III: Small: Issues in the Management of GeoMultimedia Data

University Of Maryland, College Park, College Park MD

Investigators

Abstract

The growth of the Internet has led to a dramatic increase in the rate at which information is generated and delivered. This is especially true for the news cycle which has become instantaneous and often resulting in social networks, most notably Twitter, being the medium of choice to deliver late breaking news. This is usually in the form of links to more details which include news articles and multimedia data (news photos and videos). One problem lies in identifying reliable news gatherers which is done by noting if they are the first to report on a topic in contrast to being a re-poster and their frequency. Other issues in providing access to news content involve making the news automatically indexable (instead of by human tagging) both by content (regardless of the language of creation and the nature of the media) and by the location of the content (i.e., geotagging) rather than the location or affiliation of its creator. In the case of location, the challenge lies in devising language-independent geotagging techniques. Machine learning based methods are investigated and are expected to work better than rule-based methods due to a reduced reliance on language-specific rules and a greater reliance on examples. Access to the images is facilitated by indexing them by the words associated with the news articles containing them. This indexing technique is meant to be used as a filter for finding similar or near-duplicate images where the similarity is based on image features. The access by location is facilitated by the use of a map query interface. This is important as it corresponds to enabling the use of spatial synonyms thereby permitting a wider range of queries to be posed. A novel aspect of the research is the incorporation of non-English content which is facilitated by the use of computerized translation services which will be evaluated on their ability to capture the content on the basis of clustering similar articles in different languages rather than on the basis of factors such as proper grammar, etc. In today's rapidly changing world, the tools that are developed in this project will make this information more accessible as users are enabled to focus on a geographical area of interest as well as have access to content in their own language. This is of utility to a number of organizations and attempts will be made to collaborate with potential users on tailoring the tools for their needs. In addition, the project will provide research experience to undergraduate and graduate students who will be involved in developing some of the components. These tools are also a step in the growth of the nascent field of computational journalism. The project web site (http://www.cs.umd.edu/~hjs/geomultimedia.html ) will provide access to results of this and related research.