Understanding cartoon emotion using integrated deep neural network on large dataset

AnyFace++: Deep Multi-Task, Multi-Domain Learning for Efficient Face AI

Accurate face detection and subsequent localization of facial landmarks are mandatory steps in many computer vision applications, such as emotion recognition, age estimation, and gender identification. Thanks to advancements in deep learning, numerous facial applications have been developed for human faces. However, most have to employ multiple models to accomplish several tasks simultaneously. As a result, they require more memory usage and increased inference time. Also, less attention is paid to other domains, such as animals and cartoon characters. To address these challenges, we propose an input-agnostic face model, AnyFace++, to perform multiple face-related tasks concurrently. The tasks are face detection and prediction of facial landmarks for human, animal, and cartoon faces, including age estimation, gender classification, and emotion recognition for human faces. We trained the model using deep multi-task, multi-domain learning with a heterogeneous cost function. The experimental results demonstrate that AnyFace++ generates outcomes comparable to cutting-edge models designed for specific domains.

Emotion classification of Indonesian Tweets using Bidirectional LSTM

Emotion classification can be a powerful tool to derive narratives from social media data. Traditional machine learning models that perform emotion classification on Indonesian Twitter data exist but rely on closed-source features. Recurrent neural networks can meet or exceed the performance of state-of-the-art traditional machine learning techniques using exclusively open-source data and models. Specifically, these results show that recurrent neural network variants can produce more than an 8% gain in accuracy in comparison with logistic regression and SVM techniques and a 15% gain over random forest when using FastText embeddings. This research found a statistical significance in the performance of a single-layer bidirectional long short-term memory model over a two-layer stacked bidirectional long short-term memory model. This research also found that a single-layer bidirectional long short-term memory recurrent neural network met the performance of a state-of-the-art logistic regression model with supplemental closed-source features from a study by Saputri et al. [8] when classifying the emotion of Indonesian tweets.

AccessComics2: Understanding the User Experience of an Accessible Comic Book Reader for Blind People with Textual Sound Effects

For people with visual impairments, many studies have been conducted to improve the accessibility of various types of images on the web. However, the majority of the work focused on photos or graphs. In this study, we propose AccessComics, an accessible digital comic book reader for people with visual impairments. To understand the accessibility of existing platforms, we first conducted a formative online survey with 68 participants who are blind or have low vision asking about their prior experiences with audiobooks and eBooks. Then, to learn the implications of designing an accessible comic book reader for people with visual impairments, we conducted an interview study with 8 participants and collected feedback about our system. Considering our findings that a brief description of the scene and sound effects are desired when listening to comic books, we conducted a follow-up study with 16 participants (8 blind, 8 sighted) to explore how to effectively provide scene descriptions and sound effects, generated based on the onomatopoeia and mimetic words that appear in comics. Then we assessed the impact of the overall reading experience and if it differs depending on the user group. The results show that the presence of scene descriptions was perceived to be useful for concentration and understanding the situation, while the sound effects were perceived to make the book reading experience more immersive and realistic. Based on the findings, we suggest design implications specifying features that future accessible comic book readers should support.

A deep neural network-based approach for fake news detection in regional language

Purpose

The current natural language processing algorithms are still lacking in judgment criteria, and these approaches often require deep knowledge of political or social contexts. Seeing the damage done by the spreading of fake news in various sectors have attracted the attention of several low-level regional communities. However, such methods are widely developed for English language and low-resource languages remain unfocused. This study aims to provide analysis of Hindi fake news and develop a referral system with advanced techniques to identify fake news in Hindi.

Design/methodology/approach

The technique deployed in this model uses bidirectional long short-term memory (B-LSTM) as compared with other models like naïve bayes, logistic regression, random forest, support vector machine, decision tree classifier, kth nearest neighbor, gated recurrent unit and long short-term models.

Findings

The deep learning model such as B-LSTM yields an accuracy of 95.01%.

Originality/value

This study anticipates that this model will be a beneficial resource for building technologies to prevent the spreading of fake news and contribute to research with low resource languages.

The use of Big Data Analytics in healthcare

The introduction of Big Data Analytics (BDA) in healthcare will allow to use new technologies both in treatment of patients and health management. The paper aims at analyzing the possibilities of using Big Data Analytics in healthcare. The research is based on a critical analysis of the literature, as well as the presentation of selected results of direct research on the use of Big Data Analytics in medical facilities. The direct research was carried out based on research questionnaire and conducted on a sample of 217 medical facilities in Poland. Literature studies have shown that the use of Big Data Analytics can bring many benefits to medical facilities, while direct research has shown that medical facilities in Poland are moving towards data-based healthcare because they use structured and unstructured data, reach for analytics in the administrative, business and clinical area. The research positively confirmed that medical facilities are working on both structural data and unstructured data. The following kinds and sources of data can be distinguished: from databases, transaction data, unstructured content of emails and documents, data from devices and sensors. However, the use of data from social media is lower as in their activity they reach for analytics, not only in the administrative and business but also in the clinical area. It clearly shows that the decisions made in medical facilities are highly data-driven. The results of the study confirm what has been analyzed in the literature that medical facilities are moving towards data-based healthcare, together with its benefits.

An Enhanced Deep Convolutional Neural Network for Classifying Indian Classical Dance Forms

Indian classical dance (ICD) classification is an interesting subject because of its complex body posture. It provides a stage to experiment with various computer vision and deep learning concepts. With a change in learning styles, automated teaching solutions have become inevitable in every field, from traditional to online platforms. Additionally, ICD forms an essential part of a rich cultural and intangible heritage, which at all costs must be modernized and preserved. In this paper, we have attempted an exhaustive classification of dance forms into eight categories. For classification, we have proposed a deep convolutional neural network (DCNN) model using ResNet50, which outperforms various state-of-the-art approaches. Additionally, to our surprise, the proposed model also surpassed a few recently published works in terms of performance evaluation. The input to the proposed network is initially pre-processed using image thresholding and sampling. Next, a truncated DCNN based on ResNet50 is applied to the pre-processed samples. The proposed model gives an accuracy score of 0.911.