Citation Context

...l participants in a collaborative filtering community. As a result, the users whose ratings agree with some of the filterbots’ ratings would be able to receive better recommendations [40]. Similarly, =-=[65]-=- uses a collaborative approach where the traditional user’s ratings vector is augmented with additional ratings, which are calculated using a pure content-based predictor. 2.3.3 Adding Collaborative C...

Citation Context

...t, penalizing incorrect recommendations less severely the further down the list they occur. Penalty decreases with exponential decay. (2) (3) (4) 24 Other popular decision-support metrics include ROC =-=[28, 18, 9]-=-, the “receiver operating characteristic”. ROC measures the extent to which an information filtering system is able to successfully distinguish between signal and noise. Less frequently used, NDPM [2]...

Citation Context

...assume each classifier has expertise in a subset of the decision domain [23]–[25]. Similar issues in combining decisions from multiple raters or studies arise in content-based collaborative filterin=-=g [26]-=- and in meta-analysis of diagnostic tests [27], [28]. Estimating performance in the presence of an imperfect or limited reference standard has also been explored [29], [30]. The most appropriate vote ...

Citation Context

...go allowed users to get music recommendations online and connect with other music fans [12]. Fab, and other systems like it have helped users find web pages, news articles, and other documents online =-=[1, 4, 8]-=-. Our work builds on and extends our movie recommendation research service (movielens.umn.edu), that provides movie, DVD, and VHS video recommendations, along with a search capability. In addition, we...

Citation Context

...uding various user-specific filterbots (Claypool et al., 1999) propose a modular approach where independent predictions are computed by separate content filtering or collaborative filtering modules. (=-=Melville et al., 2002-=-) uses content-based predictors (naive Bayes) to impute missing values and create pseudo-user profiles. A second family of approaches treat user rating prediction as a machine learning problem, where ...

Citation Context

... the current three. Also, using collaborative user models to initialize individual models could speed up the learning process. A more explicit combination of collaborative and individual user models (=-=Melville, Mooney, & Nagarajan, 2002-=-; Jameson & Wittig, 2001) is also a viable direction to explore. 6.4 Evaluation Finally, we are planning to carry out a larger user study. We must further verify that the differences in our study were...

Citation Context

...sparse, two users or items are unlikely to have common ratings, and consequently, neighborhood-based approaches will predict ratings using a very limited number of neighbors. Moreover, similarity weights may be computed using only a small number of ratings, resulting in biased recommendations (see Section 3.2.3 for this problem). A common solution for these problems is to fill the missing ratings with default values [10, 18], such as the middle value of the rating range, and the average user or item rating. A more reliable approach is to use content information to fill out the missing ratings [16, 25, 41, 50]. For instance, the missing ratings can be provided by autonomous agents called filterbots [25, 41], that act as ordinary users of the system and rate items based on some specific characteristics of their content. The missing ratings can instead be predicted by a content-based approach [50], such as those described in Section 1.2.1. Finally, content similarity can also be used “instead of” or “in addition to” rating correlation similarity to find the nearest-neighbors employed in the predictions [3, 46, 59, 72]. These solutions, however, also have their own drawbacks. For instance, giving a de...