|FERRARI DACREMA MAURIZIO||Cycle: XXXII |
Section: Computer Science and Engineering
Tutor: AMIGONI FRANCESCO
Advisor: CREMONESI PAOLO Major Research topic
:Boosting content-based filtering with hybrid approachesAbstract:
Among the various algorithms available to implement a recommender system we may define two main categories, content-based and collaborative. Content and collaborative approaches differ in what kind of information they exploit to build the model: content-based use information specifically related to the item (author, genre, year, tags, extracted features from multimedia data, reviews…) whereas collaborative filtering relies on how other people behave with respect to the same items. It is known that collaborative filtering yields to better recommendation quality, except when there is not a sufficient number of user-item interactions, for example in case of new items (cold start). This scenario is especially relevant in fields such as news recommendation, e-commerce and many others, where new items are added often and in significant numbers. Content based filtering on the other hand is always applicable, provided that some information for the items is available. The performance is dependent on the quality of the available features, which poses a series of challenges related to their relevance with respect to the recommending task. Features are often noisy and redundant, therefore their selection or weighting is an important step when building a content based recommender. While a lot of research has been focused on integrating content information into collaborative algorithms to boost their quality in cold start scenarios, few have explored how to exploit collaborative information in a content-based algorithm. The purpose of this thesis is therefore to take this new route. The experiments show that the proposed approach is competitive in cold item scenarios on a variety of datasets, moreover embedding collaborative information in content-based algorithms provides a series of advantages and allows to leverage the domain knowledge and user behaviour that is implicit in a collaborative model.