Rui Kong’s research while affiliated with Peking University and other places

Today's online social networks (OSNs) do little to protect users' private information especially relationships between them. This paper presents PrivOSN, an online social network system designed to be more private than existing systems while keeping neutral performance, preserving existing services, and defending against kinds of attacks already known. In PrivOSN, users' relationships are kept locally and can't be seen by OSN providers. By subscribing, users can follow and set up relationship with others, and these are transparent to OSN providers, for better privacy.

As the number of products being sold online increases, it is becoming increasingly difficult for customers to make purchasing decisions based on only pictures and short product descriptions. Thus, customer reviews, particularly the text describing the features, comparisons and experiences of using a particular product provide a rich source of information to compare products and make purchasing decisions. Especially, all kinds of reviews from various people have different degree of impact on a buyer, that is, we tend to believe our friends who always make right decisions than others. In this paper, we present an individual feature-based product ranking technique that mines thousands of customer reviews. By grouping users into unfamiliar users and familiar users according to the fact whether the client has almost always right ideas as far as one has concerned we attach different weights to them based on the friend ranking list. Friends on the top of the list are expected to be more reliable than the rest. After founding the client's friend set{F_j, S_k}, we extract crucial information from users' reviews. By realizing key words in a sentence, we classify comments into 4 representative sentences-Active Direct sentence(AD), Inactive Direct sentence(ID), Active Indirect sentence(AI), and Inactive Indirect sentence(II). Afterwards, we construct a weighted graph considering the product weight itself and the edge between every 2 relevant products, using ratios AD/ID and ID/II. The last step is that the client ranks search result with the average reliabilities of himself with respect to reviews of specific feature. Through calculation, we have a weighted score list, helping the client make purchase intentions.