Nirmalya Chowdhury

Analyzing methods for path mining with applications in metabolomics

Metabolomics is one of the key approaches of systems biology that consists of studying biochemical networks having a set of metabolites, enzymes, reactions and their interactions. As biological networks are very complex in nature, proper techniques and models need to be chosen for their better understanding and interpretation. One of the useful strategies in this regard is using path mining strategies and graph-theoretical approaches that help in building hypothetical models and perform quantitative analysis. Furthermore, they also contribute to analyzing topological parameters in metabolome networks. Path mining techniques can be based on grammars, keys, patterns and indexing. Moreover, they can also be used for modeling metabolome networks, finding structural similarities between metabolites, in-silico metabolic engineering, shortest path estimation and for various graph-based analysis. In this manuscript, we have highlighted some core and applied areas of path-mining for modeling and analysis of metabolic networks.

Unsupervised text classification using kohonen's self organizing network

A text classification method using Kohonens Self Organizing Network is presented here. The proposed method can classify a set of text documents into a number of classes depending on their contents where the number of such classes is not known a priori. Text documents from various faculties of games are considered for experimentation. The method is found to provide satisfactory results for large size of data.

Image Object Classification Using Scale Invariant Feature Transform Descriptor with Support Vector Machine Classifier with Histogram Intersection Kernel

Recently much attention have been paid to region of interests in an image as they are useful in bridging the gap between high level image semantics and low level image features. In this paper we have proposed a method for classification of image objects produced by a standard image segmentation algorithm using multiclass support vector machine classifier integrated with histogram intersection kernel. SIFT is a relatively new feature descriptor which describes a given object in terms of a number of interest points. They are invariant to scaling, translation and partially invariant to illumination changes. This paper primarily focuses on the design of a fast and efficient image object classifier by combining the robust SIFT feature descriptor with intersection kernel SVM which is comparatively better than the existing kernel functions in terms of resource utilization. The experimental results show that the proposed method has good generalization accuracy.

To predict possible profit/loss of a movie to be launched using MLP with back-propagation learning

Film industry is the most important component of entertainment industry. Both profit and loss are very high for this business. Like every other business, business prediction system plays a vital role for this industry. Before release of a particular movie, if the Production Houses or distributors get any type of prediction that how the film will do business, then it will be very useful to reduce the risk of the investors. In this paper we have proposed a method using back propagation neural network for prediction about a given movies profitability. Note that, a similar such method has been successfully applied in the field of Stock Market Prediction, Weather Prediction and Image Processing.

Finding the Natural Groupings in a Data Set Using Genetic Algorithms

Genetic Algorithms (GAs) are generally portrayed as a search procedure which can optimize functions based on a limited sample of function values. In this paper, an objective function based on minimal spanning tree (MST) of data points is proposed for clustering and GAs have been used in an attempt to optimize the specified objective function in order to detect the natural grouping in a given data set. Several experiments on synthetic data set in \({\frak R}^2\) show the utility of the proposed method. The method is also applicable to any higher dimensional data.

Opinion Classification Based on Product Reviews from an Indian E-Commerce Website

Over the past decade, Indian e-commerce sector witnessed a huge growth. Currently this industry has approximately 40 million customers and it is expanding. These people express their experiences with various products, services in several websites, blogs, and social networking sites. To identify and extract any subjective knowledge from these huge unstructured user data, we need to develop a method that can collect, analyze, and classify user opinions. Two popular learning techniques (Supervised and Unsupervised) can be used to classify an opinion into two classes—“Positive” or “Negative.” In this paper, we propose an integrated framework for product review collection and unsupervised classification. The categorization of reviews is generated by the average semantic orientation of the phrases of suggestions or opinions in the review that holds adjectives as well as adverbs. A review can be categorized as an “Endorsed” one when the average semantic orientation is “Positive” otherwise it is an “Opposed” (“Negative”) one. Our proposed method has been tested on some real-life datasets collected from an Indian e-commerce website. The experimental results obtained show the efficiency of our proposed method for classification of product reviews.

A Nonnegative Matrix Factorization Based Approach to Extract Aspects from Product Reviews

Due to the unstructured nature of review text, it is very hard to develop an automated opinion mining application to compare various product models based on their various aspects to make a purchase decision. Over the year, various data mining techniques have been proposed to extract aspects of the products. In this paper, we have proposed a technique based on the nonnegative matrix factorization to extract aspects of a product category. Performance of our proposed method has been compared with a very popular aspect extraction technique based on probabilistic latent semantic analysis. We have also given a comparison between common aspects of a particular model under a specific product category from various manufacturers. These comparisons are based on the sentiments expressed by the users on these aspects. These sentiments expressed in various aspects have been extracted using an unsupervised technique.

To Detect the Influencers in a Dynamic Co-authorship Network Using Heat-Diffusion Model

A social network consists of a collection of social entities and interactions among these entities. It performs a crucial role as a platform for spreading various ideas, informations among its members. Influence analysis in social network has always been a fascinating topic in social network analysis due to its various application areas like targeted advertisement, recommendation system, outcome of a campaign, viral marketing, etc. Most of the social networks are dynamic in nature since the state of these networks evolves over time. Majority of earlier research works have been focused on the topics like influencer detection, influence maximization, and influence diffusion in a static network due to the complexity of constantly evolving social network. In this paper, we have proposed a method based on heat-diffusion process to detect influencers in a dynamic social network. The proposed model can also rank them based on the influence he or she has on others. We have applied our proposed method on the evolving co-authorship networks to detect and rank influential persons.

Classification of Bank Direct Marketing Data Using Subsets of Training Data

Nowadays, most business organizations practice Direct Marketing. One of the promising application areas of this type of marketing practice is Banking and Financial Industry. A classification technique using subsets of training data has been proposed in this paper. We have used a real-world direct marketing campaign data for experimentation. This marketing campaign was a telemarketing campaign. The objective of our experiment is to forecast the probability of a term-deposit plan subscription. In our proposed method we have used customer segmentation process to group individual customers according to their demographic feature. We have used X-means clustering algorithm for customer segmentation process. We have extracted few appropriate collection of customers from the entire customer database using X-means cluster algorithm, on the basis of demographic feature of individual customers. We have tested our proposed method of training for classifier using three most widely used classifiers namely Naive Bayes, Decision Tree and Support Vector Machine. It has been found that the result obtained using our proposed method for classification on the banking data is better compare to that reported in some previous work on the same data.

Concepts of relative sample outlier RSO and weighted sample similarity WSS for improving performance of clustering genes: co-function and co-regulation

Performance of clustering algorithms is largely dependent on selected similarity measure. Efficiency in handling outliers is a major contributor to the success of a similarity measure. Better the ability of similarity measure in measuring similarity between genes in the presence of outliers, better will be the performance of the clustering algorithm in forming biologically relevant groups of genes. In the present article, we discuss the problem of handling outliers with different existing similarity measures and introduce the concepts of Relative Sample Outlier (RSO). We formulate new similarity, called Weighted Sample Similarity (WSS), incorporated in Euclidean distance and Pearson correlation coefficient and then use them in various clustering and biclustering algorithms to group different gene expression profiles. Our results suggest that WSS improves performance, in terms of finding biologically relevant groups of genes, of all the considered clustering algorithms.