学术论文引用预测研究进展

[目的/意义]对学术论文引用预测影响因素和预测方法进行梳理,分析现存问题并提出发展方向。[方法/过程]采用文献调研法,综述国内外研究进展,总结预测影响因素和预测方法的相关内容和特点。[结果/结论]现有影响因素指标繁多,无统一标准;预测方法理论基础薄弱;引文预测动态性研究不足;预测模型通用性受限。未来应加强引文预测的理论研究、加强传统文献计量和替代计量的结合、加强自然语言处理的深度应用、建立统一的基线标准、构建更加精准的预测模型。     

船舶交通量三阶段法预测模型及其应用

为精准预测船舶交通量,基于多源数据融合,提出用于预测船舶交通量的三阶段法,分别预测交会水域、单条航道的船舶交通量,并以长江六圩水域以及苏北运河航道为例进行实证研究。结果表明,三阶段法预测模型具有较强的预测精度。预计2040年苏北运河航道货运承载能力饱和度达到17.13%,即未来苏北运河航道利用水平较为合理,但仍处于较低水平,尤其是蔺家坝—宿迁窑湾航段和宿迁窑湾—淮安航段的利用水平更低。本文预测方法和结果可为水路交通规划与调度管理提供支撑。     

An Empirical Analysis of Design Choices in Neighborhood-Based Collaborative Filtering Algorithms

Collaborative filtering systems predict a user's interest in new items based on the recommendations of other people with similar interests. Instead of performing content indexing or content analysis, collaborative filtering systems rely entirely on interest ratings from members of a participating community. Since predictions are based on human ratings, collaborative filtering systems have the potential to provide filtering based on complex attributes, such as quality, taste, or aesthetics. Many implementations of collaborative filtering apply some variation of the neighborhood-based prediction algorithm. Many variations of similarity metrics, weighting approaches, combination measures, and rating normalization have appeared in each implementation. For these parameters and others, there is no consensus as to which choice of technique is most appropriate for what situations, nor how significant an effect on accuracy each parameter has. Consequently, every person implementing a collaborative filtering system must make hard design choices with little guidance. This article provides a set of recommendations to guide design of neighborhood-based prediction systems, based on the results of an empirical study. We apply an analysis framework that divides the neighborhood-based prediction approach into three components and then examines variants of the key parameters in each component. The three components identified are similarity computation, neighbor selection, and rating combination.     

社会化问答平台提问回复率的预测研究——以“百度知道”为例

[目的/意义] 基于社会化问答平台提问回复率较低的现状，通过预测提问回复率，能够为社会化问答平台提升用户活跃度与留存率，改善用户体验提供参考。[方法/过程] 以"百度知道"为研究平台，抓取平台设置的14个话题下共10 640条提问记录，从提问特征与提问者特征角度，构建提问回复率影响因素的研究框架。采用二元Logistic回归对影响因素进行数据验证，构建提问回复率的预测模型，对模型准确率进行验证。[结果/结论] 社会化问答平台提问回复率研究可改善平台信息服务质量与促进用户知识贡献行为，实验结果验证了研究模型在社会化问答平台提问回复率预测中的有效性。     

A local and global event sentiment based efficient stock exchange forecasting using deep learning

Stock exchange forecasting is an important aspect of business investment plans. The customers prefer to invest in stocks rather than traditional investments due to high profitability. The high profit is often linked with high risk due to the nonlinear nature of data and complex economic rules. The stock markets are often volatile and change abruptly due to the economic conditions, political situation and major events for the country. Therefore, to investigate the effect of some major events more specifically global and local events for different top stock companies (country-wise) remains an open research area. In this study, we consider four countries- US, Hong Kong, Turkey, and Pakistan from developed, emerging and underdeveloped economies’ list. We have explored the effect of different major events occurred during 2012–2016 on stock markets. We use the Twitter dataset to calculate the sentiment analysis for each of these events. The dataset consists of 11.42 million tweets that were used to determine the event sentiment. We have used linear regression, support vector regression and deep learning for stock exchange forecasting. The performance of the system is evaluated using the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The results show that performance improves by using the sentiment for these events.     

The kinematic differences between skill levels in the squash forehand drive,volley and drop strokes

ABSTRACT

Knowledge of the kinematic differences that separate highly skilled and less-skilled squash players could assist the progression of talent development. This study compared trunk, upper-limb and racket kinematics between two groups of nine highly skilled and less-skilled male athletes for forehand drive, volley and drop strokes. A 15-camera motion analysis system recorded three-dimensional trajectories, with five shots analysed per participant per stroke. The highly skilled group had significantly (p < 0.05) larger forearm pronation/supination range-of-motion and wrist extension angles at impact than the less-skilled. The less-skilled group had a significantly more “open” racket face and slower racket velocities at impact than the highly skilled. Rates of shoulder internal rotation, forearm pronation, elbow extension and wrist flexion at impact were greater in the drive stroke than in the other strokes. The position of the racket at impact in the volley was significantly more anterior to the shoulder than in the other strokes, with a smaller trunk rotation angular velocity. Players used less shoulder internal/external rotation, forearm pronation/supination, elbow and wrist flexion/extension ranges-of-motions and angular velocities at impact in the drop stroke than in the other strokes. These findings provide useful insights into the technical differences that separate highly skilled from less-skilled players and provide a kinematic distinction between stroke types.     

An Examination of Classification Accuracy in the Continuous Testing Framework

The continuous testing framework, where both successful and unsuccessful examinees have to demonstrate continued proficiency at frequent prespecified intervals, is a framework that is used in noncognitive assessment and is gaining in popularity in cognitive assessment. Despite the rigorous advantages of this framework, this paper demonstrates that there is significant inflation in false negatives as both passers and failers continually take a test, especially for examinees closer to the passing score. Several passing policies are investigated to control the inflation of false negatives while maintaining low false‐positive rates for fixed‐length tests. Lastly, recommendations are made for testing professionals who wish to utilize the rigorous nature of the continuous testing framework while also avoiding the inflation of qualified examinees failing.     

Estimating the probability of abusive head trauma after abuse evaluation

BackgroundEvidence-based, patient-specific estimates of abusive head trauma probability can inform physicians’ decisions to evaluate, confirm, exclude, and/or report suspected child abuse.ObjectiveTo derive a clinical prediction rule for pediatric abusive head trauma that incorporates the (positive or negative) predictive contributions of patients’ completed skeletal surveys and retinal exams.Participants and Setting500 acutely head-injured children under three years of age hospitalized for intensive care at one of 18 sites between 2010 and 2013.MethodsSecondary analysis of an existing, cross-sectional, prospective dataset, including (1) multivariable logistic regression to impute the results of abuse evaluations never ordered or completed, (2) regularized logistic regression to derive a novel clinical prediction rule that incorporates the results of completed abuse evaluations, and (3) application of the new prediction rule to calculate patient-specific estimates of abusive head trauma probability for observed combinations of its predictor variables.ResultsApplying a mean probability threshold of >0.5 to classify patients as abused, the 7-variable clinical prediction rule derived in this study demonstrated sensitivity 0.73 (95% CI: 0.66-0.79) and specificity 0.87 (95% CI: 0.82-0.90). The area under the receiver operating characteristics curve was 0.88 (95% CI: 0.85-0.92). Patient-specific estimates of abusive head trauma probability for 72 observed combinations of its seven predictor variables ranged from 0.04 (95% CI: 0.02-0.08) to 0.98 (95% CI: 0.96-0.99).ConclusionsSeven variables facilitate patient-specific estimation of abusive head trauma probability after abuse evaluation in intensive care settings.     

The contribution of classroom exams to formative evaluation of concept-level knowledge

Are students able to learn from exam experience about their level of knowledge for tested topics? Prior to taking an exam, undergraduates made predictive category learning judgments (CLJs) by estimating the percentage of questions they expected to answer correctly for six topics. After the exam, they made postdictive CLJs for the same topics. Supporting the postdiction superiority hypothesis, postdictive CLJs were slightly more accurate than predictive CLJs, indicating students could make better formative evaluations after taking an exam. However, postdiction accuracy was low, and in a second study, accuracy did not differ for predictive and postdictive CLJs. We also investigated two processes required to make accurate postdictive CLJs: monitoring performance for individual questions and accurately classifying each question with respect to the topic that it assessed. Although students performed these tasks adequately, their less-than-perfect performance would constrain their ability to accurately judge their topical knowledge from exam experience.     

Demand cycles and market segmentation in bicycle sharing

Consumers often display unique habitual behaviors, and knowledge of these behaviors is of great value in prediction of future demand. We investigated consumer behavior in bicycle sharing in Beijing, where demand prediction is critical for cost-effective rebalancing of bicycle locations (putting bikes where and when they will be rented) and supply (number of bicycles). We created baseline statistical demand models, borrowing methods from economics, signal processing and animal tracking to find consumption cycles of 7, 12, 24 h and 7-days. Lorenz curves of bicycle demand revealed significant stratification of consumer behavior and a long-tail of infrequent demand. To overcome the limits of traditional statistical models, we developed a deep-learning model to incorporate (1) weather and air quality, (2) time-series of demand, and (3) geographical location of demand. Customer segmentation was added at a later stage, to explore potential for improvement with customer demographics. Our final machine learning model with tuned hyperparameters yielded around 50% improvement in predictions over a discrete wavelet transform model, and 80–90% improvement in predictions over a naïve model the reflects some current industry practice. We assessed causality in the deep-learning model, finding that location and air quality had the strongest causal impact on demand. The extreme market segmentation of customer demand, and our relatively short time span of data combined to make it difficult to find sufficient data on all customers for a model fit based on segmentation. We reduced our model data to only the 10 most frequent to see whether such segmentation improves our model's predictive success. These results, though limited, suggest that customer behavior within market segments is more stable than across all customers, as was expected.