In this paper, we present agents based tool to discover new research topics from the information available on the World Wide Web (WWW). Agents are using KAROKA (Keywords Association Rides Optimizer Knobots Advisers). KAROKA is a model of discovery in text database used in WWW. The WWW sources are converted to a highly structured collection of text. Then, KAROKA tries to extract association rules, regularities and useful information in the collection of text. KAROKA techniques are described such as information retrieval similarity metrics for text, generation and pruning of keywords combination, and summary proposal of discovered information.