A5028740930- Advanced Graph Neural Networks
- Recommender Systems and Techniques
- Complex Network Analysis Techniques
- Topic Modeling
- Caching and Content Delivery
- Image and Video Quality Assessment
- Advanced Bandit Algorithms Research
- Opinion Dynamics and Social Influence
- Misinformation and Its Impacts
- Spam and Phishing Detection
- Human Mobility and Location-Based Analysis
- Ethics and Social Impacts of AI
- Image Retrieval and Classification Techniques
- Video Analysis and Summarization
- Adversarial Robustness in Machine Learning
- Cybercrime and Law Enforcement Studies
- Engineering Applied Research
- Music and Audio Processing
- Multimodal Machine Learning Applications
- Data Quality and Management
- Hate Speech and Cyberbullying Detection
- Data Stream Mining Techniques
- Bioinformatics and Genomic Networks
- Functional Brain Connectivity Studies
- Speech and Audio Processing
Ulsan National Institute of Science and Technology
2022-2024
Georgia Institute of Technology
2022-2023
Hanyang University
2015-2022
Korea Maritime and Ocean University
2012
There were fierce debates on whether the non-linear embedding propagation of GCNs is appropriate to GCN-based recommender systems. It was recently found that linear shows better accuracy than propagation. Since this phenomenon discovered especially in systems, it required we carefully analyze linearity and non-linearity issue. In work, therefore, revisit issues i) which or ii) factors users/items decide linearity/non-linearity We propose a novel Hybrid method collaborative filtering (HMLET,...
Online misinformation poses a global risk with significant real-world consequences. To combat misinformation, current research relies on professionals like journalists and fact-checkers for annotating debunking false information, while also developing automated machine learning methods detecting misinformation. Complementary to these approaches, recent has increasingly concentrated utilizing the power of ordinary social media users, a.k.a. “the crowd”, who act as eyes-on-the-ground...
We address the multimedia recommendation problem, which utilizes items' multimodal features, such as visual and textual modalities, in addition to interaction information. While a number of existing recommender systems have been developed for this we point out that none these methods individually capture influence each modality at level. More importantly, experimentally observe learning procedures works fail preserve intrinsic modality-specific properties items. To above limitations, propose...
The problem of community-level information pathway prediction (CLIPP) aims at predicting the transmission trajectory content across online communities. A successful solution to CLIPP holds significance as it facilitates distribution valuable a larger audience and prevents proliferation misinformation. Notably, solving is non-trivial inter-community relationships influence are unknown, spread multi-modal, new communities appear over time. In this work, we address by collecting large-scale,...
We investigate how to address the shortcomings of popular One-Class Collaborative Filtering (OCCF) methods in handling challenging “sparse” dataset one-class setting (e.g., clicked or bookmarked), and propose a novel graph-theoretic OCCF approach, named as gOCCF, by exploiting both positive preferences (derived from rated items) well negative unrated items). In capturing bipartite graph, further, we apply graph shattering theory determine right amount use. Then, develop suite graph-based...
Motivated by a success of generative adversarial networks (GAN) in various domains including information retrieval, we propose novel signed network embedding framework, ASiNE, which represents each node given as low-dimensional vector based on the learning. To do this, first design generator G+ and discriminator D+ that consider positive edges, well G - D- negative edges: (1) G+/G- aim to generate most indistinguishable fake positive/negative respectsupively; (2) D+/D discriminate between...
The directed network embedding problem is to represent the nodes in a given as embeddings (i.e., low-dimensional vectors) that preserve asymmetric relationships between nodes. While number of approaches have been developed for this problem, we point out existing commonly face difficulties accurately preserving proximities sparse containing large low out- and in-degree In paper, focus on addressing intrinsic difficulty caused by lack information. We first introduce concept virtual negative...
DBpia is the largest digital-bibliography service provider in Korea. It provides several convenience functions for researchers. users (i.e., researchers) can search papers via routes such as publications, publishers, authors, and keywords. Although researchers exploit functions, they may still have a number of results candidate to read. Therefore, it crucial provide function recommending most relevant an individual user. In this paper, we (1) discuss methods with four datasets context paper...
Online news providers such as Google News, Bing and NAVER News collect a large number of articles from variety presses distribute these to users via their portals. Dynamic nature domain causes the problem information overload that makes it difficult for user find her preferable articles. Motivated by this situation, Corp., largest portal company in South Korea, identified four design considerations (DCs) recommendation reflect unique characteristics domain. In paper, we introduce large-scale...
Social recommender systems (SocialRS) simultaneously leverage the user-to-item interactions as well user-to-user social relations for task of generating item recommendations to users. Additionally exploiting is clearly effective in understanding users' tastes due effects homophily and influence. For this reason, SocialRS has increasingly attracted attention. In particular, with advance graph neural networks (GNN), many GNN-based methods have been developed recently. Therefore, we conduct a...
As the number of TV channels increases, it is becoming important to recommend shows that users prefer watch. To this end, we investigate inherent characteristics implicit feedback given in show domain, and identify challenges for building an effective recommendation. Based on unique characteristics, define a user's watchable interval, most novel concept understanding users' true preferences. In order reflect new into recommendation, propose framework based collaborative filtering. Our...
The goal of temporal knowledge graph embedding (TKGE) is to represent the entities and relations in a given (TKG) as low-dimensional vectors (i.e., embeddings), which preserve both semantic information dynamics factual information. In this paper, we posit that intrinsic difficulty existing TKGE methods lies lack KG snapshots with timestamps, each contains facts co-occur at specific timestamp. To address challenge, propose novel self-supervised approach, THOR (Three-tower grapH cOnvolution...
Collaborative filtering (CF) methods for recommendation systems have been extensively researched, ranging from matrix factorization and autoencoder-based to graph filtering-based methods. Recently, lightweight that require almost no training recently proposed reduce overall computation. However, existing still room improve the trade-offs among accuracy, efficiency, robustness. In particular, there are well-designed closed-form studies balanced CF in terms of aforementioned trade-offs. this...
The problem of career trajectory prediction (CTP) aims to predict one's future employer or job position. While several CTP methods have been developed for this problem, we posit that none these (1) jointly considers the mutual ternary dependency between three key units (i.e., user, position, and company) a (2) captures characteristic shifts in over time, leading an inaccurate understanding movement patterns labor market. To address above challenges, propose novel solution, named as CAPER,...
The goal of directed network embedding is to represent the nodes in a given as embeddings that preserve asymmetric relationships between nodes. While number methods have been proposed, we empirically show existing lack out-of-distribution generalization abilities against degree-related distributional shifts. To mitigate this problem, propose ODIN (Out-of-Distribution Generalized Directed Network Embedding), new NE method where model multiple factors formation edges. Then, for each node,...
The problem of signed network embedding (SNE) aims to represent nodes in a given as low-dimensional vectors. While several SNE methods based on graph convolutional networks (GCN) have been proposed, we point out that they significantly rely the assumption decades-old balance theory always holds real world. To address this limitation, propose novel GCN-based approach, named TrustSGCN, which measures trustworthiness edge signs for high-order relationships inferred by and corrects incorrect...
Signed networks allow us to model conflicting relationships and interactions, such as friend/enemy support/oppose. These signed interactions happen in real-time. Modeling dynamics of is crucial understanding the evolution polarization network enabling effective prediction structure (i.e., link signs) future. However, existing works have modeled either (static) or dynamic (unsigned) but not networks. Since both sign inform graph different ways, it non-trivial how combine two features. In this...
In this paper, we address the personalized node ranking (PNR) problem for signed networks, which aims to rank nodes in an order most relevant a given seed network. The recently-proposed PNR methods introduce concept of random surfer, denoted as SRSurfer, that performs score propagation between using balance theory. However, real settings edge relationships often do not strictly follow rules Therefore, SRSurfer-based frequently perform incorrect nodes, thereby degrading accuracy PNR. To...
As the number of TV shows increases, designing recommendation systems to provide users with their favorable becomes more important. In a show domain, watching (i.e., giving implicit feedback show) among broadcast at same time frame implies that currently is winner in competition others losers). However, previous studies, such notion limited competitions has not been considered estimating user's preferences for shows. this paper, we propose new framework take into account based on pair-wise...