关键词： Big data   User-generated content   LDA   Sentiment analysis   Machine learning  

摘要： Marketing scholars are interested in the big data of user-generated content (UGC) from social media platforms. However, the majority of current UGC studies have been conducted in the business-to-consumer (B2C) context. To fill the knowledge gap in business-to-business (B2B) research, we investigate whether UGC has differential impacts on stock performance for B2B and B2C firms by using big data. We collect a large dataset of 84 million tweets from 20.3 million Twitter accounts and 8 years of stock data for 407 companies from the S&P500 index. The results from machine learning methods are transformed into a monthly panel data. We conduct fixed effects model on the panel data. We find that UGC has a significant impact on firms' stock performance and that its impact on stock performance is much stronger among B2C firms than among B2B firms. While consumers' positive sentiment does not play a significant role in stock performance, consumers' negative sentiment and WOM significantly impact stock prices.

关键词： Analytics   Online returns   Predictive model   Variable selection   LASSO   Machine learning  

摘要： In 2015, U.S. consumers returned goods worth $261 billion and the return rates for online sales sometimes exceeded 30%. Manufacturers and retailers have an interest in predicting return volume to address operational challenges in managing product returns. In this paper, we develop and test data-driven models for predicting return volume at the retailer, product type and period levels using a rich data set comprised of detailed operations on each product, and retailer information. The goal is to achieve a good prediction accuracy out of sample. We consider main effects and detailed interaction effects models using various machine learning methods. We find that Least Absolute Shrinkage and Selection Operator (LASSO) yields a predictive model achieving the best prediction accuracy for future return volume due to its ability to select informative interaction terms out of more than one thousand possible combinations. The LASSO model also turns in consistent performance based on several robustness tests and is easy to implement in practice. Our work provides a general predictive model framework for manufacturers to track product returns. (C) 2019 Elsevier B.V. All rights reserved.

关键词： Waste heat recovery   Organic Rankine Cycle   Dynamic Programming   Transient optimization   Real-time implementation   Machine learning  

摘要： This study aims to present a method for real-time realization of Dynamic Programming algorithm for power optimization in an organic Rankine Cycle waste heat recovery system. Different from existing studies, for the first time machine learning algorithms are utilized to extract the rules from offline Dynamic Programming results for optimal power generation. In addition, for the first time a single state Proper Orthogonal Decomposition and Galerkin Projection based reduced order model is combined with Dynamic Programming for its high accuracy and low computation cost. For a transient driving cycle, Dynamic Programming algorithm is utilized to generate the optimal working fluid pump speed. A total of eleven state-of-art machine learning algorithms are screened to predict this pump speed. Random Forest algorithm is then selected for its best pump speed prediction accuracy. A rule-based method is added to the Random Forest model to improve energy recovery. As one of the main discoveries in this study, in the rule extraction process, the Random Forest model reveals that the time delayed exhaust gas mass flow rate and exhaust temperature improve the rule extraction accuracy. This observation points out the difference between steady state and transient optimization and that the steady state optimization results do not necessarily hold true in transient conditions. Another key observation is that Random Forest - rule-based method retrieves 97.2% of the energy recovered by offline Dynamic Programming in a validation driving cycle. In addition, the inclusion of rule-based method significantly increases the Random Forest model's energy recovery from 66.5% to 97.2%. This high accuracy means that the machine learning models can be used to extract Dynamic Programming rules for real-time application.

关键词： Downscaling   Precipitation   Representative Concentration Pathway   Drainage basin   Model output statistics  

摘要： Downscaling considerably alleviates the drawbacks of regional climate simulation by general circulation models (GCMs). However, little information is available regarding the downscaling using machine learning methods, specifically at hydrological basin scale. This study developed multiple machine learning (ML) downscaling models, based on a Bayesian model average (BMA), to downscale the precipitation simulation of 8 Coupled Model Intercomparison Project Phase 5 (CMIP5) models using model output statistics (MOS) for the years 1961-2005 in the upper Han River basin. A series of statistical metrics, including Pearson's correlation coefficient (PCC), root mean squared error (RMSE), and relative bias (Rbias), were used for evaluation and comparative analyses. Moreover, the BMA and the best ML downscaling model were used to downscale precipitation in the 21st century under Representative Concentration Pathway 4.5 (RCP4.5) and RCP8.5 scenarios. The results show the following: (1) The performance of the BMA ensemble simulation is clearly better than that of the individual models and the simple mean model ensemble (MME). The PCC reaches 0.74, and the RMSE is reduced by 28%-60% for all the GCMs and 33% compared to the MME. (2) The downscaled models greatly improved station simulation performance. Support vector machine for regression (SVR) was superior to multilayer perceptron (MLP) and random forest (RF). The downscaling results based on the BMA ensemble simulation and SVR models were regarded as the best performing overall (PCC, RMSE, and Rbias were 0.82, 35.07, mm and -5.45%, respectively). (3) Based on BMA and SVR models, the projected precipitations show a weak increasing trend on the whole under RCP4.5 and RCP8.5. Specifically, the average rainfall during the mid- (2040-2069) and late (2070-2099) 21st century increased by 3.23% and 1.02%, respectively, compared to the base year (1971-2000) under RCP4.5, while they increased by 4.25% and 8.30% under RCP8.5. Additionally,

关键词： 血管造影图像分割   模型驱动方法   数据驱动方法   机器学习方法   深度学习方法  

摘要： 血管分割能够为计算机临床辅助与可视化诊疗提供关键的结构和病灶信息。由于血管造影图像涉及多种数据模态以及个体化差异和复杂病理,诸多挑战性的问题仍待解决。我们对血管分割的主流方法进行了回顾和分析,并且讨论了血管分割的研究格局和发展趋势,旨在为该领域的进一步研究提供参考。

关键词： 滑坡   灵台县   滑坡易发性评价   机器学习方法  

摘要： 笔者以甘肃省平凉市灵台县为目标研究区域,基于地理空间和历史滑坡数据,利用混合高斯聚类(GMM)优化的逻辑回归(LR)、支持向量机(SVM)、BP神经网络(BP Neural Network)、随机森林(RF)4种机器学习模型构建滑坡易发性评价分析模型。选取高程、坡度、坡向、曲率、黄土侵蚀强度、归一化植被指数、地质构造7个环境因子作为滑坡易发性影响因子,以30 m栅格建立影响因子地理空间数据库,将研究区域划分为180万栅格单元。利用混合高斯聚类模型对整个研究区域的栅格单元进行聚类,得出初步的滑坡易发分区,选择易发程度最低类别中的栅格单元作为非滑坡区域,每次随机选择500个单元作为非滑坡单元,并根据历史滑坡数据将203个已知滑坡栅格单元作为滑坡单元,建立4种机器学习分类模型。利用训练好的模型对整个研究区域进行预测,绘制各算法的受试者工作曲线(ROC曲线),对各个算法的预测结果进行对比。分析结果表明,在本目标研究区域,各模型的滑坡易发区划图与实际的滑坡分布情况总体相吻合。随机森林模型的ROC曲线下面积(AUC)最大为0.96,测试集准确率最高为0.93;BP神经网络模型的ROC曲线下面积和测试集准确率次之,为0.90和0.87;支持向量机模型和逻辑回归模型的ROC曲线下面积和测试集准确率分别为0.86、0.81和0.85、0.80,均低于随机森林和BP神经网络模型。

关键词： 地脉动   分类   地震预测  

摘要： 以"红肿"假说为基础,在由地脉动数据统计量和过往震例构成的样本集上应用数据挖掘中的分类算法开展地震预测实验。筛选符合震级、震中距、发震时间间隔以及未受台风影响等要求的地震对,并以其尾地震作为预测对象。计算地震对时间范围内各时间窗中地脉动数据的标准差,并采用z-score标准化方法对标准差数据进行标准化处理。然后,选取距震中最近三个台的最后一组标准化数据的中位数作为正样本数据,选取各台站平静期数据的中位数作为负样本数据,最后将上述正负样本数据构成样本集。使用CART算法、GBDT算法和SVM算法在此样本集上分别构建预测模型,采用5折交叉验证方法对预测模型进行评估。实验结果表明:①地震与地脉动变化存在一定的关系,且地脉动异常现象更多地出现在6.0级以上地震发生前;②6.0级以上地震构成的正样本对预测模型的构建影响较大;③SVM算法更适用于小样本数据环境。