Sanasam Ranbir Singh

Influence of edge weight on node proximity based link prediction methods

Tie weight plays an important role in maintaining cohesiveness of social networks. However, influence of the tie weight on link prediction has not been clearly understood. In few of the previous studies, conflicting observations have been reported. In this paper, we revisit the study of the effect of tie weight on link prediction. Previous studies have focused on additive weighting model. However, the additive model is not suitable for all node proximity based prediction methods. For understanding the effect of weighting models on different prediction methods, we propose two new weighting models namely, min-flow and multiplicative. The effect of all three weighting models on node proximity based prediction methods over ten datasets of different characteristics is thoroughly investigated. From several experiments, we observe that the response of different weighting models varies, subject to prediction methods and datasets. Empirically, we further show that with the right choice of a weighting model, weighted versions may perform better than their unweighted counterparts.We further extend the study to show that proper tuning of the weighting function also influences the prediction performance. We also present an analysis based on the properties of the underlying graph to justify our result. Finally, we perform an analysis of the weak tie theory, and observe that unweighted models are suitable for inter-community link prediction, and weighted models are suitable for intra-community link prediction.

Internet activity analysis through proxy log

The availability of the Internet at the click of a mouse brings with it a host of new problems. Although the World Wide Web was first started by physicists at CERN to enable collation and exchange of data, today, it is used for a wide range of applications. The requirements on bandwidth for each of the applications is also varied. An Internet Service Provider must ensure satisfaction across the entire spectrum of users. To ensure this, analysis of Internet usage becomes essential. Further, an administrator can keep a record of users Internet activity and prevent unethical activities, since the Internet is also an excellent resource for providing anonymity. This analysis can also help in resource provisioning and monitoring. In this work, a web-based tool is first proposed to analyse the Internet activity. Next, data is collected from a proxy server at a campus-wide network. Traffic patterns of different types of users are studied. Finally, the paper concludes with strategies for monitoring and control of traffic.

CloseMiner: discovering frequent closed itemsets using frequent closed tidsets

Complete set of itemsets can be grouped into non-overlapping clusters identified by closed tidsets. Each cluster has only one closed itemset and is the superset of all itemsets with the same support. Number of closed itemsets is identical to the number of clusters. Therefore, the problem of discovering closed itemsets can be considered as the problem of clustering the complete set of itemsets by closed tidsets. In this paper, we present CloseMiner, a new algorithm for discovering all frequent closed itemsets by grouping the complete set of itemsets into non-overlapping clusters identified by closed tidsets. An extensive experimental evaluation on a number of real and synthetic databases shows that CloseMiner outperforms Apriori and CHARM.

A syllable-based framework for unit selection synthesis in 13 Indian languages

In this paper, we discuss a consortium effort on building text to speech (TTS) systems for 13 Indian languages. There are about 1652 Indian languages. A unified framework is therefore attempted required for building TTSes for Indian languages. As Indian languages are syllable-timed, a syllable-based framework is developed. As quality of speech synthesis is of paramount interest, unit-selection synthesizers are built. Building TTS systems for low-resource languages requires that the data be carefully collected an annotated as the database has to be built from the scratch. Various criteria have to addressed while building the database, namely, speaker selection, pronunciation variation, optimal text selection, handling of out of vocabulary words and so on. The various characteristics of the voice that affect speech synthesis quality are first analysed. Next the design of the corpus of each of the Indian languages is tabulated. The collected data is labeled at the syllable level using a semiautomatic labeling tool. Text to speech synthesizers are built for all the 13 languages, namely, Hindi, Tamil, Marathi, Bengali, Malayalam, Telugu, Kannada, Gujarati, Rajasthani, Assamese, Manipuri, Odia and Bodo using the same common framework. The TTS systems are evaluated using degradation Mean Opinion Score (DMOS) and Word Error Rate (WER). An average DMOS score of ≈3.0 and an average WER of about 20 % is observed across all the languages.

Inference Based Query Expansion Using User’s Real Time Implicit Feedback

Query expansion is a commonly used technique to address the problem of short and under-specified search queries in information retrieval. Traditional query expansion frameworks return static results, whereas user’s information needs is dynamics in nature. User’s search goal, even for the same query, may be different at different instances. This often leads to poor coherence between traditional query expansion and user’s search goal resulting poor retrieval performance. In this study, we observe that user’s search pattern is influenced by his/her recent searches in many search instances. We further propose a query expansion framework which explores user’s real time implicit feedback provided at the time of search to determine user’s search context and identify relevant query expansion terms. From extensive experiments, it is evident that the proposed query expansion framework adapts to the changing needs of user’s information need.

Link Prediction Using Social Network Analysis over Heterogeneous Terrorist Network

Social network analysis (SNA) has been effectively used in counter-terrorism analysis by generating homogeneous network. In this paper, we consider a large dataset reporting various terrorist attacks over the globe and represent the dataset as a heterogeneous network. The objective of this paper is to the explore the effect of various link prediction frameworks such as topic modeling, network topology and graph kernels. We propose bipartite based link prediction over topic feature relationship, heterogeneous version of node proximity based link prediction and graph kernel methods. From various experimental observation, it is evident that bipartite method based on topic modeling also return comparable results (sometimes better) as that of node proximity and graph kernel.

An algorithm for discovering the frequent closed itemsets in a large database

Previous research revealed that the problem of discovering a complete set of frequent itemsets from a large database can be reduced to the problem of discovering the frequent closed itemsets, and this process results in a much smaller set of itemsets without information loss. This article is based on the observation that the set of all itemsets can be grouped into non-overlapping clusters such that each cluster is identified by a unique closed tidset. It is also found that there is only one closed itemset in each cluster and it is the superset of all itemsets with the same support. Therefore, the problem of discovering closed itemsets can be further considered as the problem of clustering the set of itemsets and then identifying each cluster by a unique closed tidset. This article presents CloseMiner, a new algorithm for discovering all frequent closed itemsets by grouping the set of itemsets into non-overlapping clusters. Experimental evaluation based on a number of real and synthetic databases has proved that CloseMiner outperforms the existing systems APRIORI and CHARM.

Exploiting reciprocity toward link prediction

This paper addresses link prediction problem in directed networks by exploiting reciprocative nature of human relationships. It first proposes a null model to present evidence that reciprocal links influence the process of “triad formation”. Motivated by this, reciprocal links are exploited to enhance link prediction performance in three ways: (a) a reciprocity-aware link weighting technique is proposed, and existing weighted link prediction methods are applied over the resultant weighted network; (b) new link prediction methods are proposed, which exploit reciprocity; and (c) existing and proposed methods are combined toward supervised prediction to enhance the prediction performance further. All experiments are carried out on two real directed network datasets.

Modeling evolution of a social network using temporalgraph kernels

Majority of the studies on modeling the evolution of a social network using spectral graph kernels do not consider temporal effects while estimating the kernel parameters. As a result, such kernels fail to capture structural properties of the evolution over the time. In this paper, we propose temporal spectral graph kernels of four popular graph kernels namely path counting, triangle closing, exponential and neumann. Their responses in predicting future growth of the network have been investigated in detail, using two large datasets namely Facebook and DBLP. It is evident from various experimental setups that the proposed temporal spectral graph kernels outperform all of their non-temporal counterparts in predicting future growth of the networks.

Effect of word density on measuring words association

The study of mining the associated words is not new. Because of its wide ranges of applications, it is still an important issue in Information Retrieval. The existing estimators such as joint probability, words association norm do not consider the density of the words present in each window. In this paper, we incorporate the word density and propose estimator based on word density to measure the association between the words. From various experimental results based on the human judgments and precision collected from search engines, we find that the precision of the estimators could be improved by incorporating word density. For all ranges of the size of the windows, our estimator outperforms all other estimators. We also observe that all these estimators (both existing and proposed one) perform relatively better when the windows contain around five sentences. We also show by using Spearman rank-order correlation coefficient that our estimator returns better quality of the ranking of the associated terms.