Representation learning (RL) has recently proven to be effective in capturing local item relationships by modeling item co-occurrence in individual user's interaction record. However, the value of RL for recommendation has not reached the full potential due to two major drawbacks: 1) recommendation is modeled as a rating prediction problem but should essentially be a personalized ranking one; 2) multi-level organizations of items are neglected for fine-grained item relationships. We design a unified Bayesian framework MRLR to learn user and item embeddings from a multi-level item organization, thus benefiting from RL as well as achieving the goal of personalized ranking. Extensive validation on real-world datasets shows that MRLR consistently outperforms state-of-the-art algorithms.

Workers of microtask crowdsourcing marketplaces strive to find a balance between the need for monetary income and the need for high reputation. Such balance is often threatened by poorly formulated tasks, as workers attempt their execution despite a sub-optimal understanding of the work to be done. In this paper we highlight the role of clarity as a characterising property of tasks in crowdsourcing. We surveyed 100 workers of the CrowdFlower platform to verify the presence of issues with task clarity in crowdsourcing marketplaces, reveal how crowd workers deal with such issues, and motivate the need for mechanisms that can predict and measure task clarity. Next, we propose a novel model for task clarity based on the goal and role clarity constructs. We sampled 7.1K tasks from the Amazon mTurk marketplace, and acquired labels for task clarity from crowd workers. We show that task clarity is coherently perceived by crowd workers, and is affected by the type of the task. We then propose a set of features to capture task clarity, and use the acquired labels to train and validate a supervised machine learning model for task clarity prediction. Finally, we perform a long-term analysis of the evolution of task clarity on Amazon mTurk, and show that clarity is not a property suitable for temporal characterisation.

Feature hierarchy (FH) has proven to be effective to improve recommendation accuracy. Prior work mainly focuses on the influence of vertically affiliated features (i.e. child-parent) on user-item interactions. The relationships of horizontally organized features (i.e. siblings and cousins) in the hierarchy, however, has only been little investigated. We show in real-world datasets that feature relationships in horizontal dimension can help explain and further model user-item interactions. To fully exploit FH, we propose a unified recommendation framework that seamlessly incorporates both vertical and horizontal dimensions for effective recommendation. Our model further considers two types of semanti-cally rich feature relationships in horizontal dimension, i.e. complementary and alternative relationships. Extensive validation on four real-world datasets demonstrates the superiority of our approach against the state of the art. An additional benefit of our model is to provide better interpretations of the generated recommendations.

Social media has emerged as one of the data backbones of urban analytics systems. Thanks to geo-located microposts (text-, image-, and video-based) created and shared through portals such as Twitter and Instagram, scientists and practitioners can capitalise on the availability of real-time and semantically rich data sources to perform studies related to cities and the people inhabiting them. Urban analytics systems usually consider the micro posts originating from within a city’s boundary uniformly, without consideration for the demographic (e.g. gender, age), geographic, technological or contextual (e.g. role in the city) differences among a platform’s users. It is well-known though, that the usage and adoption of social media profoundly differ across user segments, cities, as well as countries. We thus advocate for a better understanding of the intrinsic diversity of social media users and contents. This paper presents an observational study of the geo-located activities of users across two social media platforms, performed over a period of three weeks in four European cities. We show how demographic, geographical, technological and contextual properties of social media (and their users) can provide very different reflections and interpretations of the reality of an urban environment.

Thanks to reputation and gamification mechanisms, collaborative question answering systems coordinate the process of topical knowledge creation of thousands of users.While successful, these systems face many challenges: on one hand, the volume of submitted questions overgrows the amount of new users willing, and capable, of answering them. On the other hand, existing users need to be retained and optimally allocated. Previous work demonstrates the positive effects that two important aspects, namely engagement and expertise valorisation, can have on user quality and quantity of participation. The magnitude of their effect can greatly vary across users and across topics. In this paper we advocate for a more in-depth study of the interplay that exists between user engagement factors in question answering systems. Our working hypothesis is that the process of knowledge creation can be accelerated by better understanding and exploiting the combined effects of the interests and expertise of users, with their intrinsic and extrinsic motivations. We perform a study over 6 years of data from the StackOverflow platform. By defining metrics of expertise and (intrinsic and extrinsic) motivations, we show how they distribute and correlate across platform’s users and topics. By means of an off-line question routing experiment, we show how topic-specific combinations of motivations and expertise can help accelerating the knowledge creation process.