Xi Xiao

CenLocShare: A centralized privacy-preserving location-sharing system for mobile online social networks

Abstract Friend relationships and location information are the biggest privacy concerns for the users in mobile Online Social Networks (mOSNs). Existing solutions save these two kinds of privacy information in two different servers, i.e.,xa0Social Network Server (SNS) and Location Based Server (LBS), and employ some encryption protocols to map each other to provide the location-sharing service among friends and strangers. However, this distributed architecture incurs large communication overhead between SNS and LBS, and great storage burden as well. In order to overcome these shortages, we propose a centralized privacy-preserving location-sharing system, named CenLocShare. It integrates SNS and LBS into one server, i.e.,xa0Location-storing Social Network Server (LSSNS), and uses the dummy locations and the dedicated mapping protocols between LSSNS and Cellular Tower (CT) to share privacy-preserving locations. Its safety is validated by the security analysis. Furthermore, we implement the prototype of CenLocShare and do some comparisons with other systems. The thorough experiments indicate our system decreases the query time for friends’ locations and the storage space CenLocShare is more suitable and effective in the context of mOSNs

Weakly-supervised image captioning based on rich contextual information

Automatically generation of an image description is a challenging task which attracts broad attention in artificial intelligence. Inspired by methods of computer vision and natural language processing, different approaches have been proposed to solve the problem. However, captions generated by the existing approaches have been lack of enough contextual information to describe the corresponding images completely. The labeled captions in the training set only basically describe images and lack of enough contextual annotations. In this paper, we propose a Weakly-supervised Image Captioning Approach (WICA) to generate captions containing rich contextual information, without complete annotations for the contextual information in datasets. We utilize encoder-decoder neural networks to extract basic captioning features and leverage object detection networks to identify contextual features. Then, we encode the two levels of features by a phrase-based language model in order to generate captions with rich contextual information. The comprehensive experimental results reveal that proposed model outperforms the existing baselines in terms of on the richness and reasonability of contextual information for image captioning.

Android malware detection based on system call sequences and LSTM

As Android-based mobile devices become increasingly popular, malware detection on Android is very crucial nowadays. In this paper, a novel detection method based on deep learning is proposed to distinguish malware from trusted applications. Considering there is some semantic information in system call sequences as the natural language, we treat one system call sequence as a sentence in the language and construct a classifier based on the Long Short-Term Memory (LSTM) language model. In the classifier, at first two LSTM models are trained respectively by the system call sequences from malware and those from benign applications. Then according to these models, two similarity scores are computed. Finally, the classifier determines whether the application under analysis is malicious or trusted by the greater score. Thorough experiments show that our approach can achieve high efficiency and reach high recall of 96.6% with low false positive rate of 9.3%, which is better than the other methods.

A learnable search result diversification method

Abstract Search result diversification is to tackle the ambiguous queries and multi-faced information needs. The search result diversification problem can be formalized as a balance between the relevance score and the diversity score. Most previous diversification models utilize a predefined function to calculate the diversity score. The values of parameters need to be tuned by manual experiments. It is time-consuming and hard to reach optimal result in diversity evaluation. Proposing a learnable approach to solve the above problems is a pressing task. Therefore we introduce a Learnable Search Result Diversification model called L-SRD. On this basis, we redefine the diversity function and derive our loss function as the likelihood loss of ground truth generation. Stochastic gradient descent algorithm is employed to optimize the values of parameters. Finally we derive our ranking function to generate the diverse list sequentially. Due to the learning model, the values of parameters are determined automatically and get optimally. The experiments on TREC web tracks show that our approach outperforms several existing diversification models significantly.

Learning-based topic detection using multiple features: Topic Detection with Multiple Features

Recently, microblog sites such as Twitter attract a great deal of attention as an information resource for topic detection task. Most of existing feature‐pivot topic detection algorithms in Twitter just take a single feature into account rather than multiple features. Thus, these methods always only detect the topics related to the single feature and miss some important topics, which causes a relatively low performance. In this paper, we build a flexible term representation framework for feature‐pivot topic detection based on four features. A Learning‐based Topic Detection using Multiple Features (LTDMF) method is proposed to improve the performance of topic detection. We define a correlation function based on a specific neural network to integrate various features. A Hierarchical Agglomerative Clustering (HAC) algorithm is applied to cluster terms as topics. Based on multiple features, LTDMF detects all types of topics and improves the accuracy of topic detection to solve the problem of missing topics. Experiments show that LTDMF gets a better performance compared with several baseline methods in terms of precision and recall.

SAIDR: A New Dynamic Model for SMS-Based Worm Propagation in Mobile Networks

Recently, short message service (SMS) has become one of the most popular applications for mobile users. However, it provides convenience for worms to spread in mobile networks. Due to the differences between computers and smartphones, the current propagation models of computer worms cannot be employed in the mobile network directly, especially in the SMS scenario. In this paper, we propose a worm propagation model based on SMS, named susceptible-affected-infectious-suspended-recovered. To accurately predict the worm propagation via SMS, first, we add the affected state to represent the state of users who have received the messages but have not clicked the malicious links. Second, since an infected node does not always send malicious messages to others, a novel state, the suspended state, is introduced to describe this situation. Furthermore, related stabilities of the worm-free equilibrium and the endemic equilibrium are studied. The worm-free equilibrium is locally and globally asymptotically stable if the basic reproduction number <inline-formula> <tex-math notation=LaTeX>

Exploiting User Mobility for Time-aware POI Recommendation in Social Networks

{R}_{0}<1

Tianji: Implementation of an Efficient Tracking Engine in the Mobile Internet Era

</tex-math></inline-formula>, whereas the endemic equilibrium is locally asymptotically stable if <inline-formula> <tex-math notation=LaTeX>

Classification of ransomware families with machine learning based on N-gram of opcodes

R_{0}>1

A Learning Approach to Hierarchical Search Result Diversification.

</tex-math></inline-formula>. Finally, comprehensive experiments have been done to support our conclusions and confirm the rationality.