Abstract

Abstract
Modern online platforms for user modeling and recommendation require complex data infrastructures to collect and process data. Some of this data has to be kept to later be used in batches to train personalization models. However, since user activity data can be generated at very fast rates it is also useful to have algorithms able to process data streams online, in real time. Given the continuous and potentially fast change of content, context and user preferences or intents, stream-based models, and their synchronization with batch models can be extremely challenging. Therefore, it is important to investigate methods able to transparently and continuously adapt to the inherent dynamics of user interactions, preferably over long periods of time. Models able to continuously learn from such flows of data are gaining attention in the recommender systems community, and are being increasingly deployed in online platforms. However, many challenges associated with learning from streams need further investigation. The objective of this workshop is to foster contributions and bring together a growing community of researchers and practitioners interested in online, adaptive approaches to user modeling, recommendation and personalization, and their implications regarding multiple dimensions, such as reproducibility, privacy, fairness, diversity, transparency, auditability, and compliance with recently adopted or upcoming legal frameworks worldwide.

Abstract
Emerging evidence of the relationship between the microbiome composition and the development of numerous diseases, including cancer, has led to an increasing interest in the study of the human microbiome. Technological breakthroughs regarding DNA sequencing methods propelled microbiome studies with a large number of samples, which called for the necessity of more sophisticated data-analytical tools to analyze this complex relationship. The aim of this work was to develop a machine learning-based approach to distinguish the type of cancer based on the analysis of the tissue-specific microbial information, assessing the human microbiome as valuable predictive information for cancer identification. For this purpose, Random Forest algorithms were trained for the classification of five types of cancer-head and neck, esophageal, stomach, colon, and rectum cancers-with samples provided by The Cancer Microbiome Atlas database. One versus all and multi-class classification studies were conducted to evaluate the discriminative capability of the microbial data across increasing levels of cancer site specificity, with results showing a progressive rise in difficulty for accurate sample classification. Random Forest models achieved promising performances when predicting head and neck, stomach, and colon cancer cases, with the latter returning accuracy scores above 90% across the different studies conducted. However, there was also an increased difficulty when discriminating esophageal and rectum cancers, failing to differentiate with adequate results rectum from colon cancer cases, and esophageal from head and neck and stomach cancers. These results point to the fact that anatomically adjacent cancers can be more complex to identify due to microbial similarities. Despite the limitations, microbiome data analysis using machine learning may advance novel strategies to improve cancer detection and prevention, and decrease disease burden.

Abstract
When the price of the energy shared within an energy community is based on a local energy market, it is the responsibility of each participant to bid adequately so that participating provides a larger benefit than not participating. Alternatively, centralized energy community bill minimization may be an option, but a mechanism to share the collective benefits among the members is needed. This mechanism should be fair and easy to explain, no members should be harmed with respect to their individual optimal behavior and should provide the right economic signal. This paper analyses and compares some common pricing mechanisms for the internal compensation for the energy shared among the members of an energy community centrally managed. Simple case examples are used to identify those pricing mechanisms that are fairer and provide the righter economic signals to the participants.

Abstract
The branch-and-bound (B&B) procedure is one of the most widely used techniques to get optimal so-lutions for the resource-constrained project scheduling problem (RCPSP). Recently, various components from the literature have been assembled by Coelho and Vanhoucke (2018) into a unified search algo-rithm using the best performing lower bounds, branching schemes, search strategies, and dominance rules. However, due to the high computational time, this procedure is only suitable to solve small to medium-sized problems. Moreover, despite its relatively good performance, not much is known about which components perform best, and how these components should be combined into a procedure to maximize chances to solve the problem. This paper introduces a structured prediction approach to rank various combinations of components (configurations) of the integrated B&B procedure. More specifically, two regression methods are used to map project indicators to a full ranking of configurations. The objec-tive is to provide preference information about the quality of different configurations to obtain the best possible solution. Using such models, the ranking of all configurations can be predicted, and these predic-tions are then used to get the best possible solution for a new project with known network and resource values. A computational experiment is conducted to verify the performance of this novel approach. Fur-thermore, the models are tested for 48 different configurations, and their robustness is investigated on datasets with different numbers of activities. The results show that the two models are very competitive, and both can generate significantly better results than any single-best configuration.

Abstract
In everyday conversations, humans can take on different roles and adapt their vocabulary to their chosen roles. We explore whether LLMs can take on, that is impersonate, different roles when they generate text in-context. We ask LLMs to assume different personas before solving vision and language tasks. We do this by prefixing the prompt with a persona that is associated either with a social identity or domain expertise. In a multi-armed bandit task, we find that LLMs pretending to be children of different ages recover human-like developmental stages of exploration. In a language-based reasoning task, we find that LLMs impersonating domain experts perform better than LLMs impersonating non-domain experts. Finally, we test whether LLMs' impersonations are complementary to visual information when describing different categories. We find that impersonation can improve performance: an LLM prompted to be a bird expert describes birds better than one prompted to be a car expert. However, impersonation can also uncover LLMs' biases: an LLM prompted to be a man describes cars better than one prompted to be a woman. These findings demonstrate that LLMs are capable of taking on diverse roles and that this in-context impersonation can be used to uncover their strengths and hidden biases. Our code is available at https://github.com/ExplainableML/in-context-impersonation.