REVISTA DYNA NEW TECHNOLOGIES

Key words:

motor de búsqueda, internet, visión artificial, recuperación de información, search engines, computer visión, information retrieval

Article type:

ARTICULO DE INVESTIGACION / RESEARCH ARTICLE

Section:

RESEARCH ARTICLES

Nowadays search engines are at the core of the internet and they are means to retrieve the most relevant information for the users. Actual search engines are focused on using the textual information of the HTML pages in order to find relevant documents given a user query. To this end, these systems compute the relevance of a webpage based on a textual analysis and other reputation elements (e.g. links). Despite the extreme complexity of current search engines, these do not consider the multimedia information present in webpages., although it has been demonstrated that the multimedia information is very relevant for the users.

In this work we propose to improve textual search engines using a novel scalable system that allows the search engine to aggregate textual information and visual information of the content of the images that are present in the documents. To this end, we propose a representation model and a system to rerank the search engines result lists.

In order to validate the propose approach, we show the results on exhaustive tests done at Web-scale using more than 20 million pages, several million images and more than 200 human evaluators that had judged manually the output lists of a base search engine and our approach.