Visual content analysis has always been important yet challenging. Thanks to the popularity of social networks, images become an convenient carrier for information diffusion among online users. To understand patterns and different aspects images, we need interpret first. Similar textual content, also carry levels sentiment their viewers. However, from text, where can use easily accessible semantic context information, how extract image remains quite In this paper, propose prediction...

Strong Artificial Intelligence (Strong AI) or General (AGI) with abstract reasoning ability is the goal of next-generation AI. Recent advancements in Large Language Models (LLMs), along emerging field Multimodal (MLLMs), have demonstrated impressive capabilities across a wide range multimodal tasks and applications. Particularly, various MLLMs, each distinct model architectures, training data, stages, been evaluated broad MLLM benchmarks. These studies have, to varying degrees, revealed...

Background The number of electronic cigarette (e-cigarette) users has been increasing rapidly in recent years, especially among youth and young adults. More e-cigarette products have become available, including e-liquids with various brands flavors. Various e-liquid flavors frequently discussed by on social media. Objective This study aimed to examine the longitudinal prevalence mentions liquid (e-liquid) user perceptions Methods We applied a data-driven approach analyze trends macro-level...

Sentiment analysis on large-scale social media data is important to bridge the gaps between contents and real world activities including political election prediction, individual public emotional status monitoring analysis, so on. Although textual sentiment has been well studied based platforms such as Twitter Instagram, of role extensive emoji uses in remains light. In this paper, we propose a novel scheme for with extra attention emojis. We first learn bi-sense embeddings under positive...

Autism spectrum disorder (ASD) is a developmental that significantly impairs patients' ability to perform normal social interaction and communication. Moreover, the diagnosis procedure of ASD highly time-consuming, labor-intensive, requires extensive expertise. Although there exists no known cure for ASD, consensus among clinicians regarding importance early intervention recovery patients. Therefore, benefit autism patients by enhancing their access treatments such as intervention, we aim...

Collaborative Filtering (CF) is widely used in large-scale recommendation engines because of its efficiency, accuracy and scalability. However, practice, the fact that based on CF require interactions between users items before making recommendations, make it inappropriate for new which haven't been exposed to end interact with. This known as cold-start problem. In this paper we introduce a novel approach employs deep learning tackle problem any engine. One most important features proposed...

In recent years, flavored electronic cigarettes (e-cigarettes) have become popular among teenagers and young adults. Discussions about e-cigarettes e-cigarette use (vaping) experiences are prevalent online, making social media an ideal resource for understanding the health risks associated with flavors from users' perspective.This study aimed to investigate potential associations between cigarette liquid (e-liquid) reporting of symptoms using data.A dataset consisting 2.8 million...

Extracting fine-grained relations between entities of interest is great importance to information extraction and large-scale knowledge graph construction. Conventional approaches on relation require an existing start with or sufficient observed samples from each type in the training process. However, such resources are not always available, manual labeling extremely time-consuming requires extensive expertise for specific domains as healthcare bioinformatics. Additionally, distribution often...

We propose a Dynamic Graph-Based Spatial-Temporal Attention (DG-STA) method for hand gesture recognition. The key idea is to first construct fully-connected graph from skeleton, where the node features and edges are then automatically learned via self-attention mechanism that performs in both spatial temporal domains. further leverage spatial-temporal cues of joint positions guarantee robust recognition challenging conditions. In addition, novel mask applied significantly cut down...

Background The rise of the popular e-cigarette, JUUL, has been partly attributed to various teen-friendly e-liquid flavours offered. However, possible health risks associated with each flavour still remain unclear. This research focuses on associations between JUUL and symptoms using social media data from Reddit. Methods Keyword filtering was used obtain 5,746 flavour-related posts 7927 symptom-related June 2015 April 2019 Posts September 2016 were conduct temporal analysis for nine symptom...

In this paper, we propose Discrete Cosin TransFormer (DCFormer) that directly learn semantics from DCT-based frequency domain representation. We first show transformer-based networks are able to representation based on discrete cosine transform (DCT) without compromising the performance. To achieve desired efficiency-effectiveness trade-off, then leverage an input information compression its representation, which highlights visually significant signals inspired by JPEG compression. explore...

Cross-modal attention mechanisms have been widely applied to the image-text matching task. They achieved remarkable improvements thanks their capability of learning fine-grained relevance across different modalities. However, cross-modal models existing methods could be sub-optimal and inaccurate because there is no direct supervision provided during training process. In this work, we propose two novel strategies, namely Contrastive Content Resourcing (CCR) Swapping (CCS) constraints,...

Next item recommendation is a crucial task of session-based recommendation. However, the gap between optimization objective (Binary Cross Entropy) and ranking metric (Mean Reciprocal Rank) has not been well-explored, resulting in sub-optimal recommendations. In this paper, we propose novel function, namely Adjusted-RR, to directly optimize Mean Rank. Specifically, Adjusted-RR adopts Bayesian Average adjust Rank loss with Normal by creating position-aware weights them. plug-and-play that...

Two-tower neural networks are popularly used in review-aware recommender systems, which two encoders separately employed to learn representations for users and items from reviews. However, such an architecture isolates the information exchange between encoders, resulting suboptimal recommendation accuracy. To this end, we propose a novel two-tower style Neural Recommendation with Cross-modality Mutual Attention (NRCMA), bridges user encoder item crossing reviews ratings, order select...

Review-aware Rating Regression (RaRR) suffers the severe challenge of extreme data sparsity as multi-modality interactions ratings accompanied by reviews are costly to obtain. Although some studies semi-supervised rating regression proposed mitigate impact sparse data, they bear risk learning from noisy pseudo-labeled data. In this article, we propose a simple yet effective paradigm, called co-training-teaching ( CoT 2 ), for integrating merits both co-training and co-teaching toward robust...