We introduce normalized nonnegative models (NNM) for explorative data analysis. NNMs are partial convexifications of models from probability theory. We demonstrate their value at the example of item recommendation. We show that NNM-based recommender systems satisfy three criteria that all recommender systems should ideally satisfy: high predictive power, computational tractability, and expressive representations of users and items. Expressive user and item representations are important in practice to succinctly summarize the pool of customers and the pool of items. In NNMs, user representations are expressive because each user's preference can be regarded as normalized mixture of preferences of stereotypical users. The interpretability of item and user representations allow us to arrange properties of items (e.g., genres of movies or topics of documents) or users (e.g., personality traits) hierarchically.

Recommendation systems have been integrated into the majority of large online systems to filter and rank information according to user profiles. It thus influences the way users interact with the system and, as a consequence, bias the evaluation of the performance of a recommendation algorithm computed using historical data (via offline evaluation). This paper describes this bias and discuss the relevance of a weighted offline evaluation to reduce this bias for different classes of recommendation algorithms.

Recommender systems learn about our preferences to make targeted suggestions. In this paper we outline a novel game-with-a-purpose designed to infer preferences at scale as a side-effect of gameplay. We evaluate the utility of this data in a recommendation context as part of a small live-user trial.

Today’s businesses, government and society work and services are centered around interactions, collaborations and knowledge work. The pace, amount and veracity of data generated and processed by a worker has accelerated significantly to the level that challenged human cognitive load and productivity. On the other hand, big data has provided an unprecedented opportunity for AI to tackle one of the main challenges hindering the AI progress: building models of world in a scalable, adaptive and dynamic manner. In this paper, we describe the technology requirements of building cognitive assistance technologies that assists human workers, and present a cognitive work assistant framework that aims at offering intelligence assistance to workers to improve their productivity and agility. We then describe the design and development of a set of cognitive services offered by the framework, based on advanced NLP and machine learning methods. The cognitive services help workers in processing and linking information and identifying and tracking work items over interactions in communication channels such as email, social conversations and media, chats and messaging and calendar applications. These cognitive services are designed to be adaptive, online and personalized so that over time adapt to changing environment and knowledge, and the models become personalized through learning preferences and working language and style of the subject worker.

Storytelling can be used by conversational agents in a wide variety of domains to maintain user engagement, both within a single interaction and over dozens or hun- dreds of interactions over time. The majority of agents designed with this ability to date deliver their stories as monologues without user input. However, people rarely tell stories in conversations this way, and instead rely on listener contributions to guide the storytelling process. Corpus-based studies of human-human conversational storytelling have demonstrated greater engagement, in the form of longer stories, when listeners co-construct stories this way. We describe a research framework for the generation and evaluation of co-constructed social stories in the context of task-based conversations, and a study on the effects of degree of user-agent story co-construction on user engagement. We find that users are more en- gaged with storytelling agents that allow them to co- construct stories in a contentful manner by asking ques- tions, compared to co-construction through acknowl- edgments only.

Critically, such dimensions are uncovered based on user feedback, often in implicit form (such as purchase histories, browsing logs, etc.); in addition, some recommender systems make use of side information, such as product attributes, temporal information, or review text. However one important feature that is typically ignored by existing personalized recommendation and ranking methods is the visual appearance of the items being considered. In this paper we propose a scalable factorization model to incorporate visual signals into predictors of people's opinions, which we apply to a selection of large, real-world datasets. We make use of visual features extracted from product images using (pre-trained) deep networks, on top of which we learn an additional layer that uncovers the visual dimensions that best explain the variation in people's feedback. This not only leads to significantly more accurate personalized ranking methods, but also helps to alleviate cold start issues, and qualitatively to analyze the visual dimensions that influence people's opinions.

Recommending items to users is a challenging task due to the large amount of missing information. In many cases, the data solely consist of ratings or tags voluntarily contributed by each user on a very limited subset of the available items, so that most of the data of potential interest is actually missing. Current approaches to recommendation usually assume that the unobserved data is missing at random. In this contribution, we provide statistical evidence that existing movie recommendation datasets reveal a significant positive association between the rating of items and the propensity to select these items. We propose a computationally efficient variational approach that makes it possible to exploit this selection bias so as to improve the estimation of ratings from small populations of users. Results obtained with this approach applied to neighborhood-based collaborative filtering illustrate its potential for improving the reliability of the recommendation.

This project continued to evolve and we explored the design space of a contextual software command recommender system and completed a six-week user study (Li et al. We then expanded the scope of our project by implementing CommunityCommands, a fully functional and deployable recommender system. During a one-year period, the recommender system was used by more than 1100 users. We also present our system usage data and payoff, and provide an in-depth discussion of the challenges and design issues associated with developing and deploying the software command recommender system.

Online dating is a prime application area for recommender systems, as users face an abundance of choice, must act on limited information, and are participating in a competitive matching market. The deployment was the result of thorough evaluation and an online trial of a number of methods, including profile-based, collaborative filtering and hybrid algorithms. Results taken a few months after deployment show that the recommender system delivered its projected benefits.

In this article, we report on a multiphase R&D effort to develop a conversational second screen application for TV program discovery. Our goal is to share with the community the breadth of artificial intelligence (AI) and natural language (NL) technologies required to develop such an application along with learnings from target end-users. We first give an overview of our application from the perspective of the end-user. We then present the architecture of our application along with the main AI and NL components, which were developed over multiple phases. The first phase focuses on enabling core functionality such as effectively finding programs matching the user’s intent. The second phase focuses on enabling dialog with the user. Finally, we present two user studies, corresponding to these two phases. The results from both studies demonstrate the effectiveness of our application in the target domain.