Early recognition of risky trajectories during an Intensive Care Unit (ICU) stay is one of the key steps towards improving patient survival. Learning trajectories from physiological signals continuously measured during an ICU stay requires learning time-series features that are robust and discriminative across diverse patient populations. Patients within different ICU populations (referred here as domains) vary by age, conditions and interventions. Thus, mortality prediction models using patient data from a particular ICU population may perform suboptimally in other populations because the features used to train such models have different distributions across the groups. In this paper, we explore domain adaptation strategies in order to learn mortality prediction models that extract and transfer complex temporal features from multivariate time-series ICU data. Features are extracted in a way that the state of the patient in a certain time depends on the previous state. This enables dynamic predictions and creates a mortality risk space that describes the risk of a patient at a particular time. Experiments based on cross-ICU populations reveals that our model outperforms all considered baselines. Gains in terms of AUC range from 4% to 8% for early predictions when compared with a recent state-of-the-art representative for ICU mortality prediction. In particular, models for the Cardiac ICU population achieve AUC numbers as high as 0.88, showing excellent clinical utility for early mortality prediction. Finally, we present an explanation of factors contributing to the possible ICU outcomes, so that our models can be used to complement clinical reasoning.

/pdf/dynamic-prediction-of-icu-mortality-risk-using-domain-45j49thyls.pdf

Dynamic Prediction of ICU Mortality Risk Using Domain Adaptation

Minding other people's business: Community attachment and anticipated negative emotion in an extended norm activation model

The use of a simple categorization of emotions or even the use of universal expressions of emotions is unsuitable to properly identify sentiments in posts in some situations. The main goal of this paper is to analyze impressions of Twitter messages in the 19S Mexican earthquake of 2017 through machine learning techniques, specifically with classification algorithms. To identify impressions, we applied sentiment analysis based on supervised methods, and we identified a customized list of terms that we called impressions, which reflects the nature of tweets related to the event of study. Our proposed impressions analysis is useful to understand Twitter messages during different events since impressions adapt to each situation and context, based on emotional frameworks. We found that Twitter is useful to prove or disprove the information disseminated by the mass media and mainly for asking for help. Analyzing this kind of data in real-time will be useful for decision-making. The contribution of this paper is to fill the gap in the sentiment analysis area and the automatic identification of eleven impressions for disaster events in Twitter using machine learning techniques. This method has been called impression analysis.

Impression analysis of trending topics in Twitter with classification algorithms

Recurrent neural networks (RNNs) have shown outstanding performance for natural language processing tasks, influenced by the repeated multiplication of the recurrent weight matrices, the problem of gradient vanishing and explosion problem will be encountered when training RNN. Independently recurrent neural network (IndRNN) makes neurons independent and constrains recursive weights to effectively solve gradient problems. We combine IndRNN with long short-term memory (LSTM) and attention model, and test it for text classification, the results show that our models can effectively adapt to text classification task.

Text classification based recurrent neural network

Locations become places through personal significance and experience. While place data are not emotion data, per se, personal significance and experience are often emotional. In this paper, we explore the potential of using visual data exploration to support the qualitative analysis of place-related emotion data. To do so, we draw upon Creswell’s (2009) definition of place to define a generic data model that contains emotion data for a given location and its locale. For each data dimension in our model, we present symbolization options that can be combined to create a range of interactive visualizations, specifically supporting re-expression. We discuss the usefulness of example visualizations, created based on a data set from a pilot study on how elderly women experience their neighborhood. We find that the visualizations support four broad qualitative data analysis tasks: revising categorizations, making connections and relationships, aggregating for synthesis, and corroborating evidence by combining sense of place with locale information to support a holistic interpretation of place data. In conclusion, the paper contributes to the literature in three ways. It provides a generic data model and associated symbolization options, and uses examples to show how place-related emotion data can be visualized. Further, the example visualizations make explicit how re-expression, the combination of emotion data with locale information, and visualization of vagueness and linked data support the analysis of emotion data. Finally, we advocate for visualization-supported qualitative data analysis in interdisciplinary teams so that more suitable maps are used and so that cartographers can better understand and support qualitative data analysis.

/pdf/exploratory-geovisualizations-for-supporting-the-qualitative-1nvpyvlzg4.pdf

Christian Masdeval

Papers

Mining citizen emotions to estimate the urgency of urban issues