent neural networks (RNN) and generative adversarial networks (GAN) to learn UI design patterns from mil- lions of currently available mobile apps. Once trained, those models can be used to search for UI design samples given user-provided descriptions written in natural language and generate professional- looking UI designs from simpler, less elegant design drafts.

UX of AI & PersonalizationResearch Papers

Deep Neural Networks for YouTube RecommendationsPaul Covington, Jay Adams et al. · 2016YouTube represents one of the largest scale and most sophis- ticated industrial recommendation systems in existence. In this paper, we describe the system at a high level and fo- cus on the dramatic performance improvements brought by deep learning. The paper is split according to the classic two-stage information retrieval dichotomy: first, we detail a deep candidate generation model and then describe a sepa- rate deep ranking model. We also provide practical lessons and insights derived from designing, iterating and maintain- ing a massive recommendation system with enormous user- facing impact.

UX of AI & PersonalizationResearch Papers

ImageNet Classification with Deep Convolutional Neural NetworksAlex Krizhevsky · 2010We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 dif- ferent classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make train- ing faster, we used non-saturating neurons and a very efficient GPU implemen- tation of the convolution operation. To reduce overfitting in the fully-connected layers we employed a recently-developed regularization method called “dropout” that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

UX of AI & PersonalizationResearch Papers

Classifying Emotional States through EEG-Derived Spectrograms2024Identifying emotional state of a person plays an important role in a multitude of applications such as affective computing, human-computer interaction and most importantly healthcare. Understanding and correctly identifying human emotions can improve mental health assessments, making it possible to obtain personalized treatment plans and increase the user experience across a variety of digital applications. Our work explores the feasibility of emotion classification using EEG signals recorded from multiple users while experiencing various emotions. Working with time series data of the brain activity in various participants, who are experiencing particular emotions over an interval of 15 seconds. To train and evaluate the classification model several kinds of machine learning and deep learning models such as CNN, RNN and LSTM, are employed. We compare general sequential models with image processing models in the task of classifying signal data. Observing the model’s ability to generalize across the population.

Emotional DesignResearch Papers

A Novel User Emotional Interaction Design Model Using Long and Short-Term Memory Networks and Deep Learning2021Emotional design is an important development trend of interaction design. Emotional design in products plays a key role in enhancing user experience and inducing user emotional resonance. In recent years, based on the user's emotional experience, the design concept of strengthening product emotional design has become a new direction for most designers to improve their design thinking. In the emotional interaction design, the machine needs to capture the user's key information in real time, recognize the user's emotional state, and use a variety of clues to finally determine the appropriate user model. Based on this background, this research uses a deep learning mechanism for more accurate and effective emotion recognition, thereby optimizing the design of the interactive system and improving the user experience. First of all, this research discusses how to use user characteristics such as speech, facial expression, video, heartbeat, etc., to make machines more accurately recognize human emotions. Through the analysis of various characteristics, the speech is selected as the experimental material. Second, a speech-based emotion recognition method is proposed. The mel-Frequency cepstral coefficient (MFCC) of the speech signal is used as the input of the improved long and short-term memory network (ILSTM). To ensure the integrity of the information and the accuracy of the output at the next moment, ILSTM makes peephole connections in the forget gate and input gate of LSTM, and adds the unit state as input data to the threshold layer. The emotional features obtained by ILSTM are input into the attention layer, and the self-attention mechanism is used to calculate the weight of each frame of speech signal. The speech features with higher weights are used to distinguish different emotions and complete the emotion recognition of the speech signal. Experiments on the EMO-DB and CASIA datasets verify the effectiveness of the model for emotion recognition. Finally, the feasibility of emotional interaction system design is discussed.

Emotional DesignResearch Papers