Contrastive Collaborative Filtering for Cold-Start Item Recommendation Article Swipe
YOU?
·
· 2023
· Open Access
·
· DOI: https://doi.org/10.1145/3543507.3583286
· OA: W4319452363
The cold-start problem is a long-standing challenge in recommender systems.\nAs a promising solution, content-based generative models usually project a\ncold-start item's content onto a warm-start item embedding to capture\ncollaborative signals from item content so that collaborative filtering can be\napplied. However, since the training of the cold-start recommendation models is\nconducted on warm datasets, the existent methods face the issue that the\ncollaborative embeddings of items will be blurred, which significantly\ndegenerates the performance of cold-start item recommendation. To address this\nissue, we propose a novel model called Contrastive Collaborative Filtering for\nCold-start item Recommendation (CCFCRec), which capitalizes on the\nco-occurrence collaborative signals in warm training data to alleviate the\nissue of blurry collaborative embeddings for cold-start item recommendation. In\nparticular, we devise a contrastive collaborative filtering (CF) framework,\nconsisting of a content CF module and a co-occurrence CF module to generate the\ncontent-based collaborative embedding and the co-occurrence collaborative\nembedding for a training item, respectively. During the joint training of the\ntwo CF modules, we apply a contrastive learning between the two collaborative\nembeddings, by which the knowledge about the co-occurrence signals can be\nindirectly transferred to the content CF module, so that the blurry\ncollaborative embeddings can be rectified implicitly by the memorized\nco-occurrence collaborative signals during the applying phase. Together with\nthe sound theoretical analysis, the extensive experiments conducted on real\ndatasets demonstrate the superiority of the proposed model. The codes and\ndatasets are available on https://github.com/zzhin/CCFCRec.\n
