Durvasula V. L. N. Somayajulu

A community driven social recommendation system

Recommendation systems play an important role in suggesting relevant information to users. In this paper, we introduce community-wise social interactions as a new dimension for recommendations and present a social recommendation system using collaborative filtering and community detection approaches. We use (i) community detection algorithm to extract friendship relations among users by analyzing user-user social graph and (ii) user-item based collaborative filtering for rating prediction. We developed our approach using map-reduce framework. Our approach improves scalability, coverage and cold start issue of collaborative filtering based recommendation system. We carried out experiments on MovieLens and Facebook datasets, to predict the rating of the movie and produce top-k recommendations for new (cold start) user. The results are compared with traditional collaborative filtering based recommendation system.

A Roadmap on Improved Performance-centric Cloud Storage Estimation Approach for Database System Deployment in Cloud Environment

Cloud computing has taken the limelight with respect to the present industry scenario due to its multi-tenant and pay-as-you-use models, where users need not bother about buying resources like hardware, software, infrastructure, etc. on an permanently basis. As much as the technological benefits, cloud computing also has its downside. By looking at its financial benefits, customers who cannot afford initial investments, choose cloud by compromising on its concerns, like security, performance, estimation, availability, etc. At the same time due to its risks, customers - relatively majority in number, avoid migration towards cloud. Considering this fact, performance and estimation are being the major critical factors for any application deployment in cloud environment; this paper brings the roadmap for an improved performance-centric cloud storage estimation approach, which is based on balanced PCTFree allocation technique for database systems deployment in cloud environment. Objective of this approach is to highlight the set of key activities that have to be jointly done by the database technical team and business users of the software system in order to perform an accurate analysis to arrive at estimation for sizing of the database. For the evaluation of this approach, an experiment has been performed through varied-size PCTFree allocations on an experimental setup with 100000 data records. The result of this experiment shows the impact of PCTFree configuration on database performance. Basis this fact, we propose an improved performance-centric cloud storage estimation approach in cloud. Further, this paper applies our improved performance-centric storage estimation approach on decision support system (DSS) as a case study.

Privacy Preserving Outlier Detection Using Hierarchical Clustering Methods

Data objects which do not comply with the general behavior or model of the data are called Outliers. Outlier Detection in databases has numerous applications such as fraud detection, customized marketing, and the search for terrorism. However, the use of Outlier Detection for various purposes has raised concerns about the violation of individual privacy. Therefore, Privacy Preserving Outlier Detection must ensure that privacy concerns are addressed and balanced, so that the data analyst can get the benefits of outlier detection without being thwarted by legal counter-measures by privacy advocates. In this paper, we propose a technique for detecting outliers while preserving privacy, using hierarchical clustering methods. We analyze our technique to quantify the privacy preserved by this method and also prove that reverse engineering the perturbed data is extremely difficult.

Privacy Preserving Clustering by Cluster Bulging for Information Sustenance

Cluster analysis is a data mining approach for unsupervised learning. However, the use of clustering as a data mining tool has been a cause of growing concern as the use of this technology is violating individual privacy. This paper presents a method for privacy preserving clustering through cluster bulging. In this method, the objects of the database are first aligned into clusters based on a similarity measure. The data in these clusters is perturbed in a controlled manner by modifying the values of various objects, so that, in the perturbed data set, the clusters are bulged in comparison to those in the original data set. In order to perform this perturbation, every cluster is displaced along the line joining its centroid to the centroid of the whole data set. And, then, every object in each cluster is shifted along the line joining that object to the centroid of the cluster. The word bulging used here refers to both positive and negative bulging. The method in essence manipulates the similarity measures and recomputes the new perturbed objects of the respective clusters. Thus, every object in the bulged cluster represents its corresponding object from the original cluster. After the application of this method, the objects get perturbed, while the number of member objects and shape of each cluster remain the same as those of the original clusters, thereby the information in the two instances of the data sets is sustained, while, the privacy of sensitive data is preserved.

Scalable classifiers with dynamic pruning

The paper presents an algorithm to solve the problem of classification for data mining applications. This is a decision tree classifier which uses modified gini index as the partitioning criteria. A pre-sorting technique is used to overcome the problem of sorting at each node of the tree. This technique is integrated with a breadth first tree growth strategy which enables us to calculate the best partition for each of the leaf nodes in a single scan of a database. We have implemented this algorithm using depth first tree growth strategy also. The algorithm uses a dynamic pruning approach which reduces the number of scans of the database and does away with a separate tree pruning phase. The proof of correctness, analysis and performance study are also presented.

Paired feature constraints for latent dirichlet topic models

Non Parametric Bayes models, so called family of Latent Dirichlet Allocation (LDA) Topic Models have found application in various aspects of pattern recognition like sentiment analysis, information retrieval, question answering etc. The topics induced by LDA are used for later tasks such as classification, regression(movie ratings), ranking and recommendation. Recently various approaches are suggested to improve the utility of topics induced by LDA using various side-information such as labeled examples and labeled features. Pair-Wise feature constraints such as cannot-link and must-link, represent weak-supervision and are prevalent in domains such as sentiment analysis. Though must-link constraints are relatively easier to incorporate by using dirichlet tree, the cannot-link constraints are harder to incorporate using the dirichlet forest. In this paper we proposed an approach to address this problem using posterior constraints. We introduced additional latent variables for capturing the constraints, and modified the gibbs sampling algorithm to incorporate these constraints. Our method of Posterior Regularization has enabled us to deal with both types of constraints seamlessly in the same optimization framework. We have demonstrated our approach on a product sentiment review data set which is typically used in text analysis.

Learning automata-based trust model for user recommendations in online social networks

Abstract Nowadays, most of the online social media websites provide recommendations as service for selective decision making. Determining a recommended trust path based on the consumer’s non-functional requirements, such as availability of the products, delay for computing recommendations and response time for a good recommendation is one of the challenging issues in online social networks. In this paper, we first design a recommendation-based online social network architecture by incorporating trust information (namely, direct trust and indirect trust), relevance degree and recommended influence value. We propose a high quality of social trust associated model for evaluating a recommended trust path. The proposed model estimates utility values with associated weights based on Shannon entropy information gain. Further, for best recommended trust path selection, we propose a Learning Automata based Recommended Trust Path Selection (LA-RTPS) algorithm to identify multiple recommended trust paths and to determine an aggregate path. The experimentation using real time datasets illustrates the efficacy of the proposed algorithm.

Dynamic Replication Algorithm for Data Replication to Improve System Availability: A Performance Engineering Approach

ABSTRACT Information technology systems deployed by enterprises should not only fulfil their business functionalities but also cater to Quality of Service concerns such as Availability, Scalability, and Performance. To enhance the system performance, the system availability is an important factor and to improve the system availability, one of the strategies is replicating the frequently accessed data to multiple suitable locations which is a practical choice as the users can access the data from a nearby site. This is, however, not the case for replicas which must have a preset number of copies on several locations. How to decide a sensible number and right location for replicas have become an important issue in cloud computing. In this paper, we show a dynamic data replication strategy to enhance the performance of software system. To identify the suitable file to replicate and to decide respective number of replicas, we calculate popularity degree and replica factor. We use the fuzzy logic system to identify the system to place the replicas and we use the round robin method to place the replicas in the identified systems. We compare the performance of our technique with the existing technique.

A Scalable Algorithm for Discovering Topologies in Social Networks

Discovering topologies in a social network targets various business applications such as finding key influencers in a network, recommending music movies in virtual communities, finding active groups in network and promoting a new product. Since social networks are large in size, discovering topologies from such networks is challenging. In this paper, we present a scalable topology discovery approach using Giraph platform and perform (i) graph structural analysis and (ii) graph mining. For graph structural analysis, we consider various centrality measures. First, we find top-K centrality vertices for a specific topology (e.g. Star, ring and mesh). Next, we find other vertices which are in the neighborhood of top centrality vertices and then create the cluster based on structural density. We compare our clustering approach with DBSCAN algorithm on the basis of modularity parameter. The results show that clusters generated through structural density parameter are better in quality than generated through neighborhood density parameter.

Modified correlation based technique in micro array data analysis for searching differentially expressed genes

Modified correlation Technique has been proposed to analyze the Microarray data and to search gene related information. We have used mean absolute deviation as a new approach for the differential expression analysis instead of the existing standard deviation and variances approaches and for calculating co-expression we have used spearmens correlation as a new technique for micro array analysis.