Contrastive Learning for Cold-Start Recommendation

Abstract

Recommending purely cold-start items is a long-standing and fundamental challenge in the recommender systems. Without any historical interaction on cold-start items, the collaborative filtering (CF) scheme fails to leverage collaborative signals to infer user preference on these items. To solve this problem, extensive studies have been conducted to incorporate side information of items (e.g., content features) into the CF scheme. Specifically, they employ modern neural network techniques (e.g., dropout, consistency constraint) to discover and exploit the coalition effect of content features and collaborative representations. However, we argue that these works less explore the mutual dependencies between content features and collaborative representations and lack sufficient theoretical supports, thus resulting in unsatisfactory performance on cold-start recommendation.In this work, we reformulate the cold-start item representation learning from an information-theoretic standpoint. It aims to maximize the mutual dependencies between item content and collaborative signals. Specifically, the representation learning is theoretically lower-bounded by the integration of two terms: mutual information between collaborative embeddings of users and items, and mutual information between collaborative embeddings and feature representations of items. To model such a learning process, we devise a new objective function founded upon contrastive learning and develop a simple yet efficient Contrastive Learning-based Cold-start Recommendation framework (CLCRec). In particular, CLCRec consists of three components: contrastive pair organization, contrastive embedding, and contrastive optimization modules.

Implicit semantic-based personalized micro-videos recommendation

“…), and corrects and optimizes its recommendation results based on the feedback results of interaction with users (clicks, ratings, etc.) [17,18].…”

A multimedia recommendation model based on collaborative graph

“…In the multimodal recommendation , the main currently used recommendation models are divided into collaborative filtering-based video recommendation [24], contentbased video recommendation [25] and hybrid video recommendation [26]. YouTube [27] in 2008 used User-Video-based graph tour algorithm is a kind of propagation diffusion of video labels on the graph of collaborative filtering algorithm, but only the video label information is considered, and there is a cold start problem [28]. Mei et al [29] designed a contextual video recommendation system based on multimodal content relevance and user feedback, considering the different composition of the video and the different levels of user interest in different parts of the video, and seamlessly integrating multimodal relevance and user feedback through relevance feedback and attention fusion, but ignoring the role of user.…”

Improving Micro-video Recommendation via Contrastive Multiple Interests

“…We set the number of interests to [1,2,4,8,16] successively and conduct experiments. The experimental results are shown in Table 4.…”

“…It has obtained the success in computer vision [2], natural language processing [4,5,19] and other domains [13]. More recently, contrastive learning also has been introduced to the recommendation, such as sequential recommendation, recommendation based on graph neural network, and etc., which realizes the debiasing [22] and the denoising [14], and resolves the representation degeneration [15] and the cold start problem [16], improving the recommendation accuracy [10,18,20,21]. We note that there exists noise in the positive interactions in the micro-video scenario since micro-videos are automatically played and sometimes users cannot judge whether they like the micro-video or not until the micro-video finishes playing.…”