A5037831162- Advanced Graph Neural Networks
- Recommender Systems and Techniques
- Topic Modeling
- Complex Network Analysis Techniques
- Multimodal Machine Learning Applications
- Text and Document Classification Technologies
- Advanced Image and Video Retrieval Techniques
- Natural Language Processing Techniques
- Domain Adaptation and Few-Shot Learning
- Sentiment Analysis and Opinion Mining
- Spam and Phishing Detection
- Advanced Bandit Algorithms Research
- Misinformation and Its Impacts
- COVID-19 epidemiological studies
- Image Retrieval and Classification Techniques
- Cancer Immunotherapy and Biomarkers
- Human Mobility and Location-Based Analysis
- Caching and Content Delivery
- Immunotherapy and Immune Responses
- Graph Theory and Algorithms
- Data-Driven Disease Surveillance
- CAR-T cell therapy research
- Semantic Web and Ontologies
- Machine Learning and Data Classification
- Data Management and Algorithms
Microsoft Research Asia (China)
2020-2025
Beijing University of Posts and Telecommunications
2024-2025
Shandong First Medical University
2022-2024
Shandong Tumor Hospital
2022-2024
Beihang University
2015-2024
Fuzhou University
2024
Hong Kong University of Science and Technology
2024
Weifang Medical University
2022-2024
Peking University
2024
Hong Kong Polytechnic University
2024
Session-based recommendation (SBR) aims to predict the user's next action based on short and dynamic sessions. Recently, there has been an increasing interest in utilizing various elaborately designed graph neural networks (GNNs) capture pair-wise relationships among items, seemingly suggesting design of more complicated models is panacea for improving empirical performance. However, these achieve relatively marginal improvements with exponential growth model complexity. In this paper, we...
Nowadays, it is common for one natural person to join multiple social networks enjoy different kinds of services. Linking identical users across networks, also known as network alignment, an important problem great research challenges. Existing methods usually link identities on the pairwise sample level, which may lead undesirable performance when number available annotations limited. Motivated by isomorphism information, in this paper we consider all a whole and perform alignment from...
Learning informative representations of users and items from the historical interactions is crucial to collaborative filtering (CF). Existing CF approaches usually model solely within Euclidean space. However, sophisticated user-item inherently present highly non-Euclidean anatomy with various types geometric patterns (i.e., tree-likeness cyclic structures). The Euclidean-based models may be inadequate fully uncover intent factors beneath such hybrid-geometry interactions. To remedy this...
Sequential recommendation (SR) aims to model users dynamic preferences from a series of interactions. A pivotal challenge in user modeling for SR lies the inherent variability preferences. An effective is expected capture both long-term and short-term exhibited by users, wherein former can offer comprehensive understanding stable interests that impact latter. To more effectively such information, we incorporate locality inductive bias into Transformer amalgamating its global attention...
During the COVID-19 pandemic, forecasting trends to support planning and response was a priority for scientists decision makers alike. In United States, coordinated by large group of universities, companies, government entities led Centers Disease Control Prevention US Forecast Hub ( https://covid19forecasthub.org ). We evaluated approximately 9.7 million forecasts weekly state-level cases predictions 1–4 weeks into future submitted 24 teams from August 2020 December 2021. assessed coverage...
Image retrieval systems help users to browse and search among extensive images in real time. With the rise of cloud computing, tasks are usually outsourced servers. However, scenario brings a daunting challenge privacy protection as servers cannot be fully trusted. To this end, image-encryption-based privacy-preserving image (PPIR) schemes have been developed, which first extract features from cipher-images, then build models based on these features. Yet, most existing PPIR approaches...
Nowadays, detecting multimodal fake news has emerged as a foremost concern since the widespread dissemination of may incur adverse societal impact. Conventional methods generally focus on capturing linguistic and visual semantics within content, which fall short in effectively distinguishing heightened level meticulous fabrications. Recently, external knowledge is introduced to provide valuable background facts complementary facilitate detection. Nevertheless, existing knowledge-enhanced...
In the circumstance of social big data, sentiment analysis is attracting increasing attention for its capacity in understanding individuals' attitudes and feelings. Traditional methods focus on single modality become ineffective as enormous data are emerging websites with multiple manifestations. this article, multimodal learning approaches proposed to capture relations between image text, which only stay at region level ignore fact that channels also closely correlated semantic information....
The wide spread of fake news has caused serious societal issues. We propose a subgraph reasoning paradigm for detection, which provides crystal type explainability by revealing subgraphs the propagation network are most important verification, and concurrently improves generalization discrimination power graph-based detection models removing task-irrelevant information. In particular, we reinforced generation method, perform fine-grained modeling on generated developing Hierarchical...
Session-based recommendation (SBR) aims to predict the user next action based on ongoing sessions. Recently, there has been an increasing interest in modeling preference evolution capture fine-grained interests. While latent preferences behind sessions drift continuously over time, most existing approaches still model temporal session data discrete state spaces, which are incapable of capturing and result sub-optimal solutions. To this end, we propose Graph Nested GRU ordinary differential...
Cross-Domain Recommendation (CDR) is capable of incorporating auxiliary information from multiple domains to advance recommendation performance. Conventional CDR methods primarily rely on overlapping users, whereby knowledge conveyed between the source and target identities belonging same natural person. However, such a heuristic assumption not universally applicable due an individual may exhibit distinct or even conflicting preferences in different domains, leading potential noises. In this...
Abstract Background Tissue-resident memory T (T RM ) cells can reside in the tumor microenvironment and are considered primary response to immunotherapy. Heterogeneity functional status spatial distribution may contribute controversial role of but we know little about it. Methods Through multiplex immunofluorescence (mIF) (CD8, CD103, PD-1, Tim-3, GZMB, CK), quantity location cell subsets were recognized tissue from 274 patients with NSCLC after radical surgery. By integrating multiple...
Image-based survival prediction through deep learning techniques represents a burgeoning frontier aimed at augmenting the diagnostic capabilities of pathologists. However, directly applying existing models to may not be panacea due inherent complexity and sophistication whole slide images (WSIs). The intricate nature high-resolution WSIs, characterized by sophisticated patterns noise, presents significant challenges in terms effectiveness trustworthiness. In this paper, we propose CTUSurv,...
As an effective way of learning node representations in networks, network embedding has attracted increasing research interests recently. Most existing approaches use shallow models and only work on static networks by extracting local or global topology information each as the algorithm input. It is challenging for such to learn a desirable representation incomplete graphs with large number missing links dynamic new nodes joining in. even them deeply fuse other types data properties into...
Nowadays, it is common for one natural person to join multiple social networks enjoy different services. Linking identical users across networks, also known as the User Identity Linkage (UIL), an important problem of great research challenges and practical value. Most existing UIL models are supervised or semi-supervised a considerable number manually matched user identity pairs required, which costly in terms labor time. In addition, methods generally rely heavily on some discriminative...
Hao Wang, Bing Liu, Chaozhuo Li, Yan Yang, Tianrui Li. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.
The representation learning on textual graph is to generate low-dimensional embeddings for the nodes based individual features and neighbourhood information. Recent breakthroughs pretrained language models neural networks push forward development of corresponding techniques. existing works mainly rely cascaded model architecture: are independently encoded by at first; aggregated afterwards. However, above architecture limited due independent modeling features. In this work, we propose...
Abstract Background Anti-PD-(L)1 immunotherapy has been recommended for non-small cell lung cancer (NSCLC) patients with lymph node metastases (LNM). However, the exact functional feature and spatial architecture of tumor-infiltrating CD8 + T cells remain unclear in these patients. Methods Tissue microarrays (TMAs) from 279 IA-IIIB NSCLC samples were stained by multiplex immunofluorescence (mIF) 11 markers (CD8, CD103, PD-1, Tim3, GZMB, CD4, Foxp3, CD31, αSMA, Hif-1α, pan-CK). We evaluated...
The advent of large language models (LLMs) presents both opportunities and challenges for the information retrieval (IR) community. On one hand, LLMs will revolutionize how people access information, meanwhile techniques can play a crucial role in addressing many inherent limitations LLMs. other there are open problems regarding collaboration generation, potential risks misinformation, concerns about cost-effectiveness. To seize critical moment development, it calls joint effort from...