G. P. Sajeev

A novel content classification scheme for web caches

Web caches are useful in reducing the user perceived latencies and web traffic congestion. Multi-level classification of web objects in caching is relatively an unexplored area. This paper proposes a novel classification scheme for web cache objects which utilizes a multinomial logistic regression (MLR) technique. The MLR model is trained to classify web objects using the information extracted from web logs. We introduce a novel grading parameter worthiness as a key for the object classification. Simulations are carried out with the datasets generated from real world trace files using the classifier in Least Recently Used-Class Based (LRU-C) and Least Recently Used-Multilevel Classes (LRU-M) cache models. Test results confirm that the proposed model has good online learning and prediction capability and suggest that the proposed approach is applicable to adaptive caching.

Effective web personalization system based on time and semantic relatedness

The key aspect in building a Web personalization system is the users navigational pattern. However, the navigational pattern alone is insufficient to capture the users interest and behavior. This paper proposes a novel web personalization system that accepts the timing information, semantic information along with the navigational pattern, and classifies the users according their interest and behavior on the site. The proposed model is validated by constructing a Web personalization model using the real and synthetic data and the results are promising.

Building Web Personalization System with Time-Driven Web Usage Mining

Web personalization is a powerful tool used for personalizing the Websites. The personalization system aims at suggesting the Web pages to the users based on their navigational patterns. Use of attributes such as time, popularity of Web objects makes the model more efficient. This paper proposes a novel Web personalization model which utilizes time attributes, such as duration of visit, inter-visiting time, burst of visit, and the users navigational pattern. Test results indicate that the proposed model explores the users behaviour and their interest.

Building semi-intelligent web cache systems with lightweight machine learning techniques ☆

Abstract Web caches are used to address the problem of access delay and network congestion in the Internet. The conventional caching methods are, in general, not efficient in dealing with the problems of web cache admission control and replacement. Intelligent or machine learning-based techniques could be used to enhance the web cache performance. However, such techniques generally suffer from huge computational overheads, making them less effective. This research develops a semi-intelligent approach for web cache admission and replacement using a multinomial web object classifier. The performance of this classifier is assessed through simulation experiments using real trace data, which are compared with Least Recently Used (LRU), Least Frequently Used (LFU) and Greedy Dual Size Frequency (GDSF) schemes. The test results show that a properly trained multinomial logistic regression (MLR) model yields better cache performance in terms of hit ratios and disk space utilization. The performance of this lightweight MLR based classification and caching model is examined in comparison with the heavyweight Artificial Neural Network (ANN) based model and the results are encouraging.

A novel approach for book recommendation systems

Recommendation systems are widely used in ecommerce applications. A recommendation system intends to recommend the items or products to a particular user, based on users interests, other users preferences, and their ratings. To provide a better recommendation system, it is necessary to generate associations among products. Since e-commerce and social networking sites generates massive data, traditional data mining approaches perform poorly. Also, the pattern mining algorithm such as the traditional Apriori suffers from high latency in scanning the large database for generating association rules. In this paper we propose a novel pattern mining algorithm called as Frequent Pattern Intersect algorithm (FPIntersect algorithm), which overcomes the drawback of Apriori. The proposed method is validated through simulations, and the results are promising.

Analyzing the Long Range Dependence and Object Popularity in Evaluating the Performance of Web Caching

Web cache systems enhance Web services by reducing the client side latency. To deploy an effective Web cache, analysis of the traffic characteristics is indispensable. Various reported results of traffic analysis show evidences of long range dependence (LRD) in the data stream and rank distribution of the documents in Web traffic. This article analyzes Web cache traffic properties like LRD and rank distribution based on the traces collected from NLANR (National Laboratory of Applied Network Research) cache servers. Traces are processed to investigate the performance of Web cache servers and traffic patterns. Statistical tools are utilized to measure the strengths of the LRD and popularity. The Hurst parameter, which is a measure of the LRD, is estimated using various statistical methods. It is observed that the presence of LRD in the traffic is feeble and does not influence the Web cache performance.

Adaptive web personalization system using splay tree

Web personalization helps in understanding the user interests and creating customized experiences for users. However the user preferences changes dynamically over a period. In order to adapt with the changing information needs of the user, we have developed a novel web personalization system that captures the user changing interest by analyzing the timing information. We use splay tree, which is a self-adaptive data structure, for tracking the changing trends of the users. The proposed web personalization model is validated by building a simulation model, with real and synthetic dataset, and the quality of results are promising.

A novel web crawling method for vertical search engines

The main goal of focused web crawlers is to retrieve as many relevant pages as possible. However, most of the crawlers use page rank algorithm to lineup the pages in the crawler frontier. Since the page rank algorithm suffers from the drawback of “Richer get rich phenomenon”, focused crawlers often fail to retrieve the hidden relevant pages. This paper presents a novel approach for retrieving the hidden and relevant pages by combining rank and semantic similarity information. The model is validated by crawling the real web with different topics and the results are promising.

Task recommendation in reward-based crowdsourcing systems

Crowdsourcing systems are distributed problem solving platforms, where small tasks are channelled to a crowd in the form of open calls for solutions. Reward based crowdsourcing systems tries to attract the interested and capable workers to provide solutions in return for monetary rewards. We study the task recommendation problem in reward based crowdsourcing platforms, where we leverage both implicit and explicit features of the worker-reward and worker-task attributes. Given a set of workers, set of tasks, participation, winner attributes, we intend to recommend tasks to workers by exploiting interactions between tasks and workers. Two models based on worker-reward based features and worker task based features are presented. The proposed approach is compared with multiple related techniques using real world dataset.

Entropy based informative content density approach for efficient web content extraction

Web content extraction is a popular technique for extracting the main content from web pages and discards the irrelevant content. Extracting only the relevant content is a challenging task since it is difficult to determine which part of the web page is relevant and which part is not. Among the existing web content extraction methods, density based content extraction is one popular method. However density based methods, suffer from poor efficiency, especially when the pages containing less information and long noise. We propose a web content extraction technique build on Entropy based Informative Content Density algorithm (EICD). The proposed EICD algorithm initially analyses higher text density content. Further, the entropy-based analysis is performed for selected features. The key idea of EICD is to utilize the information entropy for representing the knowledge that correlates to the amount of informative content in a page. The proposed method is validated through simulation and the results are promising.