Search

We present a new benchmark task for graph-based machine learning, aiming to predict future air quality (PM2.5 concentration) observed by a geographically distributed network of environmental sensors. While prior work has successfully applied Graph Neural Networks (GNNs) on a wide family of spatio-temporal prediction tasks, the new benchmark task introduced here brings a technical challenge that has been less studied in the context of graph-based spatio-temporal learning: distribution shift across a long period of time. An important goal of this paper is to understand the behavior of spatio-temporal GNNs under distribution shift. We conduct a comprehensive comparative study of both graph-based and non-graph-based machine learning models under two data split methods, one results in distribution shift and one does not. Our empirical results suggest that GNN models tend to suffer more from distribution shift compared to non-graph-based models, which calls for special attention when deploying spatio-temporal GNNs in practice.

BackgroundAn effective and scalable information retrieval (IR) system plays a crucial role in enabling clinicians and researchers to harness the valuable information present in electronic health records. In a previous study, we developed a prototype medical IR system, which incorporated a semantically based query recommendation (SBQR) feature. The system was evaluated empirically and demonstrated high perceived performance by end users. To delve deeper into the factors contributing to this perceived performance, we conducted a follow-up study using query log analysis. ObjectiveOne of the primary challenges faced in IR is that users often have limited knowledge regarding their specific information needs. Consequently, an IR system, particularly its user interface, needs to be thoughtfully designed to assist users through the iterative process of refining their queries as they encounter relevant documents during their search. To address these challenges, we incorporated “query recommendation” into our Electronic Medical Record Search Engine (EMERSE), drawing inspiration from the success of similar features in modern IR systems for general purposes. MethodsThe query log data analyzed in this study were collected during our previous experimental study, where we developed EMERSE with the SBQR feature. We implemented a logging mechanism to capture user query behaviors and the output of the IR system (retrieved documents). In this analysis, we compared the initial query entered by users with the query formulated with the assistance of the SBQR. By examining the results of this comparison, we could examine whether the use of SBQR helped in constructing improved queries that differed from the original ones. ResultsOur findings revealed that the first query entered without SBQR and the final query with SBQR assistance were highly similar (Jaccard similarity coefficient=0.77). This suggests that the perceived positive performance of the system was primarily attributed to the automatic query expansion facilitated by the SBQR rather than users manually manipulating their queries. In addition, through entropy analysis, we observed that search results converged in scenarios of moderate difficulty, and the degree of convergence correlated strongly with the perceived system performance. ConclusionsThe study demonstrated the potential contribution of the SBQR in shaping participants' positive perceptions of system performance, contingent upon the difficulty of the search scenario. Medical IR systems should therefore consider incorporating an SBQR as a user-controlled option or a semiautomated feature. Future work entails redesigning the experiment in a more controlled manner and conducting multisite studies to demonstrate the effectiveness of EMERSE with SBQR for patient cohort identification. By further exploring and validating these findings, we can enhance the usability and functionality of medical IR systems in real-world settings.

Abstract Background Natural language processing (NLP) tasks in the health domain often deal with limited amount of labeled data due to high annotation costs and naturally rare observations. To compensate for the lack of training data, health NLP researchers often have to leverage knowledge and resources external to a task at hand. Recently, pretrained large-scale language models such as the Bidirectional Encoder Representations from Transformers (BERT) have been proven to be a powerful way of learning rich linguistic knowledge from massive unlabeled text and transferring that knowledge to downstream tasks. However, previous downstream tasks often used training data at such a large scale that is unlikely to obtain in the health domain. In this work, we aim to study whether BERT can still benefit downstream tasks when training data are relatively small in the context of health NLP. Method We conducted a learning curve analysis to study the behavior of BERT and baseline models as training data size increases. We observed the classification performance of these models on two disease diagnosis data sets, where some diseases are naturally rare and have very limited observations (fewer than 2 out of 10,000). The baselines included commonly used text classification models such as sparse and dense bag-of-words models, long short-term memory networks, and their variants that leveraged external knowledge. To obtain learning curves, we incremented the amount of training examples per disease from small to large, and measured the classification performance in macro-averaged $$F_{1}$$ F 1 score. Results On the task of classifying all diseases, the learning curves of BERT were consistently above all baselines, significantly outperforming them across the spectrum of training data sizes. But under extreme situations where only one or two training documents per disease were available, BERT was outperformed by linear classifiers with carefully engineered bag-of-words features. Conclusion As long as the amount of training documents is not extremely few, fine-tuning a pretrained BERT model is a highly effective approach to health NLP tasks like disease classification. However, in extreme cases where each class has only one or two training documents and no more will be available, simple linear models using bag-of-words features shall be considered.

Emotions at work have long been identified as critical signals of work motivations, status, and attitudes, and as predictors of various work-related outcomes. When more and more employees work remotely, these emotional signals of workers become harder to observe through daily, face-to-face communications. The use of online platforms to communicate and collaborate at work provides an alternative channel to monitor the emotions of workers. This paper studies how emojis, as non-verbal cues in online communications, can be used for such purposes and how the emotional signals in emoji usage can be used to predict future behavior of workers. In particular, we present how the developers on GitHub use emojis in their work-related activities. We show that developers have diverse patterns of emoji usage, which can be related to their working status including activity levels, types of work, types of communications, time management, and other behavioral patterns. Developers who use emojis in their posts are significantly less likely to dropout from the online work platform. Surprisingly, solely using emoji usage as features, standard machine learning models can predict future dropouts of developers at a satisfactory accuracy. Features related to the general use and the emotions of emojis appear to be important factors, while they do not rule out paths through other purposes of emoji use.

Abstract Background Accurately recognizing rare diseases based on symptom description is an important task in patient triage, early risk stratification, and target therapies. However, due to the very nature of rare diseases, the lack of historical data poses a great challenge to machine learning-based approaches. On the other hand, medical knowledge in automatically constructed knowledge graphs (KGs) has the potential to compensate the lack of labeled training examples. This work aims to develop a rare disease classification algorithm that makes effective use of a knowledge graph, even when the graph is imperfect. Method We develop a text classification algorithm that represents a document as a combination of a “bag of words” and a “bag of knowledge terms,” where a “knowledge term” is a term shared between the document and the subgraph of KG relevant to the disease classification task. We use two Chinese disease diagnosis corpora to evaluate the algorithm. The first one, HaoDaiFu, contains 51,374 chief complaints categorized into 805 diseases. The second data set, ChinaRe, contains 86,663 patient descriptions categorized into 44 disease categories. Results On the two evaluation data sets, the proposed algorithm delivers robust performance and outperforms a wide range of baselines, including resampling, deep learning, and feature selection approaches. Both classification-based metric (macro-averaged F 1 score) and ranking-based metric (mean reciprocal rank) are used in evaluation. Conclusion Medical knowledge in large-scale knowledge graphs can be effectively leveraged to improve rare diseases classification models, even when the knowledge graph is incomplete.

Abstract Background Numbers and numerical concepts appear frequently in free text clinical notes from electronic health records. Knowledge of the frequent lexical variations of these numerical concepts, and their accurate identification, is important for many information extraction tasks. This paper describes an analysis of the variation in how numbers and numerical concepts are represented in clinical notes. Methods We used an inverted index of approximately 100 million notes to obtain the frequency of various permutations of numbers and numerical concepts, including the use of Roman numerals, numbers spelled as English words, and invalid dates, among others. Overall, twelve types of lexical variants were analyzed. Results We found substantial variation in how these concepts were represented in the notes, including multiple data quality issues. We also demonstrate that not considering these variations could have substantial real-world implications for cohort identification tasks, with one case missing > 80% of potential patients. Conclusions Numbering within clinical notes can be variable, and not taking these variations into account could result in missing or inaccurate information for natural language processing and information retrieval tasks.

Abstract Background Active learning (AL) has shown the promising potential to minimize the annotation cost while maximizing the performance in building statistical natural language processing (NLP) models. However, very few studies have investigated AL in a real-life setting in medical domain. Methods In this study, we developed the first AL-enabled annotation system for clinical named entity recognition (NER) with a novel AL algorithm. Besides the simulation study to evaluate the novel AL algorithm, we further conducted user studies with two nurses using this system to assess the performance of AL in real world annotation processes for building clinical NER models. Results The simulation results show that the novel AL algorithm outperformed traditional AL algorithm and random sampling. However, the user study tells a different story that AL methods did not always perform better than random sampling for different users. Conclusions We found that the increased information content of actively selected sentences is strongly offset by the increased time required to annotate them. Moreover, the annotation time was not considered in the querying algorithms. Our future work includes developing better AL algorithms with the estimation of annotation time and evaluating the system with larger number of users.