The challenges for this project are many: the notion of topic is informal and vague (making it complicated to determine when a page is about a topic); the Web is too large and unruly (making it impossible to apply in-depth analysis to all pages, or storing all results locally); most Web pages contain a mix of information and noise, like advertisements and menus (making it challenging to analyze the page contents). Most crawler use Information Retrieval techniques, which are too crude to determine the true content of a Web page.

• Designing and populating a thesaurus: we use a thesaurus in order to determine what is the scope of our topic and what concepts and relations are important on it. Regular thesauri are unsuited for the task since they express this semantic knowledge in a very superficial (hypernomy and meronymy) and imprecise (related-to) form.
• Scoring pages againts thesaurus: one a high precision thesaurus is available, one must still determine how to score a given web page against the theasaurus. In this context, having many words of overlap -even if each word appears once or few times- may be better than having a few words of overlap repeated many times. Thus, the traditional concepts of frequency (tf) and inverse document frequency (idf) do not apply well here.
• Dealing with noise in pages: most web pages contain advertisement, menus, links to other pages, and other material which is not properly part of their content. Cleaning such pages to have a clear score is a very difficult task.
• Extended keyword languages: because traditional keyword search is based on enumerating several keywords, it is difficult for a user to express his/her informational needs. We are looking into ways to extend keyword search by allowing users to enter more complex expressions than simple list of keywords.