Ask or search anything...

PapersBenchmarks
Hot
Intelligence Indeed
MSL: Not All Tokens Are What You Need for Tuning LLM as a Recommender
07 Jun 2025
Zhejiang University logoZhejiang UniversityOPPO
Large language models (LLMs), known for their comprehension capabilities and extensive knowledge, have been increasingly applied to recommendation systems (RS). Given the fundamental gap between the mechanism of LLMs and the requirement of RS, researchers have focused on fine-tuning LLMs with recommendation-specific data to enhance their performance. Language Modeling Loss (LML), originally designed for language generation tasks, is commonly adopted. However, we identify two critical limitations of LML: 1) it exhibits significant divergence from the recommendation objective; 2) it erroneously treats all fictitious item descriptions as negative samples, introducing misleading training signals. To address these limitations, we propose a novel Masked Softmax Loss (MSL) tailored for fine-tuning LLMs on recommendation. MSL improves LML by identifying and masking invalid tokens that could lead to fictitious item descriptions during loss computation. This strategy can effectively avoid the interference from erroneous negative signals and ensure well alignment with the recommendation objective supported by theoretical guarantees. During implementation, we identify a potential challenge related to gradient vanishing of MSL. To overcome this, we further introduce the temperature coefficient and propose an Adaptive Temperature Strategy (ATS) that adaptively adjusts the temperature without requiring extensive hyperparameter tuning. Extensive experiments conducted on four public datasets further validate the effectiveness of MSL, achieving an average improvement of 42.24% in NDCG@10. The code is available at this https URL
View blog
#computer-science#information-retrieval
Resources41
GUI-Robust: A Comprehensive Dataset for Testing GUI Agent Robustness in Real-World Anomalies
17 Jun 2025
Zhejiang University logoZhejiang UniversityIntelligence Indeed

Researchers from Zhejiang University and Intelligence Indeed present GUI-Robust, a dataset of 5,318 GUI tasks including 200 real-world anomalies, designed to evaluate the robustness of GUI agents. Experiments reveal that state-of-the-art GUI-specific agents and general-purpose MLLMs experience substantial performance degradation in abnormal scenarios, highlighting a critical gap in real-world deployability.

View blog
#agents#computer-science#artificial-intelligence
Resources7
Distributionally Robust Graph-based Recommendation System
21 Feb 2024
Zhejiang University logoZhejiang UniversityIntelligence Indeed

Researchers from Zhejiang University developed DR-GNN, a GNN-based recommendation system that enhances robustness against various distribution shifts by integrating Distributionally Robust Optimization. The method reinterprets GNN aggregation as a graph smoothing regularizer and introduces a Graph Edge-Addition strategy to mitigate data sparsity, demonstrating consistent performance improvements over state-of-the-art baselines across multiple datasets and shift types.

View blog
#computer-science#information-retrieval
Resources31
HatLLM: Hierarchical Attention Masking for Enhanced Collaborative Modeling in LLM-based Recommendation
13 Oct 2025
Zhejiang University logoZhejiang UniversityOPPO
Recent years have witnessed a surge of research on leveraging large language models (LLMs) for sequential recommendation. LLMs have demonstrated remarkable potential in inferring users' nuanced preferences through fine-grained semantic reasoning. However, they also exhibit a notable limitation in effectively modeling collaborative signals, i.e., behavioral correlations inherent in users' historical interactions. Our empirical analysis further reveals that the attention mechanisms in LLMs tend to disproportionately focus on tokens within the same item, thereby impeding the capture of cross-item correlations. To address this limitation, we propose a novel hierarchical attention masking strategy for LLM-based recommendation, termed HatLLM. Specifically, in shallow layers, HatLLM masks attention between tokens from different items, facilitating intra-item semantic understanding; in contrast, in deep layers, HatLLM masks attention within items, thereby compelling the model to capture cross-item correlations. This progressive, layer-wise approach enables LLMs to jointly model both token-level and item-level dependencies. Extensive experiments on three real-world datasets demonstrate that HatLLM achieves significant performance gains (9.13% on average) over existing LLM-based methods.
View blog
#computer-science#information-retrieval
Resources
Field Matters: A lightweight LLM-enhanced Method for CTR Prediction
20 May 2025
Zhejiang University logoZhejiang UniversityOPPO

Researchers from Zhejiang University and OPPO Research Institute introduce LLaCTR, a lightweight method for enhancing Click-Through Rate (CTR) prediction by leveraging Large Language Models at the field level rather than instance level, achieving 2.24% AUC improvement across four real-world datasets while reducing computational overhead by 10-100x compared to existing LLM-enhanced approaches.

View blog
#computer-science#artificial-intelligence#information-retrieval
Resources
Bridging the Gap: Self-Optimized Fine-Tuning for LLM-based Recommender Systems
27 May 2025
South China University of TechnologyZhejiang University logoZhejiang University

Zhejiang University and OPPO Research Institute researchers introduce SOFT (Self-Optimized Fine-Tuning), a training framework that improves LLM-based recommender systems by combining supervised fine-tuning with self-distilled auxiliary data generation and a self-adaptive curriculum scheduler that dynamically balances training on simplified self-generated examples versus real recommendation data based on semantic distance between model outputs and target items, achieving superior performance across multiple datasets compared to traditional fine-tuning approaches, LLM-enhanced recommenders, and classical recommendation models while demonstrating that curriculum learning principles can effectively bridge the knowledge gap between general LLM capabilities and domain-specific recommendation requirements.

View blog
#computer-science#artificial-intelligence#information-retrieval
Resources
There are no more papers matching your filters at the moment.