Machine learning (ML) is increasingly used behind closed systems and APIs to make important decisions. For example, social media uses ML-based recommendation algorithms to decide what to show users, and millions of people pay to use ChatGPT for information every day. Because ML is deployed behind these closed systems, there are increasing calls for transparency, such as releasing model weights. However, these service providers have legitimate reasons not to release this information, including for privacy and trade secrets. To bridge this gap, recent work has proposed using zero-knowledge proofs (specifically a form called ZK-SNARKs) for certifying computation with private models but has only been applied to unrealistically small models.

Recent advancements in the field of extended reality (XR) have garnered significant interest in XR socialization. However, traditional social XR experiences often fall short of satisfying users' social expectations due to the negligence of the emerging opportunities in XR. In this paper, we propose a novel scenario of socializing in social XR, which has the potential to substantially enhance traditional social media through i) the recommendation of appropriate surrounding users that cater to users' individual preferences, ii) the adaptive avoidance of view occlusions to facilitate users in locating their friends, iii) the consideration of users' social presence, and iv) the development of cross-platform solutions to provide hybrid participation. To this end, we formulate Adaptive Friend Discovery for Temporal-spatial and Social-aware XR, a new NP-hard social recommendation problem aiming at satisfying social XR users. The proposed model, POSHGNN, is a deep temporal graph learning framework designed to provide efficient social recommendations for target users. Experimental results obtained from real-world social XR datasets and a user study that supports multiple XR interfaces demonstrate that the proposed method outperforms baseline approaches with an improvement of 18.5 % in solution quality.

Social metaverse with VR has been viewed as a paradigm shift for social media. However, most traditional VR social platforms ignore emerging characteristics in a metaverse, thereby failing to boost user satisfaction. In this paper, we explore a scenario of socializing in metaverse with VR, which brings major advantages over conventional social media: 1) leverage flexible display of users' 360-degree viewports to satisfy individual user interests, 2) ensure the user feelings of co-existence, 3) prevent view obstruction to help users find friends in crowds, and 4) support socializing with digital twins. Therefore, we formulate the Co-presence, and Occlusion-aware Metaverse User Recommendation (COMUR) problem to recommend a set of rendered players for users in social metaverse with VR. We prove COMUR is an NP-hard optimization problem and design a dual-module deep graph learning framework (COMURNet) to recommend appropriate users for viewport display. Experimental results on real social metaverse datasets and a user study with Occulus Quest 2 manifest that the proposed model outperforms baseline approaches by at least 36.7% of solution quality.

When spreading information over social networks, seeding algorithms selecting users to start the dissemination play a crucial role. The majority of existing seeding algorithms focus solely on maximizing the total number of reached nodes, overlooking the issue of group fairness, in particular, gender imbalance. To tackle the challenge of maximizing information spread on certain target groups, e.g., females, we introduce the concept of the community and gender-aware potential of users. We first show that the network's community structure is closely related to the gender distribution. Then, we propose an algorithm that leverages the information about community structure and its gender potential to iteratively modify a seed set such that the information spread on the target group meets the target ratio. Finally, we validate the algorithm by performing experiments on synthetic and real-world datasets. Our results show that the proposed seeding algorithm achieves not only the target ratio but also the highest information spread, compared to the state-of-the-art gender-aware seeding algorithm.

With the rising influence of current online services, it is important for service providers to discover related accounts because it helps detect suspicious account groups. Existing research on this topic mostly focuses on a variety of account behaviors. Little attention has been paid to relations among account identity, which unveils the relationship between accounts and real-world people. In this paper, we propose DRAGON for modeling an account identity network and detecting suspicious account groups among this network. To this end, identifiers for tracking physical devices are collected and uncertain graph is used for modeling uncertainty in the network. Within this network, a strategy for detecting suspicious account groups is also investigated in DRAGON. We evaluate DRAGON using a real-world dataset. The results indicate that DRAGON achieves a 280% improvement in precision and 150% improvement in recall compared to a binary classifier.