User Modeling and User-Adapted Interaction (UMUAI) provides an interdisciplinary forum for the dissemination of new research results on interactive computer systems that can be adapted or adapt themselves to their current users, and on the role of user models in the adaptation process.

UMUAI has been published since 1991 by Kluwer Academic Publishers (now merged with Springer Verlag).

UMUAI homepage with description of the scope of the journal and instructions for authors.

Springer UMUAI page with online access to the papers.

Latest Results for User Modeling and User-Adapted Interaction

02 April 2020

The latest content available from Springer
  • Session-aware news recommendations using random walks on time-evolving heterogeneous information networks


    Traditional news media Web sites usually provide generic recommendations that are not personalized to the preferences of their users. Typically, news recommendation algorithms mainly rely on the long-term preferences of users and do not adjust their model to the continuous stream of short-lived incoming stories to capture short-term intentions revealed by users’ sessions. In this paper, we therefore study the problem of session-aware recommendations by running random walks on dynamic heterogeneous graphs. Concretely, we construct a heterogeneous information network consisting of users, news articles, news categories, locations and sessions. By using different (1) sliding time window sizes, (2) sub-graphs for model learning, (3) sequential article weighting strategies and (4) more diversified random walks, we perform recommendations in a second step. Our algorithm proposal is evaluated on three real-life data sets, and we demonstrate that our method outperforms state-of-the-art methods by delivering more accurate and diversified recommendations.

  • Better targeting of consumers: Modeling multifactorial gender and biological sex from Instagram posts


    Along with the rapidly increasing influence and importance of advertisements and publicity in social networking services (SNS), considerable efforts are being made to provide user-customized services through an understanding of SNS content. Studies on online purchasing patterns based on user attributes have also been conducted; however, these studies used either only experimental methods (e.g., surveys or ethnographic accounts) or simple user attributes (e.g., age, biological sex, and location) for computational user modeling. This paper, through interviews with professional marketers, identifies their needs to understand multifactorial SNS user (potential customers) attributes—gender (i.e., masculine, feminine, androgynous) and biological sex (i.e., male and female) characteristics—for marketing purposes. Based on 33,752 Instagram posts, we develop a deep learning-based, classification model merged with three modalities—image (i.e., VGG16 feature and gesture), text (i.e., linguistic, tag, sentence, and category), and activity (i.e., reply and day). Our model achieves a better performance in classifying three gender types in the male, female, and male + female cases than the traditional machine learning models. Our study results reveal the applicability of identifying gender characteristics from posts in the marketing field.

  • Applying reranking strategies to route recommendation using sequence-aware evaluation


    Venue recommendation approaches have become particularly useful nowadays due to the increasing number of users registered in location-based social networks (LBSNs), applications where it is possible to share the venues someone has visited and establish connections with other users in the system. Besides, the venue recommendation problem has certain characteristics that differ from traditional recommendation, and it can also benefit from other contextual aspects to not only recommend independent venues, but complete routes or venue sequences of related locations. Hence, in this paper, we investigate the problem of route recommendation under the perspective of generating a sequence of meaningful locations for the users, by analyzing both their personal interests and the intrinsic relationships between the venues. We divide this problem into three stages, proposing general solutions to each case: First, we state a general methodology to derive user routes from LBSNs datasets that can be applied in as many scenarios as possible; second, we define a reranking framework that generate sequences of items from recommendation lists using different techniques; and third, we propose an evaluation metric that captures both accuracy and sequentiality at the same time. We report our experiments on several LBSNs datasets and by means of different recommendation quality metrics and algorithms. As a result, we have found that classical recommender systems are comparable to specifically tailored algorithms for this task, although exploiting the temporal dimension, in general, helps on improving the performance of these techniques; additionally, the proposed reranking strategies show promising results in terms of finding a trade-off between relevance, sequentiality, and distance, essential dimensions in both venue and route recommendation tasks.

  • Nonparametric user activity modelling and prediction


    Modelling the occupancy of buildings, rooms or the usage of machines has many applications in varying fields, exemplified by the fairly recent emergence of smart, self-learning thermostats. Typically, the aim of such systems is to provide insight into user behaviour and incentivise energy savings or to automatically reduce consumption while maintaining user comfort. This paper presents a nonparametric user activity modelling algorithm, i.e. a Dirichlet process mixture model implemented by Gibbs sampling and the stick-breaking process, to infer the underlying patterns in user behaviour from the data. The technique deals with multiple activities, such as <present, absent, sleeping>, of multiple users. Furthermore, it can also be used for modelling and predicting appliance usage (e.g. <on, standby, off>). The algorithm is evaluated, both on cluster validity and predictive performance, using three case studies of varying complexity. The obtained results indicate that the method is able to properly assign the activity data into well-defined clusters. Moreover, the high prediction accuracy demonstrates that these clusters can be exploited to anticipate future behaviour, facilitating the development of intelligent building management systems.

  • Learning to recommend similar items from human judgments


    Similar item recommendations—a common feature of many Web sites—point users to other interesting objects given a currently inspected item. A common way of computing such recommendations is to use a similarity function, which expresses how much alike two given objects are. Such similarity functions are usually designed based on the specifics of the given application domain. In this work, we explore how such functions can be learned from human judgments of similarities between objects, using two domains of “quality and taste”—cooking recipe and movie recommendation—as guiding scenarios. In our approach, we first collect a few thousand pairwise similarity assessments with the help of crowdworkers. Using these data, we then train different machine learning models that can be used as similarity functions to compare objects. Offline analyses reveal for both application domains that models that combine different types of item characteristics are the best predictors for human-perceived similarity. To further validate the usefulness of the learned models, we conducted additional user studies. In these studies, we exposed participants to similar item recommendations using a set of models that were trained with different feature subsets. The results showed that the combined models that exhibited the best offline prediction performance led to the highest user-perceived similarity, but also to recommendations that were considered useful by the participants, thus confirming the feasibility of our approach.