Chetan Kumar Verma

A Realistic Small-World Model for Wireless Mesh Networks

Small-world network concept deals with the addition of a few Long-ranged Links (LLs) to significantly bring down the average path length (APL) of the network. The existing small-world models do not consider the real constraints of wireless networks such as the transmission range of LLs, limited radios per mesh router, and limited bandwidth for wireless links, therefore, we propose C-SWAWN (Constrained Small-World Architecture for Wireless Network) model for Wireless Mesh Networks (WMNs). We then propose three LL addition strategies for reducing APL to the centrally placed Gateway node in WMNs. In moderately large WMNs, a 43% reduction in APL to Gateway can be achieved with the addition of 10% LLs (with respect to number of mesh routers) in our C-SWAWN model with greedy LL addition strategy. Detailed studies show realistic performance benefits with application of small-world concept in WMNs.

Improving Scalability of Personalized Recommendation Systems for Enterprise Knowledge Workers

Enterprise knowledge workers have been overwhelmed by the growing rate of incoming data in recent years. In this paper, we present a recommendation system with the goal of helping knowledge workers in discovering useful new content. In particular, our system builds personalized user models based on file activities on enterprise network file servers. Our models use novel features that are derived from file metadata and user collaboration. Through extensive evaluation on real-world enterprise data, we demonstrate the effectiveness of our system with high precision and recall values. Unfortunately, our experiments reveal that per-user models are unable to handle heavy workloads. To address this limitation, we propose a novel optimization technique, active feature-based model selection, that predicts the user models that should be applied on each test file. Such a technique can reduce the classification time per file by as much as 23 times without sacrificing accuracy. We also show how this technique can be extended to improve the scalability exponentially at marginal cost of prediction accuracy, e.g., we can gain 169 times faster performance on an average across all shares by sacrificing 4% of F-score.

New link addition strategies for Multi-Gateway small world Wireless Mesh Networks

Small-world network concept deals with the addition of a few Long-ranged Links (LLs) in a network to significantly bring down the average path length (APL) of the network. Existing small-world models do not consider the presence of multiple gateways and, therefore, we propose Multi-Gateway Aware LL addition Strategy (M-GAS). Further, the presence of multiple gateways brings forth the additional issue of traffic load balancing. We modify the M-GAS to propose Load balanced M-GAS (LM-GAS) for load balancing in small-world WMNs. For uniform and random placement of gateways, we present results from M-GAS and LM-GAS strategies. Our results provide early insights in achieving high load balancing in small world WMNs.

Access Prediction for Knowledge Workers in Enterprise Data Repositories

The data which knowledge workers need to conduct their work is stored across an increasing number of repositories and grows annually at a significant rate. It is therefore unreasonable to expect that knowledge workers can efficiently search and identify what they need across a myriad of locations where upwards of hundreds of thousands of items can be created daily. This paper describes a system which can observe user activity and train models to predict which items a user will access in order to help knowledge workers discover content. We specifically investigate network file systems and determine how well we can predict future access to newly created or modified content. Utilizing file metadata to construct access prediction models, we show how the performance of these models can be improved for shares demonstrating high collaboration among its users. Experiments on eight enterprise shares reveal that models based on file metadata can achieve F scores upwards of 99%. Furthermore, on an average, collaboration aware models can correctly predict nearly half of new file accesses by users while ensuring a precision of 75%, thus validating that the proposed system can be utilized to help knowledge workers discover new or modified content.

Methods to Obtain Training Videos for Fully Automated Application-Specific Classification

Personalization approaches seek to estimate user preferences in order to recommend content or social network connections, or to serve personalized advertisements to users. Such approaches are being increasingly adopted by organizations to build customized personalization applications. Leveraging the growing popularity of Web videos for such approaches necessitates the ability to classify Web videos into application-specific categories, since different applications are interested in different aspects of the user preferences. A key requirement of supervised classification models to address this is the availability of training videos labeled to the arbitrary application-specific categories. In order to address this requirement, we propose a completely automated framework to obtain training Web videos for arbitrary categories, which does not rely on any manual labeling of videos. This is achieved utilizing keywords to retrieve training videos, thereby simplifying the problem of obtaining training videos to the problem of selecting keywords to retrieve them. We show that there are two opposing objectives (proximity and diversity) that need to be considered while developing such keyword selection techniques. We propose two efficient approaches (linear combination of proximity and diversity and annealing-based alternating optimization) and study the tradeoffs between them, with respect to performance and the human input required to tune parameters of the approach. Through experiments over several sets of categories, we demonstrate the feasibility of the automated framework to select training videos for application-specific categorization. We also show that the proposed approaches lead to a substantial improvement in the performance of classification models, as compared with other automated methods.

Construction of tag ontological graphs by locally minimizing weighted average hops

We present a data-driven approach for the construction of ontological graphs on a set of image tags obtained from annotated image corpus. We treat each tag as a node in a graph, and starting with a preliminary graph obtained using WordNet, we propose the graph construction as a refinement of the preliminary graph using corpus statistics. Towards this, we formulate an optimization problem which is solved using a local search based approach. To evaluate the constructed ontological graphs, we propose a novel task which involves associating test images with tags while observing partial set of associated tags.

Fully Automated Learning for Application-Specific Web Video Classification

Personalization applications such as content recommendations, product recommendations and advertisements, and social network related recommendations, can be quite beneficial for both service providers and users. Such applications need to understand user preferences in order to provide customized services. As user engagement with web videos has grown significantly, understanding user preferences based on videos viewed looks promising. The above requires ability to classify web videos into a set of categories appropriate for the personalization application. However, such categories may be substantially different from common categories like Sports, Music, Comedy, etc. used by video sharing websites, leading to lack of labeled training videos for such categories. In this paper, we study the feasibility and effectiveness of a fully automated framework to obtain training videos to enable classification of web videos to any arbitrary set of categories, as desired by the personalization application. We investigate the desired properties in training data that can lead to high performance of the trained classification models. We then develop an approach to identify and score keywords based on their suitability to retrieve training videos, with the desired properties, for the specified set of categories. Experimental results on several sets of categories demonstrate the ability of the proposed approach to obtain effective training data, and hence achieve high video classification performance.