Вы находитесь на странице: 1из 2

PUBLICATION ARISING FROM MASTER DEGREE

Le Ba Dung
dungleba@gmail.com

1. Dung B. Le, Sunita Prasad, TS-LocalRank: A Topic Similarity Local Ranking Algorithm for Re-ranking Web Search Results, International Conference on Advanced Technologies for Communications, Hai Phong, Viet Nam, 2009.

DESCRIPTION OF THE RESEARCH UNDERTAKEN IN MASTER OF TECHNOLOGY


Le Ba Dung
dungleba@gmail.com

Motivation
Search engine has become very important for searching information in modern life. Most of search engines often return a very large number of web pages as relevant to user query. The most of users only look at the top pages. They will ignore a high relevant page if it is listed somewhere far from top pages. Re-ranking algorithm need to be used to improve the quality of top search results.

Problem Statement
Most of the popular web search engines (like Google, Yahoo!, and MSN) use link based ranking algorithms to rank web pages. These algorithms, called static ranking, assign a static rank to each document regardless of the query and the document content. There are some drawbacks of this approach: The number of pages that could be returned as relevant to user query is too large for users to digest because users often input a general search terms or short queries which relate to broad topic. Non-relevant page can appear in the top search results because link based ranking algorithms do not utilize content information of the web pages. A set of dissimilar web pages can be listed in top search results thus users have to spend much time to select a small subset which is most relevant to user query.

An alternative approach is to re-rank the search results for every query. This approach, called dynamic ranking, can be applied to a small set of top pages in order to give a better rank to the users. This research proposes a variant of PageRank algorithm to apply to re-ranking for web search results. This algorithm (named TS-LocalRank) obtains top N pages from search results of a major search engine, assigns a local rank to the web pages based on the topic similarity between the web pages, re-order the web pages based on the assigned local rank, and presents to users. The objectives of this algorithm are: Assigning a high local rank to web pages which are most relevant to user query. Minimize the Mean Absolute Deviation of the similarity between web pages in top search results.

Contributions
This research has proposed TS-LocalRank algorithm, a local ranking algorithm based on topic similarity between web pages. By applying TS-LocalRank algorithm, search engine can minimize the Mean Absolute Deviation of the similarity between web pages in top search results. The web pages which are most relevant to user query will be listed in top pages. The experience has been done by using a prototype of meta-search engine to re-rank the top search results returned by the most popular search engines (such as Google, Yahoo!, and MSN). In this experimental result, for top 30 pages returned by Yahoo! search engine with 50 popular search terms, the average MAD is about 39.75%. For top 30 pages returned by Google search engine with 50 popular search terms, the average MAD is about 47.02%.

Вам также может понравиться