A5101577090- Recommender Systems and Techniques
- Topic Modeling
- Advanced Graph Neural Networks
- Natural Language Processing Techniques
- Multimodal Machine Learning Applications
- Advanced Bandit Algorithms Research
- Domain Adaptation and Few-Shot Learning
- Advanced Text Analysis Techniques
- Machine Learning in Healthcare
- Expert finding and Q&A systems
- Data Quality and Management
- Image Retrieval and Classification Techniques
- Speech Recognition and Synthesis
- Speech and dialogue systems
- Privacy-Preserving Technologies in Data
- Caching and Content Delivery
- Advanced Image and Video Retrieval Techniques
- Sentiment Analysis and Opinion Mining
- Mental Health via Writing
- Information Retrieval and Search Behavior
- Advanced Multi-Objective Optimization Algorithms
- Spam and Phishing Detection
- Neural Networks and Applications
- Hate Speech and Cyberbullying Detection
- Image and Video Quality Assessment
Tencent (China)
2017-2025
Institute of Computing Technology
2023-2024
Chinese Academy of Sciences
2023-2024
Beihang University
2023-2024
Zhejiang University
2024
Leiden University
2024
Shandong University
2024
Tsinghua University
2016-2023
Renmin University of China
2023
Group Image (Poland)
2023
Representation learning (RL) of knowledge graphs aims to project both entities and relations into a continuous low-dimensional space. Most methods concentrate on representations with triples indicating between entities. In fact, in most there are usually concise descriptions for entities, which cannot be well utilized by existing methods. this paper, we propose novel RL method taking advantages entity descriptions. More specifically, explore two encoders, including bag-of-words deep...
Entity alignment aims to link entities and their counterparts among multiple knowledge graphs (KGs). Most existing methods typically rely on external information of such as Wikipedia links require costly manual feature construction complete alignment. In this paper, we present a novel approach for entity via joint embeddings. Our method jointly encodes both relations various KGs into unified low-dimensional semantic space according small seed set aligned entities. During process, can align...
Entity images could provide significant visual information for knowledge representation learning. Most conventional methods learn representations merely from structured triples, ignoring rich extracted entity images. In this paper, we propose a novel Image-embodied Knowledge Representation Learning model (IKRL), where are learned with both triple facts and More specifically, first construct all of an neural image encoder. These then integrated into aggregated image-based via attention-based...
Cold-start problem is still a very challenging in recommender systems. Fortunately, the interactions of cold-start users auxiliary source domain can help recommendations target domain. How to transfer user's preferences from domain, key issue Cross-domain Recommendation (CDR) which promising solution deal with problem. Most existing methods model common preference bridge for all users. Intuitively, since vary user user, bridges different should be different. Along this line, we propose novel...
In the past decades, recommender systems have attracted much attention in both research and industry communities. Existing recommendation models mainly learn underlying user preference from historical behavior data (typically forms of item IDs), then estimate user-item matching relationships for recommendations. Inspired by recent progress on large language (LLMs), we develop a different paradigm, considering as instruction following LLMs. The key idea is that needs can be expressed natural...
Pre-training models have shown their power in sequential recommendation. Recently, prompt has been widely explored and verified for tuning after pre-training NLP, which helps to more effectively parameter-efficiently extract useful knowledge from downstream tasks, especially cold-start scenarios. However, it is challenging bring prompt-tuning NLP recommendation, since the tokens of recommendation (i.e., items) are million-level do not concrete explainable semantics, sequence modeling should...
Sememes are minimum semantic units of word meanings, and the meaning each sense is typically composed by several sememes. Since sememes not explicit for word, people manually annotate form linguistic common-sense knowledge bases. In this paper, we present that, sememe information can improve representation learning (WRL), which maps words into a low-dimensional space serves as fundamental step many NLP tasks. The key idea to utilize capture exact meanings within specific contexts accurately....
Recently, embedding techniques have achieved impressive success in recommender systems. However, the are data demanding and suffer from cold-start problem. Especially, for item which only has limited interactions, it is hard to train a reasonable ID embedding, called cold major challenge techniques. The two main problems: (1) A gap existing between deep model. (2) Cold would be seriously affected by noisy interaction. most methods do not consider both issues problem, simultaneously. To...
Both explicit and implicit feedbacks can reflect user opinions on items, which are essential for learning preferences in recommendation. However, most current recommendation algorithms merely focus positive (e.g., click), ignoring other informative behaviors. In this paper, we aim to jointly consider explicit/implicit positive/negative learn unbiased Specifically, propose a novel Deep feedback network (DFN) modeling click, unclick dislike DFN has an internal interaction component that...
Integrated recommendation aims to jointly recommend heterogeneous items in the main feed from different sources via multiple channels, which needs capture user preferences on both item and channel levels. It has been widely used practical systems by billions of users, while few works concentrate integrated systematically. In this work, we propose a novel Hierarchical reinforcement learning framework for (HRL-Rec), divides into two tasks channels sequentially. The low-level agent is selector,...
Cross-domain recommendation (CDR) aims to provide better results in the target domain with help of source domain, which is widely used and explored real-world systems. However, CDR matching (i.e., candidate generation) module struggles data sparsity popularity bias issues both representation learning knowledge transfer. In this work, we propose a novel Contrastive Cross-Domain Recommendation (CCDR) framework for matching. Specifically, build huge diversified preference network capture...
Despite the advancements of open-source large language models (LLMs), e.g., LLaMA, they remain significantly limited in tool-use capabilities, i.e., using external tools (APIs) to fulfill human instructions. The reason is that current instruction tuning largely focuses on basic tasks but ignores domain. This contrast excellent capabilities state-of-the-art (SOTA) closed-source LLMs, ChatGPT. To bridge this gap, we introduce ToolLLM, a general framework encompassing data construction, model...
Pioneering efforts have verified the effectiveness of diffusion models in exploring informative uncertainty for recommendation. Considering difference between recommendation and image synthesis tasks, existing methods undertaken tailored refinements to reverse process. However, these approaches typically use highest-score item corpus user interest prediction, leading ignorance user's generalized preference contained within other items, thereby remaining constrained by data sparsity issue. To...
Knowledge graphs (KGs), which could provide essential relational information between entities, have been widely utilized in various knowledge-driven applications. Since the overall human knowledge is innumerable that still grows explosively and changes frequently, construction update inevitably involve automatic mechanisms with less supervision, usually bring plenty of noises conflicts to KGs. However, most conventional representation learning methods assume all triple facts existing KGs...
Ruidong Wu, Yuan Yao, Xu Han, Ruobing Xie, Zhiyuan Liu, Fen Lin, Leyu Maosong Sun. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.
Knowledge graphs typically undergo open-ended growth of new relations. This cannot be well handled by relation extraction that focuses on pre-defined relations with sufficient training data. To address few-shot instances, we propose a novel bootstrapping approach, Neural Snowball, to learn transferring semantic knowledge about existing More specifically, use Relational Siamese Networks (RSN) the metric relational similarities between instances based and their labeled Afterwards, given its...
Knowledge graph (KG) entity typing aims at inferring possible missing type instances in KG, which is a very significant but still under-explored subtask of knowledge completion. In this paper, we propose novel approach for KG trained by jointly utilizing local from existing assertions and global triple KGs. Specifically, present two distinct knowledge-driven effective mechanisms inference. Accordingly, build embedding models to realize the mechanisms. Afterward, joint model via connecting...
Real-world integrated personalized recommendation systems usually deal with millions of heterogeneous items. It is extremely challenging to conduct full corpus retrieval complicated models due the tremendous computation costs. Hence, most large-scale consist two modules: a multi-channel matching module efficiently retrieve small subset candidates, and ranking for precise recommendation. However, suffers from cold-start problems when adding new channels or data sources. To solve this issue,...