February 10th, 2011

Information Retrieval and Data Mining

1. Mobile location-aware search. Consider now you are in London and want to find a restaurant nearby. Can your smartphone suggest one, based on your location and navigate you through the road? Can you also access and get a suggestion from other people who have already visited to the restaurant? Can Twitter help us find a solution from word-of-mouth?

2. Web Sentence Search. Have you experienced any difficulties while you are writing your English essays?
If yes, the Internet might help you. Tons of web pages from many online news paper websites provide most authoritative examples for English writings. But the problem is how to find them. More specifically, how to retrieve the examples of sentences or paragraphs that contain the similar usage of the term that you are looking for.  Warned about plagiarism though.

Economic models for online information access.  Financial models are advanced in modelling risk. In Information Retrieval research, we have identified the analogy between Web search markets and Stock Markets,  having developed  novel methods to handle retrieval risk and search result diversification. For more information, check out our recent studies. But this is just a start…

4. Financial data mining Online materials: Data mining, Teaching Financial Data MiningData Mining for Financial Applications.

5. Items (movies, music, books, news, web pages, etc.) that are likely of interest to a given user. As one of the common techniques, collaborative filtering makes recommendation on the basis of like-minded people. Popular systems include those offered by Amazon, Netflix, LastFM, etc. Although collaborative filtering is an effective way to alleviate information overload, it may be vulnerable to recommendation attacks since it solely relies on the item ratings from users.  Attackers, who intend to have their products (items) recommended more often, introduce biased ratings (preferences) to influence recommendation predictions. This project will explore various attack methods, and particularly study the effectiveness of these attacks towards different collaborative filtering algorithms. As a result, we expect to have a new way to evaluate and detect these attacks on recommender systems.

Comments are closed.