Anuja Arora

Extended Biogeography Based Optimization for Natural Terrain Feature Classification from Satellite Remote Sensing Images

Remote sensing image classification in recent years has been a proliferating area of global research for obtaining geo-spatial information from satellite data. In Biogeography Based Optimization (BBO), knowledge sharing between candidate problem solutions or habitats depends on the migration mechanisms of the ecosystem. In this paper an extension to Biogeography Based-Optimization is proposed for image classification by incorporating the non-linear migration model into the evolutionary process. It is observed in recent literature that sinusoidal migration curves better represent the natural migration phenomenon as compared to the existing approach of using linear curves. The motivation of this paper is to apply this realistic migration model in BBO, from the domain of natural computing, for natural terrain features classification. The adopted approach calculates the migration rate using Rank- based fitness criteria. The results indicate that highly accurate land-cover features are extracted using the extended BBO technique.

Text and image based spam email classification using KNN, Naïve Bayes and Reverse DBSCAN algorithm

Internet has changed the way of communication, which has become more and more concentrated on emails. Emails, text messages and online messenger chatting have become part and parcel of our lives. Out of all these communications, emails are more prone to exploitation. Thus, various email providers employ algorithms to filter emails based on spam and ham. In this research paper, our prime aim is to detect text as well as image based spam emails. To achieve the objective we applied three algorithms namely: KNN algorithm, Naïve Bayes algorithm and reverse DBSCAN algorithm. Pre-processing of email text before executing the algorithms is used to make them predict better. This paper uses Enron corpuss dataset of spam and ham emails. In this research paper, we provide comparison performance of all three algorithms based on four measuring factors namely: precision, sensitivity, specificity and accuracy. We are able to attain good accuracy by all the three algorithms. The results have shown comparison of all three algorithms applied on same data set.

A bug Mining tool to identify and analyze security bugs using Naive Bayes and TF-IDF

Bug report contains a vital role during software development, However bug reports belongs to different categories such as performance, usability, security etc. This paper focuses on security bug and presents a bug mining system for the identification of security and non-security bugs using the term frequency-inverse document frequency (TF-IDF) weights and naïve bayes. We performed experiments on bug report repositories of bug tracking systems such as bugzilla and debugger. In the proposed approach we apply text mining methodology and TF-IDF on the existing historic bug report database based on the bug s description to predict the nature of the bug and to train a statistical model for manually mislabeled bug reports present in the database. The tool helps in deciding the priorities of the incoming bugs depending on the category of the bugs i.e. whether it is a security bug report or a non-security bug report, using naïve bayes. Our evaluation shows that our tool using TF-IDF is giving better results than the naïve bayes method.

Content management system : Comparative case study

In this paper, the various features essential for a content management system are explored. This includes a comparison of various open source CMS, their features and a conclusion on improvements that will be made in order to alleviate problems. Our aim was to find out the best available open source CMS and implement it for our website. For this we conducted a study on 13 java based open source CMS available taking into account 29 features like the technology used for creation of the CMS and features like anti plagiarism, Digital Asset Management(DAM), etc. for user convenience.

Brand analysis framework for online marketing: ranking web pages and analyzing popularity of brands on social media

Online marketing is one of the best practices used to establish a brand and to increase its popularity. Advertisements are used in a better way to showcase the company’s product/service and give rise to a worthy online marketing strategy. Posting an advertisement on utilitarian web pages helps to maximize brand reach and get a better feedback. Now-a-days companies are cautious of their brand image on the Internet due to the growing number of Internet users. Since there are billions of Web sites on the Internet, it becomes difficult for companies to really decide where to advertise on the Internet for brand popularity. What if, the company advertise on a page which is visited by less number of the interested (for a particular type of product) users instead of a web page which is visited by more number of the interested users?—this doubt and uncertainty—is a core issue faced by many companies. This research paper presents a Brand analysis framework and suggests some experimental practices to ensure efficiency of the proposed framework. This framework is divided into three components—(1) Web site network formation framework—a framework that forms a Web site network of a specific search query obtained from resultant web pages of three search engines-Google, Yahoo & Bing and their associated web pages; (2) content scraping framework—it crawls the content of web pages existing in the framework-formed Web site network; (3) rank assignment of networked web pages—text edge processing algorithm has been used to find out terms of interest and their occurrence associated with search query. We have further applied sentiment analysis to validate positive or negative impact of the sentences, having the search term and its associated terms (with reference to the search query) to identify impact of web page. Later, on the basis of both—text edge analysis and sentiment analysis results, we assigned a rank to networked web pages and online social network pages. In this research work, we present experiments for ‘Motorola smart phone,’ ‘LG smart phone’ and ‘Samsung smart phone’ as search query and sampled the Web site network of top 20 search results of all three search engines and examined up to 60 search results for each search engine. This work is useful to target the right online location for specific brand marketing. Once the brand knows the web pages/social media pages containing high brand affinity and ensures that the content of high affinity web page/social media page has a positive impact, we advertise at that respective online location. Thus, targeted brand analysis framework for online marketing not only has benefits for the advertisement agencies but also for the customers.

An improved approach to English-Hindi based Cross Language Information Retrieval system

Cross Language Information Retrieval (CLIR) is a sub domain of Information Retrieval. It deals with retrieval of information in a specified language that is different from the language of users query. In this paper, an improved English-Hindi based CLIR is proposed. There are various un-noticed domains in this broad research area that are required to be worked upon in order to improve the performance of an English-Hindi based CLIR. Not much research effort has been put up to improve the searching and ranking aspects of CLIR systems, especially in case of English-Hindi based CLIR. This paper focuses on applying algorithms like Naïve Bayes and particle swarm optimization in order to improve ranking and searching aspects of a CLIR system. We matched terms contained in documents to the query terms in same sequence as present in the search query to make our system more efficient. Along with this our approach also makes use of bilingual English-Hindi translator for query conversion in Hindi language. Further, we use Hindi query extension and synonym generation which helps in retrieving more relevant results in an English-Hindi based CLIR as compared to existing one. Both of these techniques applied to this improved approach gives user a change to choose more appropriate Hindi query than just by using the single translated query and hence improving overall performance.

Web Search Personalization Using Ontological User Profiles

In web, users with different interest and goal enter queries to the search engine. Search engines provide all these users with the same search results irrespective of their context and interest. Therefore, the user has to browse through many results most of which are irrelevant to his goal. Personalization of search results involves understanding the user’s preferences based on his interaction and then re-ranking the search results to provide more relevant searches. We present a method for search engine to personalize search results leading to better search experience. In this method, a user profile is generated using reference ontology. The user profile is updated dynamically with interest scores whenever, he clicks on a webpage. With the help of these interest scores in the user profile, the search results are re-ranked to give personalized results. Our experimental results show that personalized search results are effective and efficient.

State based test case generation using VCL-GA

This paper presents a method for test cases generation from a state machine. This paper investigates variable chromosome length genetic algorithm for automatically generating state based test cases. For efficiency we developed a Genetic algorithm that optimize and select test case which are covering maximum number of states and transitions on distinct chromosome length. Variable chromosome Length Genetic algorithm was requisite because of throttle in representing FSM as a binary chromosome discussed in this research paper and various issues discussed in paper. In this paper, Genetic Algorithm starts with a short chromosome length and retrieves the best solution with respect to optimized test cases covering maximum states and maximum transition of chosen state machine. The best solutions are then moved to the next stage with longer chromosome length. In this paper population size is again varying on the basis of chromosome length. To validate the introduced technique, experiment with the test-data-generation technique has been applied on numerous state machines.

Cross domain recommendation using multidimensional tensor factorization

Abstract In the era of social media, exponential growth of information generated by online social media and e-commerce applications demands expert and intelligent recommendation systems. It has become one of the most valuable tools as it reduces problems such as information overload while selecting and suggesting friends, items, products, jobs etc according to users’ interests. Recommendation uses Collaborative Filtering as one of the most popular approaches but the major limitations of this approach are sparsity and cold-start issues. Mostly existing recommendation systems focus on a single domain, on the other end cross-domain collaborative filtering is able to alleviate the degree of sparsity and cold-start problems to a better extent. To avoid these problems, cross domain evolution comes in limelight and has become an emerging topic of research nowadays. This paper mainly discusses the notion of cross-domain recommendation, its techniques and proposes a generalized Cross Domain- Multi Dimension Tensor Factorization (CD-MDTF) approach to trade off influence among domains optimally. Cross Domain recommendation system employs knowledge from source domain and commingles it to target domain which covers the aspect of intelligent behavior and brings it to the category of an expert system. Finally, to evaluate the proposed CD-MDTF approach, experiments are performed on two real-world datasets, Movie-Lens and Book-Crossing. Results validate that sparsity and cold start problem is reduced by 16% and 25% respectively in comparison to single-domain recommendation systems. Further, the proposed CD-MDTF recommendation system accuracy is validated using precision and recall as evaluation performance metrics which shows an improvement of 41% in precision and 21% in recall. The results show that embedding of multiple domains and multiple dimensions for recommendation helps in result improvement, thereby augmenting the recommendation system performance like an expert and intelligent system.

Performance evaluation of LSB and LSD in steganography

In this work we have implemented and discussed the performance evaluation of Least Significant Bit (LSB) and Least Significant Digit (LSD) on various formats of multimedia data. We have shown the performance variation on different formats for which these two techniques have been applied to hide the messages. Implementation of both the algorithms has been done to explore the security and distortion level in different formats. Based on the implementation and exhaustive testing of the methods on documents, it was found that they help in proper hiding of messages so that it is not recovered by the intruder during the transfer of data from sender to the receiver.