Rank-based self-training for graph convolutional networks

Graph Convolutional Networks (GCNs) have been established as a fundamental approach for representation learning on graphs, based on convolution operations on non-Euclidean domain, defined by graph-structured data. GCNs and variants have achieved state-of-the-art results on classification tasks, especially in semi-supervised learning scenarios. A central challenge in semi-supervised classification consists in how to exploit the maximum of useful information encoded in the unlabeled data. In this paper, we address this issue through a novel self-training approach for improving the accuracy of GCNs on semi-supervised classification tasks. A margin score is used through a rank-based model to identify the most confident sample predictions. Such predictions are exploited as an expanded labeled set in a second-stage training step. Our model is suitable for different GCN models. Moreover, we also propose a rank aggregation of labeled sets obtained by different GCN models. The experimental evaluation considers four GCN variations and traditional benchmarks extensively used in the literature. Significant accuracy gains were achieved for all evaluated models, reaching results comparable or superior to the state-of-the-art. The best results were achieved for rank aggregation self-training on combinations of the four GCN models.     

Pre-analytical mysteries: A case of severe hypervitaminosis D and mild hypercalcaemia

We describe a case of severe hypervitaminosis D and mild hypercalcaemia in a 68-year-old woman who presented with fatigue and weight loss. Her 25-hydroxy vitamin D (25OHD) was > 400 nmol/L (50-150) and corrected serum calcium was 2.83 mmol/L (2.1-2.6). Her intact parathyroid hormone (PTH) was 4.9 pmol/L (2.0-9.5). Further investigation revealed an IgM kappa paraprotein, and a bone marrow aspirate confirmed a diagnosis of lymphoplasmacytic lymphoma/Waldenstrom’s macroglobulinemia (LPL/WM). As the vitamin D level was discordant with the patient’s other results and presentation, the presence of an assay interferent was suspected. A 1-in-2 dilution of the sample returned a 25OHD result of 84 nmol/L in keeping with the presence of an interferent. Testing for rheumatoid factor was negative. The sample was treated with an antibody blocking reagent (Scantibodies) and results were not consistent with heterophile antibody interference. The sample was then analysed using liquid chromatography tandem mass spectrometry (LC-MS/MS), which returned a 25OHD result of 82 nmol/L. Testing on an alternative immunoassay platform produced a 25OHD result of 75 nmol/L. Reapeted testing on the original platform following reduction of the monoclonal paraprotein with chemotherapy, returned a result of 64 nmol/L. The patient’s mild hypercalcaemia persisted following resolution of the monoclonal paraprotein, in keeping with a diagnosis of primary hyperparathyroidism. This case highlights the potential for paraproteins to cause assay interference, and the importance of considering interference when results are incongruous with the clinical presentation.     

Social media monitoring: Responsive governance in the shadow of surveillance?

Social media monitoring is gradually becoming a common practice in public organizations in the Netherlands. The main purposes of social media monitoring are strategic control and responsiveness. Social media monitoring poses normative questions in terms of transparency, accountability and privacy. We investigate practices of social media monitoring in four Dutch public organizations. Policy departments seem to be more strongly orientated towards monitoring, whereas organizations involved in policy implementation seem to be more inclined to progress to webcare. The paper argues for more transparency on social media monitoring.     

Extending the language modeling framework for sentence retrieval to include local context

Employing effective methods of sentence retrieval is essential for many tasks in Information Retrieval, such as summarization, novelty detection and question answering. The best performing sentence retrieval techniques attempt to perform matching directly between the sentences and the query. However, in this paper, we posit that the local context of a sentence can provide crucial additional evidence to further improve sentence retrieval. Using a Language Modeling Framework, we propose a novel reformulation of the sentence retrieval problem that extends previous approaches so that the local context is seamlessly incorporated within the retrieval models. In a series of comprehensive experiments, we show that localized smoothing and the prior importance of a sentence can improve retrieval effectiveness. The proposed models significantly and substantially outperform the state of the art and other competitive sentence retrieval baselines on recall-oriented measures, while remaining competitive on precision-oriented measures. This research demonstrates that local context plays an important role in estimating the relevance of a sentence, and that existing sentence retrieval language models can be extended to utilize this evidence effectively.     

Second Screening as Convergence in Brazil and the United States

Second screening is widespread worldwide, particularly in younger populations. We analyze a survey of college students in Brazil and the United States to compare second screening frequency, types, platforms, and motivations between the two countries. Despite lower Internet penetration, Brazilians second screen significantly more than Americans, a result of the country’s tradition of interacting with producers of television. In both countries, those who use the interactive affordances of social media are more likely to second screen. As such, we posit this unique audience-driven act works to bridge Web-connected devices and television to create a converged atmosphere.     

Communicative Antecedents of Political Persuasion: Political Discussion,Citizen News Creation,and the Moderating Role of Strength of Partisanship

Although much attention has been paid to how media use and interpersonal discussion motivate people to engage in political persuasion, and despite recent efforts to study the role of digital media technologies, less is known about the creation of news and public affairs content online. This study sheds light on how online content creation works alongside other communicative behaviors, such as news use and political discussion, to affect attempted political persuasion. Using two-wave panel survey data, we find that political discussion and citizen news creation mediate the relationships between online and traditional news use, on one hand, and attempted persuasion, on the other. Furthermore, strength of partisanship moderates the relationship between content creation and attempted persuasion. Findings are discussed in light of their implications for the political communication and public sphere processes.     

Analysis of the distances covered and technical actions performed by professional tennis players during official matches

The purpose of this study was to analyse the physical and technical performances of professional tennis players during official matches. The trajectories of eight players were obtained during matches, using an automatic tracking method. The distances covered and technical performances were analysed for the first and second sets. The athletes covered (mean ± standard deviation) a total of 1702.4 ± 448.2 m in the first set, 1457.6 ± 678.1 m in the second set and 3160.0 ± 880.1 in the entire match. No differences were found between the sets for the physical variables (lateral and forward displacements, distance covered per rally, per game and per set, and the percentage of time spent in each range of velocity). However, the distances covered by the athletes during the rallies in which they were serving (median = 5.2; interquartile range (IQR) = 6.7 m) were statistically smaller than when they were returning (median = 6.2; IQR = 7.7 m). Forehand ground stroke proficiency decreased from the first (mean ± standard deviation: 75.2 ± 4.11%) to the second set (mean ± standard deviation = 65.5 ± 14.3%). In conclusion, tennis players did not present reduced physical performance from the first to the second set.     

Test-retest reliability and agreement of the SPI-Questionnaire to detect symptoms of digital ischemia in elite volleyball players

The Shoulder posterior circumflex humeral artery Pathology and digital Ischemia – questionnaire (SPI-Q) has been developed to enable periodic surveillance of elite volleyball players, who are at risk for digital ischemia. Prior to implementation, assessing reliability is mandatory. Therefore, the test-retest reliability and agreement of the SPI-Q were evaluated among the population at risk.

A questionnaire survey was performed with a 2-week interval among 65 elite male volleyball players assessing symptoms of cold, pale and blue digits in the dominant hand during or after practice or competition using a 4-point Likert scale (never, sometimes, often and always). Kappa (κ) and percentage of agreement (POA) were calculated for individual symptoms, and to distinguish symptomatic and asymptomatic players. For the individual symptoms, κ ranged from “poor” (0.25) to “good” (0.63), and POA ranged from “moderate” (78%) to “good” (97%). To classify symptomatic players, the SPI-Q showed “good” reliability (κ = 0.83; 95%CI 0.69–0.97) and “good” agreement (POA = 92%). The current study has proven the SPI-Q to be reliable for detecting elite male indoor volleyball players with symptoms of digital ischemia.     

Biomechanics,energetics and coordination during extreme swimming intensity: effect of performance level

The present study aimed to examine how high- and low-speed swimmers organise biomechanical, energetic and coordinative factors throughout extreme intensity swim. Sixteen swimmers (eight high- and eight low-speed) performed, in free condition, 100-m front crawl at maximal intensity and 25, 50 and 75-m bouts (at same pace as the previous 100-m), and 100-m maximal front crawl on the measuring active drag system (MAD-system). A 3D dual-media optoelectronic system was used to assess speed, stroke frequency, stroke length, propelling efficiency and index of coordination (IdC), with power assessed by MAD-system and energy cost by quantifying oxygen consumption plus blood lactate. Both groups presented a similar profile in speed, power output, stroke frequency, stroke length, propelling efficiency and energy cost along the effort, while a distinct coordination profile was observed (F_{(3, 42)} = 3.59, P = 0.04). Speed, power, stroke frequency and propelling efficiency (not significant, only a tendency) were higher in high-speed swimmers, while stroke length and energy cost were similar between groups. Performing at extreme intensity led better level swimmers to achieve superior speed due to higher power and propelling efficiency, with consequent ability to swim at higher stroke frequencies. This imposes specific constraints, resulting in a distinct IdC magnitude and profile between groups.     

The sensitivity of the alternative maximal accumulated oxygen deficit method to discriminate training status

The purpose of the study was to investigate the sensitivity of an alternative maximal accumulated oxygen deficit (MAOD_ALT) method to discriminate the “anaerobic” capacity while comparing: least trained (LT) participants (n = 12), moderately trained (MT) participants (n = 12), endurance trained (ET) participants (n = 16), and rugby (RG) players (n = 11). Participants underwent a graded exercise test on a treadmill and a supramaximal effort for assessing MAOD_ALT. MAOD_ALT was calculated as the sum of oxygen equivalents from the phosphagen and glycolytic metabolic pathways. MAOD_ALT was significantly higher (P < 0.05) in RG (64.4 ± 12.1 mL · kg^?1) than in ET (56.8 ± 5.4 mL · kg^?1; effect size [ES] = 0.77; +13.5%), MT (53.8 ± 5.3 mL · kg^?1; ES = 1.08; +19.8%), and LT (49.9 ± 4.5 mL · kg^?1; ES = 1.50; +36.4%). In addition, the magnitude-based inference analysis revealed that MAOD_ALT was likely (LT vs. MT), very likely (MT vs. RG, and ET vs. RG) and most likely (LT vs. ET, and LT vs. RG) different between all groups, except for MT and ET, which presented an unclear difference. In conclusion, MAOD_ALT was sensitive enough to distinguish the “anaerobic” capacity in individuals with different training status, especially for RG players compared with LT participants and MT participants.