Most of the existing recommender systems use the ratings provided by users on individual items. An additional source of preference information is to use the ratings that users provide on sets of items. The advantages of using preferences on sets are two-fold. First, a rating provided on a set conveys some preference information about each of the set's items, which allows us to acquire a user's preferences for more items that the number of ratings that the user provided. Second, due to privacy concerns, users may not be willing to reveal their preferences on individual items explicitly but may be willing to provide a single rating to a set of items, since it provides some level of information hiding. This paper investigates two questions related to using set-level ratings in recommender systems. First, how users' item-level ratings relate to their set-level ratings. Second, how collaborative filtering-based models for item-level rating prediction can take advantage of such set-level ratings. We have collected set-level ratings from active users of Movielens on sets of movies that they have rated in the past. Our analysis of these ratings shows that though the majority of the users provide the average of the ratings on a set's constituent items as the rating on the set, there exists a significant number of users that tend to consistently either under- or over-rate the sets. We have developed collaborative filtering-based methods to explicitly model these user behaviors that can be used to recommend items to users. Experiments on real data and on synthetic data that resembles the under- or over-rating behavior in the real data, demonstrate that these models can recover the overall characteristics of the underlying data and predict the user's ratings on individual items.

Recommender systems face several challenges, e.g., recommending novel and diverse items and generating helpful explanations. Where algorithms struggle, people may excel. We therefore designed CrowdLens to explore different workflows for incorporating people into the recommendation process. We did an online experiment, finding that: compared to a state-of-the-art algorithm, crowdsourcing workflows produced more diverse and novel recommendations favored by human judges;some crowdworkers produced high-quality explanations for their recommendations, and we created an accurate model for identifying high-quality explanations;volunteers from an online community generally performed better than paid crowdworkers, but appropriate algorithmic support erased this gap. We conclude by reflecting on lessons of our work for those considering a crowdsourcing approach and identifying several fundamental issues for future work.

This paper builds upon several recent research efforts that have explored the nature and qualities of questions asked on these social Q&A sites by offering a focused examination of answers posted to three of the most popular Q&A sites. Specifically, this paper examines sets of answers responding to specific types of questions and explores the degree to which question types are predictive of answer quantity and answer quality. Blending qualitative and quantitative methods, the paper builds upon rich coding of a representative sets of real questions — drawn from Answerbag, (Ask) MetaFilter, and Yahoo! Answers — in order to better understand whether the explicit and implicit theories and predictions drawn from coding of these questions were borne out in the corresponding answer sets found on these sites. Quantitative findings include data underscoring the general overall success of social Q&A sites in producing answers that can satisfy the needs of those who pose questions. Additionally, this paper presents a predictive model that can anticipate the archival value of answers based on the category and qualities of questions asked. Qualitative findings include an analysis of the variation in responses to questions that are primarily seeking objective, grounded information relative to those seeking subjective opinions.