We present multi-modal adversarial autoencoders for recommendation and evaluate them on two different tasks: citation recommendation and subject label recommendation. We analyze the effects of adversarial regularization, sparsity, and different input modalities. By conducting 408 experiments, we show that adversarial regularization consistently improves the performance of autoencoders for recommendation. We demonstrate, however, that the two tasks differ in the semantics of item co-occurrence in the sense that item co-occurrence resembles relatedness in case of citations, yet implies diversity in case of subject labels. Our results reveal that supplying the partial item set as input is only helpful, when item co-occurrence resembles relatedness. When facing a new recommendation task it is therefore crucial to consider the semantics of item co-occurrence for the choice of an appropriate model.

Recommender systems (RS) are software tools that use analytic technologies to suggest different items of interest to an end user. Linked Data is a set of best practices for publishing and connecting structured data on the Web. This paper presents a systematic literature review to summarize the state of the art in RS that use structured data published as Linked Data for providing recommendations of items from diverse domains. It considers the most relevant research problems addressed and classifies RS according to how Linked Data have been used to provide recommendations. Furthermore, it analyzes contributions, limitations, application domains, evaluation techniques, and directions proposed for future research. We found that there are still many open challenges with regard to RS based on Linked Data in order to be efficient for real applications. The main ones are personalization of recommendations, use of more datasets considering the heterogeneity introduced, creation of new hybrid RS for adding information, definition of more advanced similarity measures that take into account the large amount of data in Linked Data datasets, and implementation of testbeds to study evaluation techniques and to assess the accuracy scalability and computational complexity of RS.Use URI (uniform resource identifiers) as names for things. Use HTTP (Hypertext Transfer Protocol) URIs, so that people can look up those names. Use of standard mechanisms to provide useful information when someone looks up a URI, for example, RDF (Resource Description Framework) to represent data as graphs and SPARQL (SPARQL Protocol and RDF Query Language) to query Linked Data. Include links to other URIs, so that they can discover more things.The main benefit of using Linked Data as a source for generating recommendations is the large amount of available concepts and the relationships between them that can be used to infer relations more effectively in comparison to derive the same kind of relationships from text [12]. As Linked Data information is machine-readable, it is possible to query datasets on a fine-grained level in order to collect information without having to take manual actions; therefore, information is explicitly represented, which allows for applying reasoning techniques when querying datasets and making implicit knowledge explicit. Recommender systemsRS are software tools and techniques that provide suggestions of items to a user. These items can belong to different categories or types, for example, songs, places, news, books, films, and events. ‡ 4661 According to Adomavicius and Tuzhilin [4], the roots of RS can be traced back to the works in cognitive science, approximation theory, information retrieval, forecasting theories, management science, and consumer choice modeling in marketing.Nowadays, RS are focused on the recommendation problem of guiding users in a personalized way to interesting items in a large space of possible options [10]. Typically, RS are classified as content based, collaborative filtering, knowledge based, and ...

Background The number of studies applying machine learning (ML) to predict acute kidney injury (AKI) has grown steadily over the last decade. We assess and critically appraise the state of the art in ML models for AKI prediction considering performance, methodological soundness, and applicability. Methods We searched PubMed and ArXiv, extracted data, and critically appraised studies based on the TRIPOD, CHARMS, and PROBAST guidelines. Results Forty-six studies from 3166 titles were included. Thirty-eight studies developed a model, five developed and externally validated one, and three studies externally validated one. Flexible ML methods were used more often than deep learning, although the latter was common with temporal variables and text as predictors. Predictive performance showed an area under receiver operating curves ranging from 0.49 to 0.99. Our critical appraisal identified a high risk of bias in thirty-nine studies. Some studies lacked internal validation, whereas external validation and interpretability of results were rarely considered. Fifteen studies focused on AKI prediction in the intensive care setting, and the USA-derived MIMIC dataset was commonly used. Reproducibility was limited as data and code were usually unavailable. Conclusions Flexible ML methods are popular for the prediction of AKI, although more complex models based on deep learning are emerging. Our critical appraisal identified a high risk of bias in most models: Studies should use calibration measures and external validation more often, improve model interpretability, and share data and code to improve reproducibility.

No abstract