Sujeet Pradhan

A query model to synthesize answer intervals from indexed video units

While a query result in a traditional database is a subset of the database, in a video database, it is a set of subintervals extracted from the raw video sequence. It is very hard, if not impossible, to predetermine all the queries that will be issued in the future, and all the subintervals that will become necessary to answer them. As a result, conventional query frameworks are not applicable to video databases. We propose a new video query model that computes query results by dynamically synthesizing needed subintervals from fragmentary indexed intervals in the database. We introduce new interval operations required for that computation. We also propose methods to compute relative relevance of synthesized intervals to a given query. A query result is a list of synthesized intervals sorted in the order of their degree of relevance.

Towards a novel desktop search technique

The most serious challenges Personal Information Management Systems face today are the results of having to deal with a large number of heterogeneous types of data from diverse data sources, but having no means of managing and searching them in a convenient, unified fashion. We argue that simplicity and flexibility are essential attributes for the next-generation search tools to respond to these challenges. This paper lays out specific issues to realizing such a tool in the context of desktop search and ties them to existing search techniques employed by Database Management Systems and Information Retrieval -- the two leading disciplines in search technology. We propose a novel technique for desktop search and show how our combined database and information retrieval approach to searching heterogeneous desktop data is going to benefit a large community of users.

A query model for retrieving relevant intervals within a video stream

The nature of video is such that even an hour long video data may contain a large number of meaningful intervals. Manual identification of all such intervals is practically infeasible. There has been some success in automatically parsing and indexing video data through the integration of technologies such as image processing, speech/character recognition, and natural language understanding. However, even by applying such techniques, complete identification of all the intervals required for answering all possible queries cannot be achieved. As a result, using the current state-of-art techniques, whether automatic or manual, it is only fragmentary video intervals that can be successfully indexed. Our goal is to retrieve meaningful intervals within such fragmentarily indexed video streams. We propose a new set of algebraic operations which enable us to compose all the intervals that are conceivably relevant to a query. Since these operations may compose even irrelevant intervals, we provide a mechanism to exclude as many of them as possible from the answer set.

Retrieving Relevant Portions from Structured Digital Documents

Retrieving relevant portions from structured documents consisting of logical components has been a challenging task in both the database and the information retrieval world, since an answer related to a query may be split across multiple components. In this paper, we propose a query mechanism that applies database style query evaluation in response to IR style keyword-based queries for retrieving relevant answers from a logically structured document. We first define an appropriate semantics of keywords-based queries and then propose an algebra that is capable of computing every relevant portion of a document, which can be considered answer to a set of arbitrary keywords. The ordering and structural relationship among the components are preserved in the answer. We also introduce several practically useful filters that saves users from having to deal with an overwhelming number of answers.

A New Algebraic Approach to Retrieve Meaningful Video Intervals from Fragmentarily Indexed Video Shots

Video data consists of a sequence of shots. Over the past several years, substantial progress has been made in automatically detecting shot boundaries based on changes of visual and/or audio characteristics. There has also been considerable progress in indexing such video shots by automatically extracting keywords using techniques such as speech and text recognition. Shots detected by those techniques, however, are very fragmental. A single shot itself is rarely self-contained and therefore may not carry enough information to be a meaningful unit. A meaningful interval that interests common users generally spans several consecutive shots. There hardly exists any reliable technique for identifying all such meaningful intervals in advance so that any possible query can be answered.

Towards an integrated framework for querying collection of heterogeneous data

Keywords queries on various data models have been studied in the past. Regardless of the nature of target data, a keyword query poses a common problem: the difficulty in determining a suitable retrieval unit. In this paper, we discuss this problem in a new scenario that assumes several different types of data being stored and queried in a unified framework. A hypothetical general query model that borrows the ideas of our previous model designed for XML and linear video data is presented. We then state the requirements and challenges to realize such a model.

Two-Dimensional Retrieval in Synchronized Media Streams

We propose a two-dimensional information retrieval model for querying inside multimedia database where numerous media streams are semantically synchronized. The model exploits spatio-temporal arrangement of synchronized media streams during query evaluation phase. The output is a list of on-demand, dynamically integrated multimedia presentations satisfying a set of query keywords.

A Framework for the Relational Implementation of Tree Algebra to Retrieve Structured Document Fragments

Naive users typically query documents with keywords. The problem of retrieval unit when keyword queries are posed against a structured document consisting of several logical components has been studied in the past. We developed a new query model based on tree algebra, which successfully resolves this problem. However, one important issue any such effective theoretical model has to deal with, is the difficulty in its equally effective implementation. In this paper, we overview our query model and explore how this model can be successfully implemented using an existing relational database technology. Tree nodes representing logical components of a structured document are indexed with their pre-order and post-order rankings and stored as a relation. We then show how the basic algebraic operation defined in our query model can be transformed into a simple SQL query against this relation. We also discuss various issues regarding query optimization on the implementation level of the model.