Search

Recent developments in large language models (LLM) and generative AI have unleashed the astonishing capabilities of text-to-image generation systems to synthesize high-quality images that are faithful to a given reference text, known as a "prompt". These systems have immediately received lots of attention from researchers, creators, and common users. Despite the plenty of efforts to improve the generative models, there is limited work on understanding the information needs of the users of these systems at scale. We conduct the first comprehensive analysis of large-scale prompt logs collected from multiple text-to-image generation systems. Our work is analogous to analyzing the query logs of Web search engines, a line of work that has made critical contributions to the glory of the Web search industry and research. Compared with Web search queries, text-to-image prompts are significantly longer, often organized into special structures that consist of the subject, form, and intent of the generation tasks and present unique categories of information needs. Users make more edits within creation sessions, which present remarkable exploratory patterns. There is also a considerable gap between the user-input prompts and the captions of the images included in the open training data of the generative models. Our findings provide concrete implications on how to improve text-to-image generation systems for creation purposes.

The gig economy provides workers with the benefits of autonomy and flexibility but at the expense of work identity and coworker bonds. Among the many reasons why gig workers leave their platforms, one unexplored aspect is the lack of an organization identity. In this study, we develop a team formation and interteam contest field experiment at a ride-sharing platform. We assign drivers to teams either randomly or based on similarity in age, hometown location, or productivity. Having these teams compete for cash prizes, we find that (1) compared with those in the control condition, treated drivers work longer hours and earn 12% higher revenue during the contest; (2) the treatment effect persists two weeks postcontest, albeit with half of the effect size; and (3) drivers in hometown-similar teams are more likely to communicate with each other, whereas those in age-similar teams continue to work longer hours and earn higher revenue during the two weeks after the contest ends. Together, our results show that platform designers can leverage team identity and team contests to increase revenue and worker engagement in a gig economy. This paper was accepted by David Simchi-Levi, behavioral economics and decision analysis. Funding: Financial support from the platform through the Michigan Institute for Data Science is gratefully acknowledged. Supplemental Material: The e-companion are data are available at https://doi.org/10.1287/mnsc.2022.4624 .

While the gig economy provides flexible jobs for millions of workers globally, a lack of organization identity and coworker bonds contributes to their low engagement and high attrition rates. To test the impact of virtual teams on worker productivity and retention, we conduct a field experiment with 27,790 drivers on a ride-sharing platform. We organize drivers into teams that are randomly assigned to receiving their team ranking, or individual ranking within their team, or individual performance information (control). We find that treated drivers work longer hours and generate significantly higher revenue. Furthermore, drivers in the team-ranking treatment continue to be more engaged 3 mo after the end of the experiment. A machine-learning analysis of 149 team contests in 86 cities suggests that social comparison, driver experience, and within-team similarity are the key predictors of the virtual team efficacy.

Social news aggregator services generate readers’ subjective reactions to news opinion articles. Can we use those as a resource to classify articles as liberal or conservative, even without knowing the self-identified political leaning of most users? We applied three semi-supervised learning methods that propagate classifications of political news articles and users as conservative or liberal, based on the assumption that liberal users will vote for liberal articles more often, and similarly for conservative users and articles. Starting from a few labeled articles and users, the algorithms propagate political leaning labels to the entire graph. In cross-validation, the best algorithm achieved 99.6% accuracy on held-out users and 96.3% accuracy on held-out articles. Adding social data such as users’ friendship or text features such as cosine similarity did not improve accuracy. The propagation algorithms, using the subjective liking data from users, also performed better than an SVM based text classifier, which achieved 92.0% accuracy on articles.

Existing tools for exploratory analysis of information diffusion in social media focus on the message senders who actively diffuse the meme. We develop a tool for audience analysis, focusing on the people who are passively exposed to the messages, with a special emphasis on competing memes such as propagations and corrections of a rumor. In such competing meme diffusions, important questions include which meme reached a bigger total audience, the overlap in audiences of the two, and whether exposure to one meme inhibited propagation of the other. We track audience members’ states of interaction, such as having been exposed to one meme or another or both. We analyze the marginal impact of each message in terms of the number of people who transition between states as a result of that message. These marginal impacts can be computed efficiently, even for diffusions involving thousands of senders and millions of receivers. The marginal impacts provide the raw material for an interactive tool, RumorLens, that includes a Sankey diagram and a network diagram. We validate the utility of the tool through a case study of nine rumor diffusions. We validate the usability of the tool through a user study, showing that nonexperts are able to use it to answer audience analysis questions.

Serendipity is a beneficial discovery that happens in an unexpected way. It has been found spectacularly valuable in various contexts, including scientific discoveries, acquisition of business, and recommender systems. Although never formally proved with large-scale behavioral analysis, it is believed by scientists and practitioners that serendipity is an important factor of positive user experience and increased user engagement. In this paper, we take the initiative to study the ubiquitous occurrence of serendipitious information diffusion and its effect in the context of microblogging communities. We refer to serendipity as unexpected relevance, then propose a principled statistical method to test the unexpectedness and the relevance of information received by a microblogging user, which identifies a serendipitous diffusion of information to the user. Our findings based on large-scale behavioral analysis reveal that there is a surprisingly strong presence of serendipitous information diffusion in retweeting, which accounts for more than 25% of retweets in both Twitter and Weibo. Upon the identification of serendipity, we are able to conduct observational analysis that reveals the benefit of serendipity to microblogging users. Results show that both the discovery and provision of serendipity increase the level of user activities and social interactions, while the provision of serendipitous information also increases the influence of Twitter users.

We present an experimental assessment of the impact of feature attribution-style explanations on human performance in predicting the consensus toxicity of social media posts with advice from an unreliable machine learning model. By doing so we add to a small but growing body of literature inspecting the utility of interpretable machine learning in terms of human outcomes. We also evaluate interpretable machine learning for the first time in the important domain of online toxicity, where fully-automated methods have faced criticism as being inadequate as a measure of toxic behavior.We find that, contrary to expectations, explanations have no significant impact on accuracy or agreement with model predictions, through they do change the distribution of subject error somewhat while reducing the cognitive burden of the task for subjects. Our results contribute to the recognition of an intriguing expectation gap in the field of interpretable machine learning between the general excitement the field has engendered and the ambiguous results of recent experimental work, including this study.

BackgroundNatural 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.MethodWe 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$$F_{1}$$F1score.ResultsOn 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.ConclusionAs 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.

Text simplification reduces the language complexity of professional content for accessibility purposes. End-to-end neural network models have been widely adopted to directly generate the simplified version of input text, usually functioning as a blackbox. We show that text simplification can be decomposed into a compact pipeline of tasks to ensure the transparency and explainability of the process. The first two steps in this pipeline are often neglected: 1) to predict whether a given piece of text needs to be simplified, and 2) if yes, to identify complex parts of the text. The two tasks can be solved separately using either lexical or deep learning methods, or solved jointly. Simply applying explainable complexity prediction as a preliminary step, the out-of-sample text simplification performance of the state-of-the-art, black-box simplification models can be improved by a large margin., ACL 2021

Millions of drivers worldwide have enjoyed financial benefits and work schedule flexibility through a ride-sharing economy, but meanwhile they have suffered from the lack of a sense of identity and career achievement. Equipped with social identity and contest theories, financially incentivized team competitions have been an effective instrument to increase drivers' productivity, job satisfaction, and retention, and to improve revenue over cost for ride-sharing platforms. While these competitions are overall effective, the decisive factors behind the treatment effects and how they affect the outcomes of individual drivers have been largely mysterious. In this study, we analyze data collected from more than 500 large-scale team competitions organized by a leading ride-sharing platform, building machine learning models to predict individual treatment effects. Through a careful investigation of features and predictors, we are able to reduce out-sample prediction error by more than 24%. Through interpreting the best-performing models, we discover many novel and actionable insights regarding how to optimize the design and the execution of team competitions on ride-sharing platforms. A simulated analysis demonstrates that by simply changing a few contest design options, the average treatment effect of a real competition is expected to increase by as much as 26%. Our procedure and findings shed light on how to analyze and optimize large-scale online field experiments in general., Accepted to KDD 2020

Sentiment classification typically relies on a large amount of labeled data. In practice, the availability of labels is highly imbalanced among different languages. To tackle this problem, cross-lingual sentiment classification approaches aim to transfer knowledge learned from one language that has abundant labeled examples (i.e., the source language, usually English) to another language with fewer labels (i.e., the target language). The source and the target languages are usually bridged through off-the-shelf machine translation tools. Through such a channel, cross-language sentiment patterns can be successfully learned from English and transferred into the target languages. This approach, however, often fails to capture sentiment knowledge specific to the target language. In this paper, we employ emojis, which are widely available in many languages, as a new channel to learn both the cross-language and the language-specific sentiment patterns. We propose a novel representation learning method that uses emoji prediction as an instrument to learn respective sentiment-aware representations for each language. The learned representations are then integrated to facilitate cross-lingual sentiment classification.

We describe our participation at the SemEval 2020 “Detection of Propaganda Techniques in News Articles” - Techniques Classification (TC) task, designed to categorize textual fragments into one of the 14 given propaganda techniques. Our solution leverages pre-trained BERT models. We present our model implementations, evaluation results and analysis of these results. We also investigate the potential of combining language models with resampling and ensemble learning methods to deal with data imbalance and improve performance.

The prevalence of smart mobile devices has promoted the popularity of mobile applications (a.k.a. apps). Supporting mobility has become a promising trend in software engineering research. This article presents an empirical study of behavioral service profiles collected from millions of users whose devices are deployed with Wandoujia, a leading Android app-store service in China. The dataset of Wandoujia service profiles consists of two kinds of user behavioral data from using 0.28 million free Android apps, including (1) app management activities (i.e., downloading, updating, and uninstalling apps) from over 17 million unique users and (2) app network usage from over 6 million unique users. We explore multiple aspects of such behavioral data and present patterns of app usage. Based on the findings as well as derived knowledge, we also suggest some new open opportunities and challenges that can be explored by the research community, including app development, deployment, delivery, revenue, etc.

ObjectiveMedical word sense disambiguation (WSD) is challenging and often requires significant training with data labeled by domain experts. This work aims to develop an interactive learning algorithm that makes efficient use of expert’s domain knowledge in building high-quality medical WSD models with minimal human effort.MethodsWe developed an interactive learning algorithm with expert labeling instances and features. An expert can provide supervision in 3 ways: labeling instances, specifying indicative words of a sense, and highlighting supporting evidence in a labeled instance. The algorithm learns from these labels and iteratively selects the most informative instances to ask for future labels. Our evaluation used 3 WSD corpora: 198 ambiguous terms from Medical Subject Headings (MSH) as MEDLINE indexing terms, 74 ambiguous abbreviations in clinical notes from the University of Minnesota (UMN), and 24 ambiguous abbreviations in clinical notes from Vanderbilt University Hospital (VUH). For each ambiguous term and each learning algorithm, a learning curve that plots the accuracy on the test set against the number of labeled instances was generated. The area under the learning curve was used as the primary evaluation metric.ResultsOur interactive learning algorithm significantly outperformed active learning, the previous fastest learning algorithm for medical WSD. Compared to active learning, it achieved 90% accuracy for the MSH corpus with 42% less labeling effort, 35% less labeling effort for the UMN corpus, and 16% less labeling effort for the VUH corpus.ConclusionsHigh-quality WSD models can be efficiently trained with minimal supervision by inviting experts to label informative instances and provide domain knowledge through labeling/highlighting contextual features.

App marketplaces host millions of mobile apps that are downloaded billions of times. Investigating how people manage mobile apps in their everyday lives creates a unique opportunity to understand the behavior and preferences of mobile device users, infer the quality of apps, and improve user experience. Existing literature provides very limited knowledge about app management activities, due to the lack of app usage data at scale. This article takes the initiative to analyze a very large app management log collected through a leading Android app marketplace. The dataset covers 5 months of detailed downloading, updating, and uninstallation activities, which involve 17 million anonymized users and 1 million apps. We present a surprising finding that the metrics commonly used to rank apps in app stores do not truly reflect the users’ real attitudes. We then identify behavioral patterns from the app management activities that more accurately indicate user preferences of an app even when no explicit rating is available. A systematic statistical analysis is designed to evaluate machine learning models that are trained to predict user preferences using these behavioral patterns, which features an inverse probability weighting method to correct the selection biases in the training process.

Display Omitted A user-centered evaluation is conducted to assess the value of query recommendation.The feature is designed to facilitate retrieval of information from EHRs.The algorithm utilizes MetaMap to identify medical concepts.The performance is rated consistently higher with query recommendation turned on.Perceived usefulness and perceived ease of use scores are overwhelmingly positive. ObjectiveThe utility of biomedical information retrieval environments can be severely limited when users lack expertise in constructing effective search queries. To address this issue, we developed a computer-based query recommendation algorithm that suggests semantically interchangeable terms based on an initial user-entered query. In this study, we assessed the value of this approach, which has broad applicability in biomedical information retrieval, by demonstrating its application as part of a search engine that facilitates retrieval of information from electronic health records (EHRs). Materials and MethodsThe query recommendation algorithm utilizes MetaMap to identify medical concepts from search queries and indexed EHR documents. Synonym variants from UMLS are used to expand the concepts along with a synonym set curated from historical EHR search logs. The empirical study involved 33 clinicians and staff who evaluated the system through a set of simulated EHR search tasks. User acceptance was assessed using the widely used technology acceptance model. ResultsThe search engines performance was rated consistently higher with the query recommendation feature turned on vs. off. The relevance of computer-recommended search terms was also rated high, and in most cases the participants had not thought of these terms on their own. The questions on perceived usefulness and perceived ease of use received overwhelmingly positive responses. A vast majority of the participants wanted the query recommendation feature to be available to assist in their day-to-day EHR search tasks. Discussion and ConclusionChallenges persist for users to construct effective search queries when retrieving information from biomedical documents including those from EHRs. This study demonstrates that semantically-based query recommendation is a viable solution to addressing this challenge.

We investigate the effects of team competition on pro-social lending activity on Kiva.org , the first microlending website to match lenders with entrepreneurs in developing countries. Using naturally occurring field data, we find that lenders who join teams contribute 1.2 more loans ($30–$42) per month than those who do not. To further explore factors that differentiate successful teams from dormant ones, we run a large-scale randomized field experiment ( n = 22 , 233 ) by posting forum messages. Compared to the control, we find that lenders make significantly more loans when exposed to a goal-setting and coordination message, whereas goal-setting alone significantly increases lending activities of previously inactive teams. Our findings suggest that goal-setting and coordination are effective mechanisms to increase pro-social behavior in teams.

Objective Active Learning (AL) attempts to reduce annotation cost (ie, time) by selecting the most informative examples for annotation. Most approaches tacitly (and unrealistically) assume that the cost for annotating each sample is identical. This study introduces a cost-aware AL method, which simultaneously models both the annotation cost and the informativeness of the samples and evaluates both via simulation and user studies. Materials and Methods We designed a novel, cost-aware AL algorithm (Cost-CAUSE) for annotating clinical named entities; we first utilized lexical and syntactic features to estimate annotation cost, then we incorporated this cost measure into an existing AL algorithm. Using the 2010 i2b2/VA data set, we then conducted a simulation study comparing Cost-CAUSE with noncost-aware AL methods, and a user study comparing Cost-CAUSE with passive learning. Results Our cost model fit empirical annotation data well, and Cost-CAUSE increased the simulation area under the learning curve (ALC) scores by up to 5.6% and 4.9%, compared with random sampling and alternate AL methods. Moreover, in a user annotation task, Cost-CAUSE outperformed passive learning on the ALC score and reduced annotation time by 20.5%–30.2%. Discussion Although AL has proven effective in simulations, our user study shows that a real-world environment is far more complex. Other factors have a noticeable effect on the AL method, such as the annotation accuracy of users, the tiredness of users, and even the physical and mental condition of users. Conclusion Cost-CAUSE saves significant annotation cost compared to random sampling.

Sentiment analysis has various application scenarios in software engineering (SE), such as detecting developers' emotions in commit messages and identifying their opinions on Q&A forums. However, commonly used out-of-the-box sentiment analysis tools cannot obtain reliable results on SE tasks and the misunderstanding of technical jargon is demonstrated to be the main reason. Then, researchers have to utilize labeled SE-related texts to customize sentiment analysis for SE tasks via a variety of algorithms. However, the scarce labeled data can cover only very limited expressions and thus cannot guarantee the analysis quality. To address such a problem, we turn to the easily available emoji usage data for help. More specifically, we employ emotional emojis as noisy labels of sentiments and propose a representation learning approach that uses both Tweets and GitHub posts containing emojis to learn sentiment-aware representations for SE-related texts. These emoji-labeled posts can not only supply the technical jargon, but also incorporate more general sentiment patterns shared across domains. They as well as labeled data are used to learn the final sentiment classifier. Compared to the existing sentiment analysis methods used in SE, the proposed approach can achieve significant improvement on representative benchmark datasets. By further contrast experiments, we find that the Tweets make a key contribution to the power of our approach. This finding informs future research not to unilaterally pursue the domain-specific resource, but try to transform knowledge from the open domain through ubiquitous signals such as emojis., Accepted by the 2019 ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE 2019). Please include ESEC/FSE in any citations

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.

We conduct a large-scale, systematic study to evaluate the existing evaluation methods for natural language generation in the context of generating online product reviews. We compare human-based evaluators with a variety of automated evaluation procedures, including discriminative evaluators that measure how well machine-generated text can be distinguished from human-written text, as well as word overlap metrics that assess how similar the generated text compares to human-written references. We determine to what extent these different evaluators agree on the ranking of a dozen of state-of-the-art generators for online product reviews. We find that human evaluators do not correlate well with discriminative evaluators, leaving a bigger question of whether adversarial accuracy is the correct objective for natural language generation. In general, distinguishing machine-generated text is challenging even for human evaluators, and human decisions correlate better with lexical overlaps. We find lexical diversity an intriguing metric that is indicative of the assessments of different evaluators. A post-experiment survey of participants provides insights into how to evaluate and improve the quality of natural language generation systems.

Significance With three billion people subsisting on the equivalent of $2.50 per day, alleviating poverty is one of the most urgent challenges facing the world today. One solution to this problem has been to encourage the growth of small enterprises through microlending. A successful innovation is represented by Kiva.org , which matches citizen lenders with low-income entrepreneurs in developing countries. To increase prosocial lending, we use a large-scale field experiment and machine-learning methods to recommend lending teams to lenders. We find that lenders who join a team contribute significantly more compared with those who do not. Our results suggest team recommendation can be an effective and low-cost behavioral mechanism to increase charitable contributions.

We introduce an adversarial method for producing high-recall explanations of neural text classifier decisions. Building on an existing architecture for extractive explanations via hard attention, we add an adversarial layer which scans the residual of the attention for remaining predictive signal. Motivated by the important domain of detecting personal attacks in social media comments, we additionally demonstrate the importance of manually setting a semantically appropriate `default' behavior for the model by explicitly manipulating its bias term. We develop a validation set of human-annotated personal attacks to evaluate the impact of these changes., Accepted to EMNLP 2018 Code and data available at https://github.com/shcarton/rcnn

Much work has been done recently on learning word embeddings from large corpora, which attempts to find the coordinates of words in a static and high dimensional semantic space. In reality, such corpora often span a sufficiently long time period, during which the meanings of many words may have changed. The co-evolution of word meanings may also result in a distortion of the semantic space, making these static embeddings unable to accurately represent the dynamics of semantics. In this paper, we present a novel computational method to capture such changes and to model the evolution of word semantics. Distinct from existing approaches that learn word embeddings independently from time periods and then align them, our method explicitly establishes the stable topological structure of word semantics and identifies the surprising changes in the semantic space over time through a principled statistical method. Empirical experiments on large-scale real-world corpora demonstrate the effectiveness of the proposed approach, which outperforms the state-of-the-art by a large margin.

A critical research problem about information cascades, which is a central topic of social network analysis, is to predict the potential influence or the future growth of cascades. Recent developments of deep learning have provided promising alternatives, which no longer rely on heavy feature engineering efforts and instead learn the representation of cascade graphs in an end-to-end manner. In reality, however, the influence of a cascade not only depends on the cascade graph and the global network structure, but also largely relies on the content of the cascade and the preferences of users. In this work, we extend the deep learning approaches to cascade prediction by jointly modeling the content and the structure of cascades. We find that text information provides a valuable addition for the learning of cascade graphs, especially when some users (nodes) have rarely participated in the past cascades. To this end, a gating mechanism is introduced to dynamically fuse the structural and textual representations of nodes based on their respective properties. Attentions are employed to incorporate the text information associated with both cascade items and nodes. Empirical experiments demonstrate that incorporating text information brings a significant improvement to cascade prediction, and that the proposed model outperforms alternative ways to combine text and networks.

Display Omitted We developed novel active learning algorithms for clinical named entity recognition.Equal cost per sample is not a practical annotation cost assumption in this task.We evaluated methods based on two types of estimated annotation cost.To achieve 0.8 in F-measure, active learning could save 42% annotation cost in words.The actual benefit of active learning should be further evaluated in real time. ObjectivesNamed entity recognition (NER), a sequential labeling task, is one of the fundamental tasks for building clinical natural language processing (NLP) systems. Machine learning (ML) based approaches can achieve good performance, but they often require large amounts of annotated samples, which are expensive to build due to the requirement of domain experts in annotation. Active learning (AL), a sample selection approach integrated with supervised ML, aims to minimize the annotation cost while maximizing the performance of ML-based models. In this study, our goal was to develop and evaluate both existing and new AL methods for a clinical NER task to identify concepts of medical problems, treatments, and lab tests from the clinical notes. MethodsUsing the annotated NER corpus from the 2010 i2b2/VA NLP challenge that contained 349 clinical documents with 20,423 unique sentences, we simulated AL experiments using a number of existing and novel algorithms in three different categories including uncertainty-based, diversity-based, and baseline sampling strategies. They were compared with the passive learning that uses random sampling. Learning curves that plot performance of the NER model against the estimated annotation cost (based on number of sentences or words in the training set) were generated to evaluate different active learning and the passive learning methods and the area under the learning curve (ALC) score was computed. ResultsBased on the learning curves of F-measure vs. number of sentences, uncertainty sampling algorithms outperformed all other methods in ALC. Most diversity-based methods also performed better than random sampling in ALC. To achieve an F-measure of 0.80, the best method based on uncertainty sampling could save 66% annotations in sentences, as compared to random sampling. For the learning curves of F-measure vs. number of words, uncertainty sampling methods again outperformed all other methods in ALC. To achieve 0.80 in F-measure, in comparison to random sampling, the best uncertainty based method saved 42% annotations in words. But the best diversity based method reduced only 7% annotation effort. ConclusionIn the simulated setting, AL methods, particularly uncertainty-sampling based approaches, seemed to significantly save annotation cost for the clinical NER task. The actual benefit of active learning in clinical NER should be further evaluated in a real-time setting.

Objective ClinicalTrials.gov serves critical functions of disseminating trial information to the public and helping the trials recruit participants. This study assessed the readability of trial descriptions at ClinicalTrials.gov using multiple quantitative measures.Materials and Methods The analysis included all 165 988 trials registered at ClinicalTrials.gov as of April 30, 2014. To obtain benchmarks, the authors also analyzed 2 other medical corpora: (1) all 955 Health Topics articles from MedlinePlus and (2) a random sample of 100 000 clinician notes retrieved from an electronic health records system intended for conveying internal communication among medical professionals. The authors characterized each of the corpora using 4 surface metrics, and then applied 5 different scoring algorithms to assess their readability. The authors hypothesized that clinician notes would be most difficult to read, followed by trial descriptions and MedlinePlus Health Topics articles.Results Trial descriptions have the longest average sentence length (26.1 words) across all corpora; 65% of their words used are not covered by a basic medical English dictionary. In comparison, average sentence length of MedlinePlus Health Topics articles is 61% shorter, vocabulary size is 95% smaller, and dictionary coverage is 46% higher. All 5 scoring algorithms consistently rated CliniclTrials.gov trial descriptions the most difficult corpus to read, even harder than clinician notes. On average, it requires 18 years of education to properly understand these trial descriptions according to the results generated by the readability assessment algorithms.Discussion and Conclusion Trial descriptions at CliniclTrials.gov are extremely difficult to read. Significant work is warranted to improve their readability in order to achieve CliniclTrials.gov’s goal of facilitating information dissemination and subject recruitment.

OBJECTIVE: Traditional Chinese medicine (TCM) is a unique and complex medical system that has developed over thousands of years. This article studies the problem of automatically extracting meaningful relations of entities from TCM literature, for the purposes of assisting clinical treatment or poly-pharmacology research and promoting the understanding of TCM in Western countries. METHODS: Instead of separately extracting each relation from a single sentence or document, we propose to collectively and globally extract multiple types of relations (eg, herb-syndrome, herb-disease, formula-syndrome, formula-disease, and syndrome-disease relations) from the entire corpus of TCM literature, from the perspective of network mining. In our analysis, we first constructed heterogeneous entity networks from the TCM literature, in which each edge is a candidate relation, then used a heterogeneous factor graph model (HFGM) to simultaneously infer the existence of all the edges. We also employed a semi-supervised learning algorithm estimate the model's parameters. RESULTS: We performed our method to extract relations from a large dataset consisting of more than 100,000 TCM article abstracts. Our results show that the performance of the HFGM at extracting all types of relations from TCM literature was significantly better than a traditional support vector machine (SVM) classifier (increasing the average precision by 11.09%, the recall by 13.83%, and the F1-measure by 12.47% for different types of relations, compared with a traditional SVM classifier). CONCLUSION: This study exploits the power of collective inference and proposes an HFGM based on heterogeneous entity networks, which significantly improved our ability to extract relations from TCM literature. ispartof: Journal of the American Medical Informatics Association vol:23 issue:2 pages:356-365 ispartof: location:England status: published

Information cascades, effectively facilitated by most social network platforms, are recognized as a major factor in almost every social success and disaster in these networks. Can cascades be predicted? While many believe that they are inherently unpredictable, recent work has shown that some key properties of information cascades, such as size, growth, and shape, can be predicted by a machine learning algorithm that combines many features. These predictors all depend on a bag of hand-crafting features to represent the cascade network and the global network structure. Such features, always carefully and sometimes mysteriously designed, are not easy to extend or to generalize to a different platform or domain. Inspired by the recent successes of deep learning in multiple data mining tasks, we investigate whether an end-to-end deep learning approach could effectively predict the future size of cascades. Such a method automatically learns the representation of individual cascade graphs in the context of the global network structure, without hand-crafted features and heuristics. We find that node embeddings fall short of predictive power, and it is critical to learn the representation of a cascade graph as a whole. We present algorithms that learn the representation of cascade graphs in an end-to-end manner, which significantly improve the performance of cascade prediction over strong baselines that include feature based methods, node embedding methods, and graph kernel methods. Our results also provide interesting implications for cascade prediction in general.

We consider the task of identifying attitudes towards a given set of entities from text. Conventionally, this task is decomposed into two separate subtasks: target detection that identifies whether each entity is mentioned in the text, either explicitly or implicitly, and polarity classification that classifies the exact sentiment towards an identified entity (the target) into positive, negative, or neutral. Instead, we show that attitude identification can be solved with an end-to-end machine learning architecture, in which the two subtasks are interleaved by a deep memory network. In this way, signals produced in target detection provide clues for polarity classification, and reversely, the predicted polarity provides feedback to the identification of targets. Moreover, the treatments for the set of targets also influence each other -- the learned representations may share the same semantics for some targets but vary for others. The proposed deep memory network, the AttNet, outperforms methods that do not consider the interactions between the subtasks or those among the targets, including conventional machine learning methods and the state-of-the-art deep learning models., Accepted to WSDM'17

Based on a large data set of emoji using behavior collected from smartphone users over the world, this paper investigates gender-specific usage of emojis. We present various interesting findings that evidence a considerable difference in emoji usage by female and male users. Such a difference is significant not just in a statistical sense; it is sufficient for a machine learning algorithm to accurately infer the gender of a user purely based on the emojis used in their messages. In real world scenarios where gender inference is a necessity, models based on emojis have unique advantages over existing models that are based on textual or contextual information. Emojis not only provide language-independent indicators, but also alleviate the risk of leaking private user information through the analysis of text and metadata., Comment: The Web Conference 2018 (WWW 2018)

Objectives This study was to assess whether active learning strategies can be integrated with supervised word sense disambiguation (WSD) methods, thus reducing the number of annotated samples, while keeping or improving the quality of disambiguation models. Methods We developed support vector machine (SVM) classifiers to disambiguate 197 ambiguous terms and abbreviations in the MSH WSD collection. Three different uncertainty sampling-based active learning algorithms were implemented with the SVM classifiers and were compared with a passive learner (PL) based on random sampling. For each ambiguous term and each learning algorithm, a learning curve that plots the accuracy computed from the test set as a function of the number of annotated samples used in the model was generated. The area under the learning curve (ALC) was used as the primary metric for evaluation. Results Our experiments demonstrated that active learners (ALs) significantly outperformed the PL, showing better performance for 177 out of 197 (89.8%) WSD tasks. Further analysis showed that to achieve an average accuracy of 90%, the PL needed 38 annotated samples, while the ALs needed only 24, a 37% reduction in annotation effort. Moreover, we analyzed cases where active learning algorithms did not achieve superior performance and identified three causes: (1) poor models in the early learning stage; (2) easy WSD cases; and (3) difficult WSD cases, which provide useful insight for future improvements. Conclusions This study demonstrated that integrating active learning strategies with supervised WSD methods could effectively reduce annotation cost and improve the disambiguation models.

We study the problem of visualizing large-scale and high-dimensional data in a low-dimensional (typically 2D or 3D) space. Much success has been reported recently by techniques that first compute a similarity structure of the data points and then project them into a low-dimensional space with the structure preserved. These two steps suffer from considerable computational costs, preventing the state-of-the-art methods such as the t-SNE from scaling to large-scale and high-dimensional data (e.g., millions of data points and hundreds of dimensions). We propose the LargeVis, a technique that first constructs an accurately approximated K-nearest neighbor graph from the data and then layouts the graph in the low-dimensional space. Comparing to t-SNE, LargeVis significantly reduces the computational cost of the graph construction step and employs a principled probabilistic model for the visualization step, the objective of which can be effectively optimized through asynchronous stochastic gradient descent with a linear time complexity. The whole procedure thus easily scales to millions of high-dimensional data points. Experimental results on real-world data sets demonstrate that the LargeVis outperforms the state-of-the-art methods in both efficiency and effectiveness. The hyper-parameters of LargeVis are also much more stable over different data sets., WWW 2016

With the rapid decrease in cost of genome sequencing, the classification of gene function is becoming a primary problem. Such classification has been performed by human curators who read biological literature to extract evidence. BeeSpace Navigator is a prototype software for exploratory analysis of gene function using biological literature. The software supports an automatic analogue of the curator process to extract functions, with a simple interface intended for all biologists. Since extraction is done on selected collections that are semantically indexed into conceptual spaces, the curation can be task specific. Biological literature containing references to gene lists from expression experiments can be analyzed to extract concepts that are computational equivalents of a classification such as Gene Ontology, yielding discriminating concepts that differentiate gene mentions from other mentions. The functions of individual genes can be summarized from sentences in biological literature, to produce results resembling a model organism database entry that is automatically computed. Statistical frequency analysis based on literature phrase extraction generates offline semantic indexes to support these gene function services. The website with BeeSpace Navigator is free and open to all; there is no login requirement at www.beespace.illinois.edu for version 4. Materials from the 2010 BeeSpace Software Training Workshop are available at www.beespace.illinois.edu/bstwmaterials.php.

he prevalence of smart devices has promoted the popular- ity of mobile applications (a.k.a. apps) in recent years. A number of interesting and important questions remain unan- swered, such as why a user likes/dislikes an app, how an app becomes popular or eventually perishes, how a user selects apps to install and interacts with them, how frequently an app is used and how much traffic it generates, etc. This paper presents an empirical analysis of app usage behaviors collected from millions of users of Wandoujia, a leading An- droid app marketplace in China. The dataset covers two types of user behaviors of using over 0.2 million Android apps, including (1) app management activities (i.e., installa- tion, updating, and uninstallation) of over 0.8 million unique users and (2) app network traffic from over 2 million unique users. We explore multiple aspects of such behavior data and present interesting patterns of app usage. The results provide many useful implications to the developers, users, and disseminators of mobile apps.

We present the problem of click-through prediction for advertising in Twitter timeline, which displays a stream of Tweets from accounts a user choose to follow. Traditional computational advertising usually appears in two forms: sponsored search that places ads onto the search result page when a query is issued to a search engine, and contextual advertising that places ads onto a regular, usually static Web page. Compared with these two paradigms, placing ads into a Tweet stream is particularly challenging given the nature of the data stream: the context into which an ad can be placed updates dynamically and never replicates. Every ad is therefore placed into a unique context. This makes the information available for training a machine learning model extremely sparse. In this study, we propose a learning-to-rank method which not only addresses the sparsity of training signals but also can be trained and updated online. The proposed method is evaluated using both offline experiments and online A/B tests, which involve very large collections of Twitter data and real Twitter users. Results of the experiments prove the effectiveness and efficiency of our solution, and its superiority over the current production model adopted by Twitter.

Unsupervised text embedding methods, such as Skip-gram and Paragraph Vector, have been attracting increasing attention due to their simplicity, scalability, and effectiveness. However, comparing to sophisticated deep learning architectures such as convolutional neural networks, these methods usually yield inferior results when applied to particular machine learning tasks. One possible reason is that these text embedding methods learn the representation of text in a fully unsupervised way, without leveraging the labeled information available for the task. Although the low dimensional representations learned are applicable to many different tasks, they are not particularly tuned for any task. In this paper, we fill this gap by proposing a semi-supervised representation learning method for text data, which we call the \textit{predictive text embedding} (PTE). Predictive text embedding utilizes both labeled and unlabeled data to learn the embedding of text. The labeled information and different levels of word co-occurrence information are first represented as a large-scale heterogeneous text network, which is then embedded into a low dimensional space through a principled and efficient algorithm. This low dimensional embedding not only preserves the semantic closeness of words and documents, but also has a strong predictive power for the particular task. Compared to recent supervised approaches based on convolutional neural networks, predictive text embedding is comparable or more effective, much more efficient, and has fewer parameters to tune., KDD 2015

Community-generated text corpora can be a valuable resource to extract consumer health vocabulary (CHV) and link them to professional terminologies and alternative variants. In this research, we propose a pattern-based text-mining approach to identify pairs of CHV and professional terms from Wikipedia, a large text corpus created and maintained by the community. A novel measure, leveraging the ratio of frequency of occurrence, was used to differentiate consumer terms from professional terms. We empirically evaluated the applicability of this approach using a large data sample consisting of MedLine abstracts and all posts from an online health forum, MedHelp. The results show that the proposed approach is able to identify synonymous pairs and label the terms as either consumer or professional term with high accuracy. We conclude that the proposed approach provides great potential to produce a high quality CHV to improve the performance of computational applications in processing consumer-generated health text.

This paper studies the problem of embedding very large information networks into low-dimensional vector spaces, which is useful in many tasks such as visualization, node classification, and link prediction. Most existing graph embedding methods do not scale for real world information networks which usually contain millions of nodes. In this paper, we propose a novel network embedding method called the "LINE," which is suitable for arbitrary types of information networks: undirected, directed, and/or weighted. The method optimizes a carefully designed objective function that preserves both the local and global network structures. An edge-sampling algorithm is proposed that addresses the limitation of the classical stochastic gradient descent and improves both the effectiveness and the efficiency of the inference. Empirical experiments prove the effectiveness of the LINE on a variety of real-world information networks, including language networks, social networks, and citation networks. The algorithm is very efficient, which is able to learn the embedding of a network with millions of vertices and billions of edges in a few hours on a typical single machine. The source code of the LINE is available online., Comment: WWW 2015

Objective We describe experiments designed to determine the feasibility of distinguishing known from novel associations based on a clinical dataset comprised of International Classification of Disease, V.9 (ICD-9) codes from 1.6 million patients by comparing them to associations of ICD-9 codes derived from 20.5 million Medline citations processed using MetaMap. Associations appearing only in the clinical dataset, but not in Medline citations, are potentially novel. Methods Pairwise associations of ICD-9 codes were independently identified in both the clinical and Medline datasets, which were then compared to quantify their degree of overlap. We also performed a manual review of a subset of the associations to validate how well MetaMap performed in identifying diagnoses mentioned in Medline citations that formed the basis of the Medline associations. Results The overlap of associations based on ICD-9 codes in the clinical and Medline datasets was low: only 6.6% of the 3.1 million associations found in the clinical dataset were also present in the Medline dataset. Further, a manual review of a subset of the associations that appeared in both datasets revealed that cooccurring diagnoses from Medline citations do not always represent clinically meaningful associations. Discussion Identifying novel associations derived from large clinical datasets remains challenging. Medline as a sole data source for existing knowledge may not be adequate to filter out widely known associations. Conclusions In this study, novel associations were not readily identified. Further improvements in accuracy and relevance for tools such as MetaMap are needed to realize their expected utility.

Medical documents provided to patients at the end of an episode of care, such as discharge summaries and referral letters, serve as an important vehicle to convey critical information to patients and families. Increasingly, healthcare institutions are also experimenting with granting patients direct electronic access to other types of clinical narratives that are not typically shared unless explicitly requested, such as progress notes. While these efforts have great potential to improve information transparency, their value can be severely diminished if patients are unable to read and thus unable to properly interpret the medical documents shared to them. In this study, we approached the problem by contrasting the 'readability' of two types of medical documents: referral letters vs. other genres of narrative clinician notes not explicitly intended for direct viewing by patients. To establish a baseline for comparison, we also computed readability scores of MedlinePlus articles - exemplars of fine patient education materials carefully crafted for lay audiences. We quantified document readability using four different measures. Differences in the results obtained through these measures are also discussed.

In this study, we quantified the use of uncertainty expressions, referred to as ‘hedge’ phrases, among a corpus of 100,000 clinical documents retrieved from our institution’s electronic health record system. The frequency of each hedge phrase appearing in the corpus was characterized across document types and clinical departments. We also used a natural language processing tool to identify clinical concepts that were spatially, and potentially semantically, associated with the hedge phrases identified. The objective was to delineate the prevalence of hedge phrase usage in clinical documentation which may have a profound impact on patient care and provider–patient communication, and may become a source of unintended consequences when such documents are made directly accessible to patients via patient portals.