By: Avanti Kumar, MIS Asia in InterGovWorld.com

A locally-developed search engine, that uses linguistic analytics instead of keywords, is among the first batch of projects under Singapore government’s Interactive Digital Media Research and Development Programme Office (IDMPO).The project, developed by researchers at the Lab for Media Search (LMS) from the National University of Singapore, uses a question-answering (Q&A) system that aims to provide users with precise answers, rather than a long list of Web pages like other search engines.

This year, the LMS Q&A search engine, will be joined by a new wave of 13 projects proposed by researchers from Singapore tertiary institutions. Some SG$18 million in funding will be provided.

Focusing on the research theme “Co-space”, the projects aim to make explore a new generation of the Internet that is more pervasive, immersive and integrated into people’s lives.

Up to 80 per cent accuracy

LMS Director, Professor Chua Tat-Seng said the Q&A search engine results offer 70 to 80 per cent accuracy, compared to the 20 per cent by typical search engines that use keywords search.

And, when a user keys in a question, instead of just displaying websites with related words, the application will display a list of exact answers. Each comes with the name of the source, date of publication and the Internet address.

Professor Chua said that such a system, with its ability to give answers directly, will do well in a knowledge management space within an enterprise.

Already, the system, together with another that is able to search for video clips on the Internet, has been adopted by US-based television cable company Comcast to help its customers search for programs more accurately. “For instance, the viewers can specify their search into fighting scenes and will be displayed with a search result of shows with fighting scenes,” said Professor Chua.

Mimicking human emotions

Other new IDMPO projects include one that mimics and simulates the sense of touch and other human emotions, to enrich the virtual experience, and another that addresses privacy and trust challenges in virtual worlds.

The latter measures and deduces the trust worthiness of users in the e-commerce space and makes the information available to interested parties. This is done through examination of past behaviors and others’ experiences and interactions with the particular user.

This project addresses the technology’s social aspect, which is often overlooked in many innovations, said Dr John Seely Brown, Visiting Scholar, Annenberg Center at University of Southern California. “This helps bring a level of trust to the Internet,” he said.

Dr Brown is part of IDMPO’s International Review Panel (IRP), which comprises of academics and industry luminaries in the field of new media. The IRP met in Singapore from 1 to 4 September to select the projects, apart from giving their recommendations, the judges felt that the new batch can trump the previous projects.

“I was very impressed to see such a progress since our last meeting in 2007,” said IRP member Professor Dr Jose Luis Encarnacao, Professor of Computer Science at the Technische Universitat Darmstadt and head of the Interactive Graphics Research Group. “The quality of the research is very high and the talent being developed shows a high-level of competence and motivation.”

Text Processing – Passage Retrieval and Question-Answering (QA)

We work mainly on question-answering from open domain (free-) text corpus, mainly in news. Depending on the type of questions, we develop techniques to return precise answers either at the phrase level, sentence level, or as query-focused summary.

Additionally, we are exploring techniques to answer deductive and exploratory questions. While extending the techniques for precise information retrieval, we are also applying them to vertical domains of education and legal search.
Question-Answering

Question-answering (QA) aims to find exact answers to users’ natural language queries, instead of ranked lists of documents as is done in current search engines. It is a major step towards information retrieval instead of document retrieval. Our QA system employs a pipeline structure that consists of several modules to get short and precise answers to users’ questions. It searches for answers at increasingly finer-grained units of: (1) locating the relevant documents, (2) retrieving passages that may contain the answer, and (3) pinpointing the exact answer from candidate passages. The main research focuses of our work are three-fold.

First we search the Web for relevance context information to supplement the often inexact query. In particular, we perform semantic clustering of information retrieved from the Web to induce different facets of queries in supporting event-based QA.

Second, we employ dependency relations, in additional to density-based word matching, to perform passage retrieval at sentence level. We will explore the integration of statistic, dependency and semantic relations in QA.

Third, we develop document concept lattice model together with definitional patterns and human interests model to perform task-oriented summarization.

Our studies on large-scale TREC-QA corpus demonstrate that our approaches are effective in performing factoid, list and definitional QA. Our system has been ranked consistently at second position in last 3 years (2003-2005) in public TREC-QA evaluations organized by NIST, USA. Our summarization system also achieved top position in DUC forum in 2005. Our technology has been licensed by industry to perform precise legal search.

This entry was posted on Tuesday, September 9th, 2008 at 10:46 pm and is filed under Global, News. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.