The amount of text data has been growing exponentially in recent years, giving rise to automatic information extraction methods that store text annotations in a database. The current state-of-the-art structured prediction methods, however, are likely to contain errors and it is important to be able to manage the overall uncertainty of the database. On the other hand, the advent of crowdsourcing has enabled humans to aid machine algorithms at scale. In this article, we introduce pi-CASTLE, a system that optimizes and integrates human and machine computing as applied to a complex structured prediction problem involving Conditional Random Fields (CRFs). We propose strategies grounded in information theory to select a token subset, formulate questions for the crowd to label, and integrate these labelings back into the database using a method of constrained inference. On both a text segmentation task over academic citations and a named entity recognition task over tweets we show an order of magnitude improvement in accuracy gain over baseline methods.

No abstract

ObjectivesWe examined the use of data from social media for surveillance of physical activity prevalence in the USA.MethodsWe obtained data from the social media site Twitter from April 2015 to March 2016. The data consisted of 1 382 284 geotagged physical activity tweets from 481 146 users (55.7% men and 44.3% women) in more than 2900 counties. We applied machine learning and statistical modelling to demonstrate sex and regional variations in preferred exercises, and assessed the association between reports of physical activity on Twitter and population-level inactivity prevalence from the US Centers for Disease Control and Prevention.ResultsThe association between physical inactivity tweet patterns and physical activity prevalence varied by sex and region. Walking was the most popular physical activity for both men and women across all regions (15.94% (95% CI 15.85% to 16.02%) and 18.74% (95% CI 18.64% to 18.88%) of tweets, respectively). Men and women mentioned performing gym-based activities at approximately the same rates (4.68% (95% CI 4.63% to 4.72%) and 4.13% (95% CI 4.08% to 4.18%) of tweets, respectively). CrossFit was most popular among men (14.91% (95% CI 14.52% to 15.31%)) among gym-based tweets, whereas yoga was most popular among women (26.66% (95% CI 26.03% to 27.19%)). Men mentioned engaging in higher intensity activities than women. Overall, counties with higher physical activity tweets also had lower leisure-time physical inactivity prevalence for both sexes.ConclusionsThe regional-specific and sex-specific activity patterns captured on Twitter may allow public health officials to identify changes in health behaviours at small geographical scales and to design interventions best suited for specific populations.

Automatic extraction of product attribute values is an important enabling technology in e-Commerce platforms. This task is usually modeled using sequence labeling architectures, with several extensions to handle multi-attribute extraction. One line of previous work constructs attribute-specific models, through separate decoders or entirely separate models. However, this approach constrains knowledge sharing across different attributes. Other contributions use a single multiattribute model, with different techniques to embed attribute information. But sharing the entire network parameters across all attributes can limit the model's capacity to capture attribute-specific characteristics. In this paper we present AdaTag, which uses adaptive decoding to handle extraction. We parameterize the decoder with pretrained attribute embeddings, through a hypernetwork and a Mixture-of-Experts (MoE) module. This allows for separate, but semantically correlated, decoders to be generated on the fly for different attributes. This approach facilitates knowledge sharing, while maintaining the specificity of each attribute. Our experiments on a realworld e-Commerce dataset show marked improvements over previous methods. * Most of the work was done during an internship at Amazon.

Online privacy has become immensely important with the growth of technology and the expansion of communication. Social Media Networks have risen to the forefront of current communication trends. With the current trends in social media, the question now becomes how can we actively protect ourselves on these platforms? Users of social media networks share billions of images a day. Whether intentional or unintentional, users tend to share private information within these images. In this study, we investigate (1) the users’ perspective of privacy, (2) pervasiveness of privacy leaks on Twitter, and (3) the threats and dangers on these platforms. In this study, we incorporate techniques such as text analysis, analysis of variance, and crowdsourcing to process the data received from these sources. Based on the results, the participants’ definitions of privacy showed overlap regardless of age or gender identity. After looking at the survey results, most female participants displayed a heightened fear of dangers on social media networks because of threats in the following areas: assets and identity. When the participants were asked to rank the threats on social media, they showed a high concern for burglary and kidnapping. We find that participants need more education about the threats of visual content and how these privacy leaks can lead to physical, mental, and emotional danger.