Biblio

Social networks are considered to be heterogeneous graph neural networks (HGNNs) with deep learning technological advances. HGNNs, compared to homogeneous data, absorb various aspects of information about individuals in the training stage. That means more information has been covered in the learning result, especially sensitive information. However, the privacy-preserving methods on homogeneous graphs only preserve the same type of node attributes or relationships, which cannot effectively work on heterogeneous graphs due to the complexity. To address this issue, we propose a novel heterogeneous graph neural network privacy-preserving method based on a differential privacy mechanism named HeteDP, which provides a double guarantee on graph features and topology. In particular, we first define a new attack scheme to reveal privacy leakage in the heterogeneous graphs. Specifically, we design a two-stage pipeline framework, which includes the privacy-preserving feature encoder and the heterogeneous link reconstructor with gradients perturbation based on differential privacy to tolerate data diversity and against the attack. To better control the noise and promote model performance, we utilize a bi-level optimization pattern to allocate a suitable privacy budget for the above two modules. Our experiments on four public benchmarks show that the HeteDP method is equipped to resist heterogeneous graph privacy leakage with admirable model generalization.

Recommender system helps people in decision making by asking their preferences about various items and recommends other items that have not been rated yet and are similar to their taste. A traditional recommendation system aims at generating a set of recommendations based on inter-user similarity that will satisfy the target user. Positive preferences as well as negative preferences of the users are taken into account so as to find strongly related users. Weighted entropy is usedz as a similarity measure to determine the similar taste users. The target user is asked to fill in the ratings so as to identify the closely related users from the knowledge base and top N recommendations are produced accordingly. Results show a considerable amount of improvement in accuracy after using weighted entropy and opposite preferences as a similarity measure.

Research in combating misinformation reports many negative results: facts may not change minds, especially if they come from sources that are not trusted. Individuals can disregard and justify lies told by trusted sources. This problem is made even worse by social recommendation algorithms which help amplify conspiracy theories and information confirming one's own biases due to companies' efforts to optimize for clicks and watch time over individuals' own values and public good. As a result, more nuanced voices and facts are drowned out by a continuous erosion of trust in better information sources. Most misinformation mitigation techniques assume that discrediting, filtering, or demoting low veracity information will help news consumers make better information decisions. However, these negative results indicate that some news consumers, particularly extreme or conspiracy news consumers will not be helped. We argue that, given this background, technology solutions to combating misinformation should not simply seek facts or discredit bad news sources, but instead use more subtle nudges towards better information consumption. Repeated exposure to such nudges can help promote trust in better information sources and also improve societal outcomes in the long run. In this article, we will talk about technological solutions that can help us in developing such an approach, and introduce one such model called Trust Nudging.

Mobile and web application security, particularly the areas of data privacy, has raised much concerns from the public in recent years. Most applications, or apps for short, are installed without disclosing full information to users and clearly stating what the application has access to, which often raises concern when users become aware of unnecessary information being collected. Unfortunately, most users have little to no technical expertise in regards to what permissions should be turned on and can only rely on their intuition and past experiences to make relatively uninformed decisions. To solve this problem, we developed DroidNet, which is a crowd-sourced Android recommendation tool and framework. DroidNet alleviates privacy concerns and presents users with high confidence permission control recommendations based on the decision from expert users who are using the same apps. This paper explains the general framework, principles, and model behind DroidNet while also providing an experimental setup design which shows the effectiveness and necessity for such a tool.

Ubiquitous deployment of low-cost mobile positioning devices and the widespread use of high-speed wireless networks enable massive collection of large-scale trajectory data of individuals moving on road networks. Trajectory data mining finds numerous applications including understanding users' historical travel preferences and recommending places of interest to new visitors. Privacy-preserving trajectory mining is an important and challenging problem as exposure of sensitive location information in the trajectories can directly invade the location privacy of the users associated with the trajectories. In this paper, we propose a differentially private trajectory analysis algorithm for points-of-interest recommendation to users that aims at maximizing the accuracy of the recommendation results while protecting the privacy of the exposed trajectories with differential privacy guarantees. Our algorithm first transforms the raw trajectory dataset into a bipartite graph with nodes representing the users and the points-of-interest and the edges representing the visits made by the users to the locations, and then extracts the association matrix representing the bipartite graph to inject carefully calibrated noise to meet έ-differential privacy guarantees. A post-processing of the perturbed association matrix is performed to suppress noise prior to performing a Hyperlink-Induced Topic Search (HITS) on the transformed data that generates an ordered list of recommended points-of-interest. Extensive experiments on a real trajectory dataset show that our algorithm is efficient, scalable and demonstrates high recommendation accuracy while meeting the required differential privacy guarantees.

There are vast amounts of information in our world. Accessing the most accurate information in a speedy way is becoming more difficult and complicated. A lot of relevant information gets ignored which leads to much duplication of work and effort. The focuses tend to provide rapid and intelligent retrieval systems. Information retrieval (IR) is the process of searching for information that is related to some topics of interest. Due to the massive search results, the user will normally have difficulty in identifying the relevant ones. To alleviate this problem, a recommendation system is used. A recommendation system is a sort of filtering information system, which predicts the relevance of retrieved information to the user's needs according to some criteria. Hence, it can provide the user with the results that best fit their needs. The services provided through the web normally provide massive information about any requested item or service. An efficient recommendation system is required to classify this information result. A recommendation system can be further improved if augmented with a level of trust information. That is, recommendations are ranked according to their level of trust. In our research, we produced a recommendation system combined with an efficient level of trust system to guarantee that the posts, comments and feedbacks from users are trusted. We customized the concept of LoT (Level of Trust) [1] since it can cover medical, shopping and learning through social media. The proposed system TRS\_LoT provides trusted recommendations to the users with a high percentage of accuracy. Whereas a 300 post with more than 5000 comments from ``Amazon'' was selected to be used as a dataset, the experiment has been conducted by using same dataset based on ``post rating''.

Information shared on Twitter is ever increasing and users-recipients are overwhelmed by the number of tweets they receive, many of which of no interest. Filters that estimate the interest of each incoming post can alleviate this problem, for example by allowing users to sort incoming posts by predicted interest (e.g., "top stories" vs. "most recent" in Facebook). Global and personal filters have been used to detect interesting posts in social networks. Global filters are trained on large collections of posts and reactions to posts (e.g., retweets), aiming to predict how interesting a post is for a broad audience. In contrast, personal filters are trained on posts received by a particular user and the reactions of the particular user. Personal filters can provide recommendations tailored to a particular user's interests, which may not coincide with the interests of the majority of users that global filters are trained to predict. On the other hand, global filters are typically trained on much larger datasets compared to personal filters. Hence, global filters may work better in practice, especially with new users, for which personal filters may have very few training instances ("cold start" problem). Following Uysal and Croft, we devised a hybrid approach that combines the strengths of both global and personal filters. As in global filters, we train a single system on a large, multi-user collection of tweets. Each tweet, however, is represented as a feature vector with a number of user-specific features.

Collaborative filtering plays an essential role in a recommender system, which recommends a list of items to a user by learning behavior patterns from user rating matrix. However, if an attacker has some auxiliary knowledge about a user purchase history, he/she can infer more information about this user. This brings great threats to user privacy. Some methods adopt differential privacy algorithms in collaborative filtering by adding noises to a rating matrix. Although they provide theoretically private results, the influence on recommendation accuracy are not discussed. In this paper, we solve the privacy problem in recommender system in a different way by applying the differential privacy method into the procedure of recommendation. We design two differentially private recommender algorithms with sampling, named Differentially Private Item Based Recommendation with sampling (DP-IR for short) and Differentially Private User Based Recommendation with sampling(DP-UR for short). Both algorithms are based on the exponential mechanism with a carefully designed quality function. Theoretical analyses on privacy of these algorithms are presented. We also investigate the accuracy of the proposed method and give theoretical results. Experiments are performed on real datasets to verify our methods.