Zhuangzhuang He

Zhuangzhuang He

Auto-generated profile, based on publicly available information

Zhuangzhuang He

Hefei University of Technology, Hefei, China

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
24 August 2024
Verified
Conference Paper
Open access
YYYonghui Yang
Yonghui Yang
LWLe Wu
Le Wu
ZWZihan Wang
Zihan Wang
ZHZhuangzhuang He
Zhuangzhuang He
RHRichang Hong
Richang Hong
MWMeng Wang
Meng Wang
With the emergence of social networks, social recommendation has become an essential technique for personalized services. Recently, graph-based social recommendations have shown promising results by capturing the high-order social influence. Most empirical studies of graph-based social recommendations directly take the observed social networks into formulation, and produce user preferences based on social homogeneity. Despite the effectiveness, we argue that social networks in the real-world are inevitably noisy~(existing redundant social relations), which may obstruct precise user preference characterization. Nevertheless, identifying and removing redundant social relations is challenging due to a lack of labels. In this paper, we focus on learning the denoised social structure to facilitate recommendation tasks from an information bottleneck perspective. Specifically, we propose a novel Graph Bottlenecked Social Recommendation (GBSR) framework to tackle the social noise issue. GBSR is a model-agnostic social denoising framework, that aims to maximize the mutual information between the denoised social graph and recommendation labels, meanwhile minimizing it between the denoised social graph and the original one. This enables GBSR to learn the minimal yet sufficient social structure, effectively reducing redundant social relations and enhancing social recommendations. Technically, GBSR consists of two elaborate components, preference-guided social graph refinement, and HSIC-based bottleneck learning. Extensive experimental results demonstrate the superiority of the proposed GBSR, including high performances and good generality combined with various backbones. Our code is available at: https://github.com/yimutianyang/KDD24-GBSR.

10 July 2024
Verified
Conference Paper
Open access
HBHaoyue Bai
Haoyue Bai
LWLe Wu
Le Wu
MHMin Hou
Min Hou
MCMiaomiao Cai
Miaomiao Cai
ZHZhuangzhuang He
Zhuangzhuang He
MWMeng Wang
Meng Wang
Multimedia-based recommendation provides personalized item suggestions by learning the content preferences of users. With the proliferation of digital devices and APPs, a huge number of new items are created rapidly over time. How to quickly provide recommendations for new items at the inference time is challenging. What's worse, real-world items exhibit varying degrees of modality missing(e.g., many short videos are uploaded without text descriptions). Though many efforts have been devoted to multimedia-based recommendations, they either could not deal with new multimedia items or assumed the modality completeness in the modeling process. In this paper, we highlight the necessity of tackling the modality missing issue for new item recommendation. We argue that users' inherent content preference is stable and better kept invariant to arbitrary modality missing environments. Therefore, we approach this problem from a novel perspective of invariant learning. However, how to construct environments from finite user behavior training data to generalize any modality missing is challenging. To tackle this issue, we propose a novel Multimodality Invariant Learning reCommendation (a.k.a. MILK) framework. Specifically, MILK first designs a cross-modality alignment module to keep semantic consistency from pretrained multimedia item features. After that, MILK designs multi-modal heterogeneous environments with cyclic mixup to augment training data, in order to mimic any modality missing for invariant user preference learning.Extensive experiments on three real datasets verify the superiority of our proposed framework.The code is available at https://github.com/HaoyueBai98/MILK.

24 August 2024
Verified
Conference Paper
Open access
ZHZhuangzhuang He
Zhuangzhuang He
YWYifan Wang
Yifan Wang
YYYonghui Yang
Yonghui Yang
PSPeijie Sun
Peijie Sun
LWLe Wu
Le Wu
MZMin Zhang
Min Zhang
As its availability and generality in online services, implicit feedback is more commonly used in recommender systems. However, implicit feedback usually presents noisy samples in real-world recommendation scenarios (such as misclicks or non-preferential behaviors), which will affect precise user preference learning. To overcome the noisy samples problem, a popular solution is based on dropping noisy samples in the model training phase, which follows the observation that noisy samples have higher training losses than clean samples. Despite the effectiveness, we argue that this solution still has limits. (1) High training losses can result from model optimization instability or hard samples, not just noisy samples. (2) Completely dropping of noisy samples will aggravate the data sparsity, which lacks full data exploitation. To tackle the above limitations, we propose a Double Correction Framework for Denoising Recommendation (DCF), which contains two correction components from views of more precise sample dropping and avoiding more sparse data. In the sample dropping correction component, we use the loss value of the samples over time to determine whether it is noise or not, increasing dropping stability. Instead of averaging directly, we use the damping function to reduce the bias effect of outliers. Furthermore, due to the higher variance exhibited by hard samples, we derive a lower bound for the loss through concentration inequality to identify and reuse hard samples. In progressive label correction, we iteratively re-label highly deterministic noisy samples and retrain them to further improve performance. Finally, extensive experimental results on three datasets and four backbones demonstrate the effectiveness and generalization of our proposed framework.

13 July 2025
Verified
Conference Paper
Open access
YYYonghui Yang
Yonghui Yang
LWLe Wu
Le Wu
YLYuxin Liao
Yuxin Liao
ZHZhuangzhuang He
Zhuangzhuang He
PSPengyang Shao
Pengyang Shao
MWMeng Wang
Meng Wang

25 May 2021 in IEEE Access
Verified
Journal Article
Open access
HWHao Wu
Hao Wu
CSChunshan Shen
Chunshan Shen
ZHZhuangzhuang He
Zhuangzhuang He
YWYongmei Wang
Yongmei Wang
XXXinyuan Xu
Xinyuan Xu
Human-Computer natural language interaction is helpful to reduce the operation and maintenance cost of the SCADA system, and it is necessary to solve the complex natural language interface problem that supports data query and real-time control. According to the complexity of natural language instructions, a hierarchical classification of semantic parsing algorithm is adopted. Firstly, the KWECS method is used to classify the intent of natural language instruction, then the TF-IDF keyword extraction algorithm combined with the cosine similarity is used to structure the key-value of the classified natural language instructions which was used into SCADA control intermediate language and then formally converted into actual control or query instruction. If the analysis fails, the complex control and query instruction analysis are carried out according to the classification results, structuring instruction parsing based on dependency parsing and SQL natural language parsing based on deep learning are adopted respectively to implement real-time control interface and database query interface. Our experimental results show that the proposed hierarchical classification of natural language comprehensive query and control interface can better solve the problem of human-computer natural language interaction in the SCADA system, and the accuracy of intent recognition reaches 96.5%. In more detail, the accuracy, precision, recall, and F-score of instruction parsing reach 88.47%, 90.21%, 89.48%, and 89.72% respectively. Especially, it provides more convenient interactive means for industrial and agricultural information management and control.

26 August 2021 in IEEE Access
Verified
Journal Article
Open access
ZHZhuangzhuang He
Zhuangzhuang He
HWHao Wu
Hao Wu
GWGuodong Wu
Guodong Wu
Hyperspectral images are rich in both spectral information and spatial dependence information between pixels; however, hyperspectral images are characterized by the high dimensionality of small data sets and the spectral variance. Facing these problems, spatial dependence information as supplementary information is a relatively effective means to solve them. And the label dependence characteristic of hyperspectral images is excellent spatial dependence information. Therefore, to address the above issues, based on residual network and spatial information extractor(RAS), which is based on a residual network, pixel embedding(PE), and a spatial information extractor(SIE). At the stage of mining spectral information, we use the residual network to mine spectral features; At the stage of mining spatial information, we utilize the label dependency characteristic to feed the set of pixels containing the target pixels into PE. Then, a pixel vector with location information and self-defined dimensionality is obtained. Next, this vector is fed into our proposed SIE to mine the spatial dependency information. In multi-group ablation experiments, our proposed model achieves overall accuracy (OA) scores of 79.16% on the 5% Indian Pines test set, 90.82% on the 1% Pavia University test set, and 92.17% on the 1% Salinas test set. Especially, the experimental results demonstrate that the joint spectral-spatial approach is effective in improving the accuracy of hyperspectral image classification.