基于趋势一致性学习的对比聚类算法OA

In recent years,contrastive clustering has become a research hotspot in the fields of data mining and machine learning,aiming to enhance clustering performance by leveraging the powerful feature representation capabilities of contrastive learning.However,the use of contrastive learning often introduces the problem of false negative examples due to category conflicts,thereby reducing the performance of contrastive clustering.To address this issue,this paper proposes a contrastive clustering algorithm based on a trend consistency constraint strategy(CCTC).By marking high-confidence sample pairs with consistent category information in the trend consistency array and using this semantic in-formation to calculate the trend constraint matrix to assist in selecting positive samples,the algorithm achieves dynamic interaction between cluster-level and instance-level sample information through the combination of instance-level and cluster-level consistency loss functions,thereby enhancing sample consistency and inter-class distinguishability.Com-pared with other contrastive clustering algorithms,this method can utilize the pseudo-label change trends in the multi-round training process to obtain sample pairs with high-confidence category trend consistency,thus improving the clus-tering performance of the model.Experiments have demonstrated the effectiveness of the algorithm.