ArabAgent Project is my Master Research Project. I worked on it during my master period and I continued working on it 2 years after I earned the degree.
I develop a personalization system, called ArabAgent, to individualize the access into the Web i.e. through personalizing search and filtering news articles. The ArabAgent has the characteristics of an intelligent agent. It depends only on pure content-based approach to personalize the Web and it is mainly concerned with the Arabic content on the Web. ArabAgent is based on content to describe and represent their users‟ interests and knowledge. It uses documents submitted explicitly as relevant to extract the features that may interest the user. These features are any entities (such as objects, person names, place names, events, concepts, etc) that are mentioned through the text in each document. It uses such data to build a model to the user in the form of a time frame over a semantic network that keeps track of the entities that frequently occur in his documents and entities that frequently co-occur with each other within a 30-day time frame. If the entities didn‟t occur in the user‟s feedback documents, the user interest attenuates until it vanishes after 30 days.
In Evaluation, the average accuracy of the 30 user profiles of the 10 test days takes about 15 days to converge. We consider only an average of 25 relevant news articles (chosen randomly) of an average of 80 relevant articles. The convergence, definitely, takes less time if more than 25 relevant news articles are chosen and more time if less relevant articles are chosen.