Amit Bagga

Categorizing images in Web documents

The Web provides an increasingly powerful and popular publication mechanism. Web documents often contain a large number of images that serve various purposes. Identifying the functional categories of these images is an important task in Web repurposing. This article describes a study on the functional categorization of Web images using data collected from news Web sites. As the popularity of the Web soars, the content on the Web is increasingly accessed from wireless devices that have small screens and different bandwidths. Because many Web documents contain a large number of images serving different purposes, how to identify the function of each image so that it can be handled accordingly is an important issue in Web content repurposing. Much work remains to be done in function-based image classification of all images. Icons that appear regularly on Web sites (for example, newspaper logos) could be classified by analyzing different editions of the pages for repetitions. For the host class, a combination of detecting repetitive images and face recognition should help significantly.

Call Center Customer Verification by Query-Directed Passwords

We introduce an authentication framework called Query-Directed Passwords (QDP) that incorporates the convenience of authentication by long-term knowledge questions and offers stronger security than from traditional types of personal questions. Security is strengthened for this scheme by imposing several restrictions on the questions and answers, and specifying how QDP is implemented in conjunction with other factors. Four QDP implementations are examined for call center applications. We examine the security and convenience of one of these implementations in detail. This implementation involves client-end storage of questions in a computer file or a wallet card, and follows a basic challenge-response authentication protocol.

Query-directed passwords

A classical tradeoff in the field of user authentication is between user convenience and system security. Should users authenticate themselves with their mothers maiden name, which is easily recalled but not very secure; or should they memorize a long, random password that is secure but unmemorable? In recent years, tokens and biometrics have been offered as the answer to this convenience-versus-security conflict; however, these require infrastructure modifications. We introduce query-directed passwords (QDP), an authentication procedure based on questions and answers - where the answers are known, not memorized. QDP is particularly convenient for infrequent use, such as monthly or yearly authentication to seldom-accessed accounts. Applications are described that capitalize on advantages of QDP. One of these is an automated password recovery system where testing showed a reduced use of Help Desk personnel for repeated, forgotten passwords from 20% to 2.7%. We discuss other applications, experimental results, and future research directions.

Email classification for contact centers

The explosive growth of the Internet has made email an integral part of business communication. Therefore, business customer service centers, or contact centers, are processing larger amounts of email interactions with customers. In this paper we discuss a preliminary email routing and classification system that filters and classifies incoming email messages upon their content. A module first attempts to identify and filter those email messages that do not require immediate (if any) responses. We call such email messages single messages. The emails that do require immediate responses are called root messages. A second module classifies messages in categories that characterize the type of interaction between the contact center operators and the customers. Emails that are involved in such interactions form a thread and can be classified broadly into one of three categories: root, inner, and leaf. Root messages are those that start a thread while a leaf message is the final email sent in an interaction. All other emails in the interaction are considered to be inner messages.

Identifying story and preview images in news web pages

The World Wide Web provides an increasingly powerfuland popular publication mechanism. Web documents oftencontain a large number of images serving various differentpurposes. This paper focuses on images that are associatedwith a story or preview to a story. Such images often accompanythe key content on a web page, thus their identificationis important for applications such as web page summarizationand mobile access. We present a novel algorithmfor automatic identification of story/preview images whichcombines features extracted from both the image itself andthe surrounding text. The effectiveness of this algorithm isdemonstrated by experimental results on over 1500 imagescollected from 25 news web sites.

Multi-source combined-media video tracking for summarization

Video summarization is receiving increasing attention due to the large amount of video content made available on the Internet. We present an idea to track video from multiple sources for video summarization. An algorithm that takes advantage of both video and closed caption text information for video scene clustering is described. Experimental results are given followed by discussion on future directions.

Combined-media video tracking for summarization

Video summarization is receiving increasing attention due to the large amount of video content made available on the Internet. In this paper we present a novel idea to track video from multiple sources for video summarization. An algorithm that takes advantage of both video and close caption text information for video scene clustering is described. Experimental results are given followed by discussion on future directions.

Categorizing images in web documents

The World Wide Web provides an increasingly powerful and popular publication mechanism. Web documents often contain a large number of images serving various different purposes. Identifying the functional categories of these images ahs important applications including information extraction, web mining, web page summarization and mobile access. An important first step towards designing algorithms for automatic categorization of images on the web is to identify the common categories and examine their properties and characteristics. This paper describes results from such an initial study using data collected from news web sites. We describe the image categories found in such web pages and their distributions, and identify the main research issues involved in automatically classifying images into these categories.

Avaya Interactive Dashboard (AID): An Interactive Tool for Mining the Avaya Problem Ticket Database

In this paper we describe an interactive tool called the Avaya Interactive Dashboard, or AID, that was designed to help Avaya’s services organization to mine the text fields in Maestro. AID allows engineers to quickly and conveniently drill down to discover patterns and/or verify intuitions with a few simple clicks of the mouse. The interface has a web-based front-end that interacts through CGI scripts with a central server implemented mostly in Java. The central server in turn interacts with Maestro as needed.