A5100341816- Topic Modeling
- Natural Language Processing Techniques
- Recommender Systems and Techniques
- Disaster Management and Resilience
- Machine Learning in Healthcare
- Radiomics and Machine Learning in Medical Imaging
- Advanced Graph Neural Networks
- Sentiment Analysis and Opinion Mining
- Misinformation and Its Impacts
- Digital Marketing and Social Media
- Multimodal Machine Learning Applications
- Big Data Technologies and Applications
- Smart Cities and Technologies
- Radiology practices and education
- Generative Adversarial Networks and Image Synthesis
- Job Satisfaction and Organizational Behavior
- Bayesian Methods and Mixture Models
- Biomedical Text Mining and Ontologies
- Facility Location and Emergency Management
- Hate Speech and Cyberbullying Detection
- Domain Adaptation and Few-Shot Learning
- Public Relations and Crisis Communication
- Computational and Text Analysis Methods
- Human Mobility and Location-Based Analysis
- Evacuation and Crowd Dynamics
Huawei Technologies (China)
2023-2025
Huawei Technologies (United Kingdom)
2023
Peking University
2020-2022
Harbin Institute of Technology
2012-2022
Wenzhou Medical University
2011
Illinois Institute of Technology
2006-2009
South China University of Technology
2002
With the rapid development of online services and web applications, recommender systems (RS) have become increasingly indispensable for mitigating information overload matching users’ needs by providing personalized suggestions over items. Although RS research community has made remarkable progress past decades, conventional recommendation models (CRM) still some limitations, e.g. , lacking open-domain world knowledge, difficulties in comprehending underlying preferences motivations....
Traditional click-through rate (CTR) prediction models convert the tabular data into one-hot vectors and leverage collaborative relations among features for inferring user’s preference over items. This modeling paradigm discards essential semantic information. Though some works like P5 KAR have explored potential of using Pre-trained Language Models (PLMs) to extract signals CTR prediction, they are computationally expensive suffer from low efficiency. Besides, beneficial not considered,...
Multi-Domain Click-Through Rate (MDCTR) prediction is crucial for online recommendation platforms, which involves providing personalized services to users in different domains. However, current MDCTR models are confronted with the following limitations. Firstly, due varying data sparsity domains, can easily be dominated by some specific leads significant performance degradation other domains (i.e., “seesaw phenomenon”). Secondly, when new domain emerges, scalability of existing methods...
Click-through rate (CTR) prediction plays as a core function module in various personalized online services. The traditional ID-based models for CTR take inputs the one-hot encoded ID features of tabular modality, which capture collaborative signals via feature interaction modeling. But encoding discards semantic information included textual features. Recently, emergence Pretrained Language Models (PLMs) has given rise to another paradigm, takes sentences modality obtained by hard prompt...
Scoring a large number of candidates precisely in several milliseconds is vital for industrial pre-ranking systems. Existing systems primarily adopt the two-tower model since "user-item decoupling architecture" paradigm able to balance efficiency and effectiveness. However, cost high neglect potential information interaction between user item towers, hindering prediction accuracy critically. In this paper, we show it possible design that emphasizes both interactions inference efficiency. The...
Chaojun Xiao, Zhengyan Zhang, Xu Han, Chi-Min Chan, Yankai Lin, Zhiyuan Liu, Xiangyang Li, Zhonghua Zhao Cao, Maosong Sun. Proceedings of the 61st Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2023.
Considering the balance between revenue and resource consumption for industrial recommender systems, intelligent recommendation computing has been emerging recently. Existing solutions deploy same model to serve users indiscriminately, which is sub-optimal total maximization. We propose a multi-model service solution by deploying different-complexity models different-valued users. An automated dynamic generation framework AutoGen elaborated efficiently derive multiple parameter-sharing with...
As the demand for more personalized recommendation grows and a dramatic boom in commercial scenarios arises, study on multi-scenario (MSR) has attracted much attention, which uses data from all to simultaneously improve their performance. However, existing methods tend integrate insufficient scenario knowledge neglect learning cross-scenario preferences, thus leading sub-optimal Meanwhile, though large language model (LLM) shown great capability of reasoning capturing semantic information,...
Information sharing is a vital component of unified planning among multiple agencies performing varied tasks and activities toward effective emergency response, which promotes coordination. Insufficient information ineffective coordination organizations during disaster response creates bottleneck in need urgent resolution to effect preparedness. Building highly management system would help solve this problem. Modern technology processing techniques have produced tools approaches for across...
The large-scale disaster has characteristics of complexity, uncertainty and dynamics, leads to accident scene, so it brought difficulties challenges the emergency management process. In order response effectively, according business continuity management, this paper studies on fault-tolerant planning framework with power network infrastructure as an example. post-disaster restoration stage preparation stage, should be designed, which includes three types way, such task approach, resource...
Differentiated service (DiffServ) is a mechanism to provide the quality-of-service (QoS) with certain performance guarantee. In this paper, we study how design DiffServ multicast when every relay link an independent selfish agent. We assume that each e/sub i/ associated (privately known) cost coefficient c/sub such of transmission bandwidth demand x i//spl middot/x. Further, there fixed source node s and set R receivers, which requires from data minimum demand. The problem compute...
This research investigates how reviews and attendance of predecessor movie influence sequels' performance. The empirical analysis reveals the influencing mechanism movie' characteristics following sequels. In particular, using 32-year-long data authors find that commercial success parent enhances first sequel attendance, but have little impact on high number However, positive ratings from both critics amateurs do not affect performance sequels (neither nor sequel). They also induces more...
Text style transfer (TST) without parallel data has achieved some practical success. However, most of the existing unsupervised text methods suffer from (i) requiring massive amounts non-parallel to guide transferring different styles. (ii) colossal performance degradation when fine-tuning model in new domains. In this work, we propose DAML-ATM (Domain Adaptive Meta-Learning with Adversarial Transfer Model), which consists two parts: DAML and ATM. is a domain adaptive meta-learning approach...
Text style transfer is a hot issue in recent natural language processing,which mainly studies the text to adapt different specific situations, audiences and purposes by making some changes. The of usually includes many aspects such as morphology, grammar, emotion, complexity, fluency, tense, tone so on. In traditional model, generally relied on experts knowledge hand-designed rules, but with application deep learning field processing, method based Started be heavily researched. years,...
Exploration of mechanisms underlying the emergence collective cooperation remains a focal point in field evolution cooperation. Prevailing studies often neglect historical information, relying on latest rewards as primary criterion for individual decision-making-a method incongruent with human cognition and decision-making modes. This limitation impedes comprehensive understanding spontaneous Integrating memory factors into evolutionary game models to formulate decision criteria delayed...
Recommender systems aim to predict user interest based on historical behavioral data. They are mainly designed in sequential pipelines, requiring lots of data train different sub-systems, and hard scale new domains. Recently, Large Language Models (LLMs) have demonstrated remarkable generalized capabilities, enabling a singular model tackle diverse recommendation tasks across various scenarios. Nonetheless, existing LLM-based utilize LLM purely for single task the pipeline. Besides, these...
Click-Through Rate (CTR) prediction holds a paramount position in recommender systems. The prevailing ID-based paradigm underperforms cold-start scenarios due to the skewed distribution of feature frequency. Additionally, utilization single modality fails exploit knowledge contained within textual features. Recent efforts have sought mitigate these challenges by integrating Pre-trained Language Models (PLMs). They design hard prompts structure raw features into text for each interaction and...
CTR prediction plays a vital role in recommender systems. Recently, large language models (LLMs) have been applied systems due to their emergence abilities. While leveraging semantic information from LLMs has shown some improvements the performance of systems, two notable limitations persist these studies. First, LLM-enhanced encounter challenges extracting valuable lifelong user behavior sequences within textual contexts for recommendation tasks. Second, inherent variability human behaviors...
Based on scenario analysis, this paper makes a definition of natural disaster type and concept from different aspects, describes the mechanization development chain applies Dynamic Bayesian Network. It uses probability assessment to analyze transformation rules in scenario, make up shortage traditional method capture dynamic changes chain, provide support for prediction, decision response disasters.