Junlin Zhou

Information filtering on coupled social networks.

In this paper, based on the coupled social networks (CSN), we propose a hybrid algorithm to nonlinearly integrate both social and behavior information of online users. Filtering algorithm, based on the coupled social networks, considers the effects of both social similarity and personalized preference. Experimental results based on two real datasets, Epinions and Friendfeed, show that the hybrid pattern can not only provide more accurate recommendations, but also enlarge the recommendation coverage while adopting global metric. Further empirical analyses demonstrate that the mutual reinforcement and rich-club phenomenon can also be found in coupled social networks where the identical individuals occupy the core position of the online system. This work may shed some light on the in-depth understanding of the structure and function of coupled social networks.

Embedding a mesh of trees in the crossed cube

Crossed cubes are an important class of variants of hypercubes as interconnection topologies in parallel computing. In this paper, we study the embedding of a mesh of trees in the crossed cube. Let n be a multiple of 4 and N=2^(^n^-^2^)^/^2. We prove that an NxN mesh of trees (containing 3N^2-2N nodes) can be embedded in an n-dimensional crossed cube (containing 4N^2 nodes) with dilation 1 and expansion about 4/3. This result shows that crossed cubes are promising interconnection networks since mesh of trees enables fast parallel computation.

A personalized recommendation algorithm via biased random walk

With the rapid development of Internet, Recommender Systems can help us efficiently find the useful objects in the information era. Generally, the traditional random walk algorithm has high accuracy but low personality and diversity. In this paper, we propose an improved random walk algorithm by depressing the influence of large-degree objects. Experimental results on MovieLens and Netflix data sets show that this algorithm can effectively improve not only the accuracy (improved by 5.5% and 5.9%, respectively) but also the diversity.

Personalized Privacy-Preserving Frequent Itemset Mining Using Randomized Response

Frequent itemset mining is the important first step of association rule mining, which discovers interesting patterns from the massive data. There are increasing concerns about the privacy problem in the frequent itemset mining. Some works have been proposed to handle this kind of problem. In this paper, we introduce a personalized privacy problem, in which different attributes may need different privacy levels protection. To solve this problem, we give a personalized privacy-preserving method by using the randomized response technique. By providing different privacy levels for different attributes, this method can get a higher accuracy on frequent itemset mining than the traditional method providing the same privacy level. Finally, our experimental results show that our method can have better results on the frequent itemset mining while preserving personalized privacy.

Intelligent spam filtering for massive short message stream

Purpose – The purpose of this paper is to provide an intelligent spam filtering method to meet the real‐time processing requirement of the massive short message stream and reduce manual operation of the system.Design/methodology/approach – An integrated framework based on a series of algorithms is proposed. The framework consists of message filtering module, log analysis module and rules handling module, and dynamically filters the short message spam, while generating the filtering rules. Experiments using Java are used to execute the proposed work.Findings – The experiments are carried out both on the simulation model (off‐line) and on the actual plant (on‐line). All experiment data are considered in both normal and spam real short messages. The results show that use of the integrated framework leads to a comparable accuracy and meet the real‐time filtration requirement.Originality/value – The approach in the design of the filtering system is novel. In addition, implementation of the proposed integrated ...

UNSUPERVISED LEARNING BASED DISTRIBUTED DETECTION OF GLOBAL ANOMALIES

Anomaly detection has recently become an important problem in many industrial and financial applications. Very often, the databases from which anomalies have to be found are located at multiple local sites and cannot be merged due to privacy reasons or communication overhead. In this paper, a novel general framework for distributed anomaly detection is proposed. The proposed method consists of three steps: (i) building local models for distributed data sources with unsupervised anomaly detection methods and computing quality measure of local models; (ii) transforming local unsupervised local models into sharing models; and (iii) reusing sharing models for new data and combining their results by considering both quality and diversity of them to detect anomalies in a global view. In experiments performed on synthetic and real-life large data set, the proposed distributed anomaly detection method achieved prediction performance comparable or even slightly better than the global anomaly detection algorithm applied on the data set obtained when all distributed data set were merged.

Spectral clustering algorithm based on adaptive neighbor distance sort order

Spectral clustering has been widely used in data mining in the past years. The performance of spectral clustering is very sensitive to the selection of scale parameter. Especially, when data has multi-scale it is very difficult to find a proper value for the scale parameter. To solve the problem, an improved method based on adaptive neighbor distance sort order has been proposed in this paper. The method enlarges the affinity between two points in the same cluster and reduces that in different clusters. Our experiments on the synthetic and real life datasets have shown promising results comparing with tradition method and k-means.

TaDb: A time-aware diffusion-based recommender algorithm

Traditional recommender algorithms usually employ the early and recent records indiscriminately, which overlooks the change of user interests over time. In this paper, we show that the interests of a user remain stable in a short-term interval and drift during a long-term period. Based on this observation, we propose a time-aware diffusion-based (TaDb) recommender algorithm, which assigns different temporal weights to the leading links existing before the target users collection and the following links appearing after that in the diffusion process. Experiments on four real datasets, Netflix, MovieLens, FriendFeed and Delicious show that TaDb algorithm significantly improves the prediction accuracy compared with the algorithms not considering temporal effects.

Time series analysis based on enhanced NLCS

Similarity analysis plays a key role in clustering of time series. Normalized longest common subsequence (NLCS) is a similarity measurement widely used in comparing character sequences. In this paper, we developed the NLCS and present a novel algorithm to precisely calculate the similarity of time series. The algorithm used the sum of all common subsequence instead of longest common subsequence which can not represent the similarity of sequences accurately. The experiments based on synthetic and real-life datasets shown that the proposed algorithm performed better in comparing the similarity of time series. Comparing with Euclidean distance on four cluster validity indices, the results lead to a better performance by k-means or self-organize map.

Privacy-Preserving Distributed Collaborative Filtering Using Random Orthogonal Transformation

Collaborative Filtering is a powerful recommendation technique and has been widely used in e-commerce, search engine, and etc. Typically, the collaborative filtering model is built on a central storage of user preferences to generate the personalized recommendation. To produce a better recommendation, many data owners (small or medium company) collaborate with each other for building a shared collaborative filtering model. This leads to the privacy problem that the data owner is reluctant to reveal its data to others. To protect the user privacy, we design an privacy-preserving approach based on the random orthogonal transformation under the semi-honest model. We show that the distributed collaborative filtering based on our approach can provide zero loss of accuracy in the recommendation while preserving the privacy of different data owners.