Using structural topic modelling to predict users’ sentiment towards intelligent personal agents. An application for Amazon’s echo and Google Home

Despite growing levels of usage of Intelligent Personal Assistants (hereinafter, IPA), their in-home usage has not been studied in depth by scholars. To increase our understanding of user interactions with IPA, our research created a theoretical framework rooted in technology acceptance models and Uses and Gratification Theory. Our empirical method designs an ambitious analysis of natural and non-structured narratives (user-generated content) on Amazon’s Echo and Google Home. And to identify key aspects that differentially influence the evaluation of IPA our method employs machine-learning algorithms based on text summarisation, structural topic modelling and cluster analysis, sentiment analysis, and XGBoost regression, among other approaches. Our results reveal that (hedonic and utilitarian) benefits gratification, social influence and facilitating conditions have a direct impact on the users’ sentiment for IPA. To sum up, designers and managers should recognise the challenge of increasing the customer satisfaction of current and potential users by adjusting doubtful users’ technical skills and the (hedonic, cognitive, and social) benefits and functionalities of IPA to avoid boredom after a short lapse of time. Finally, the discussion section outlines future lines of research and theoretical and managerial implications.     

Text mining-based four-step framework for smart speaker product improvement and sales planning

The smart speaker market, which is considered an early-stage market, is expected to grow rapidly as smart speakers become a part of daily life. Consequently, manufacturers are trying to dominate the market. To achieve this, they must analyze users' reactions to their products and find insights for product improvement through comparison with competitors. We propose a four-step methodological framework for identifying meaningful opinions from a large number of online user reviews. First, network analysis is conducted to compare differences between brands. Next, through topic modeling, the attributes of speakers that users consider the most crucial are extracted for each brand. Third, sentiment analysis is conducted to examine how users' emotional polarities differ for each attribute. Through this, product improvement and product sales plans can be derived. Finally, in order to clarify the strengths and weaknesses of each brand, brand positioning is conducted and user opinions that have changed along with the evolution of the speaker's generation are analyzed. Our study identified the factors that positively or negatively affect the experience of smart speaker users. In addition, the proposed method is highly useful because it can be used to derive insights from a large amount of user opinion data regardless of the search term.     

Effect of personal branding stereotypes on user engagement on short-video platforms

The emergence of short-video platforms offers individuals an opportunity to build personal branding. This study selected two personal brandings on Bilibili.com, Li Ziqi and HuaNong Brothers, to explore the differences in the effect of personal branding stereotypes on user engagement on short-video platforms. We first identified the perception of users based on the Stereotype Content Model, then adopted regression models to construct the relationship between personal branding stereotypes and user engagement. The results show that perceived warmth exerts a more significant influence on user emotional engagement, while perceived competence impacts more on user cognitive and behavioral engagement. Moreover, we find the live comments’ sentiment moderates these effects. This research provided references for individuals and managers to benefit from personal brandings.     

An Investigation of Brand-Related User-Generated Content on Twitter

The big data of user-generated content (UGC) on social media are laden with potential value for brand managers. However, there are many obstacles to using big data to answer brand-management questions. This article presents a framework that automatically derives latent brand topics and classifies brand sentiments. It applies text mining with latent Dirichlet allocation (LDA) and sentiment analysis on 1.7 million unique tweets for 20 brands across five industries: fast food, department store, footwear, electronics, and telecommunications. The framework is used to explore four brand-related questions on Twitter. There are three main findings. First, product, service, and promotions are the dominant topics of interest when consumers interact with brands on Twitter. Second, consumer sentiments toward brands vary within and across industries. Third, separate company-specific analyses of positive and negative tweets generate a more accurate understanding of Twitter users' major brand topics and sentiments. Our findings provide brand managers with actionable insights in targeted advertising, social customer relationship management (CRM), and brand management.     

Dynamic impact of negative public sentiment on agricultural product prices during COVID-19

The COVID-19 pandemic has had a significantly negative impact on public sentiment, which has resulted in panic and some irrational buying behavior, which in turn has had a complex impact on agricultural product prices. This study quantified online negative sentiment using micro-blog text mining and a time-varying parameter vector autoregressive model (TVP-VAR) to empirically analyze the dynamic impact of negative public emotions on agricultural product prices during the COVID-19 pandemic in China. It was found that the online negative sentiment impacted agricultural products prices during COVID-19 and had significant time-varying, lag, and life cycle characteristics, with the responses being most significant in the spread and recession periods. Differences were found in the price responses for different agricultural products and in different risk areas. The online negative sentiment was found to have the greatest impact on vegetable prices, with livestock products and vegetable prices being mainly positively impacted, fruit prices being mainly negatively impacted, and aquatic product prices being negatively impacted in the early stage and positively impacted in the middle and late stages. The online negative sentiment had the greatest impact on medium-risk area agricultural product prices, followed by low-risk areas, with the lowest impact found on the high-risk area agricultural product prices. Three policy suggestions for epidemic monitoring, public opinion guidance and control, and the timely release of agricultural product information are given based on the results.     

基于深度学习的毫米波系统波束成形

针对高移动用户在毫米波系统中用户与基站之间频繁切换增大延迟开销问题,提出了一种新颖的集成深度学习协调波束成形（Deep Learning Coordinated Beamforming,DLCBF）解决方案。该方案通过协调多个基站同时为一个移动用户服务,接收用户上行导频序列和预编码码本训练模型,进而预测最佳下行波束向量,克服了大型天线阵列毫米波系统中选取最佳波束成形向量需要的巨大训练开销。仿真结果表明,所提方案与传统的毫米波波束成形策略相比有更高的频谱有效率。     

基于谱聚类的用户关联关系挖掘

为了从用户地理空间分布数据中挖掘用户间关联关系,提出了一种基于谱聚类的关联关系挖掘算法。首先定义了关联度,用以衡量用户之间空间分布的相似性,基于关联度构造相似矩阵,再利用谱聚类方法对用户进行聚类分析,聚类结果表征了用户的关联关系。采用Silhouette指标和聚类准确率来衡量用户关系挖掘质量,同时与传统的K-Means方法进行了比较,通过真实数据集实验,结果表明该算法在实验数据集上能达到90%以上的聚类准确率,证明方法有效、可行。     

Health as a Means Towards Profitable Ends: mHealth Apps,User Autonomy,and Unfair Commercial Practices

In this article, we discuss mHealth apps and their potential to influence the user’s behaviour in increasingly persuasive ways. More specifically, we call attention to the fact that mHealth apps often seek to not only influence the health behaviour of users but also their economic behaviour by merging health and commercial content in ways that are hard to detect. We argue that (1) such merging of health and commercial content raises specific questions concerning the autonomy of mHealth app users, and (2) consumer law offers a promising legal lens to address questions concerning user protection in this context. Based on an empirically informed ethical analysis of autonomy, we develop a fine-grained framework that incorporates three different requirements for autonomy that we call “independence,” “authenticity,” and “options.” This framework also differentiates between three different stages of mHealth app use, namely installing, starting to use, and continuing to use an app. As a result, user autonomy can be analysed in a nuanced and precise manner. Since the concept of autonomy plays a prominent, yet poorly understood role in unfair commercial practice law, we utilize the ethical analysis of autonomy to guide our legal analysis of the proper application of unfair commercial practice law in the mHealth app domain.     

基于网络效应的SNS网站用户参与动机和参与强度研究

以用户使用社交性网络服务(SNS)网站的参与动机和参与强度关系为研究目的,运用动机理论和网络效应理论,构建基于网络外部性的SNS网站用户参与动机和参与强度关系模型,选取人人网用户作为调研对象,通过问卷调查方式进行实证研究。研究表明:娱乐性动机是影响用户参与强度的主要因素;间接网络效应(感知附加功能)对用户参与动机全部具有显著影响,但直接网络效应(参与成员数量)对用户的参与动机只有局部显著影响,且娱乐性动机在网络效应和参与强度的关系中起到了中介效应作用。依据研究结论,SNS网站的运营商应在明确网络环境下的"动机—行为"理论的基础上,从鼓励用户分享、丰富娱乐资源,构建服务平台、探索新的盈利模式两个方面加强自身建设。     

基于DOA估计的认知无线网络频谱感知

认知无线电技术中频谱感知性能的优劣直接影响认知通信系统的性能。针对该特点提出了认知无线电网络中基于波达方向（DOA）估计的主用户频谱感知模型,即单主用户多次用户和多主用户多次用户的系统模型,选取基于特征分解的多重信号分类（MUSIC）算法分析两种模型的感知性能,包括虚警概率、漏检概率、最小总错误概率、算法复杂度等,获得了闭值表达式,最后在两种模型下对算法进行了仿真。仿真结果表明：各参数主要影响虚警概率,而漏检概率几乎不受影响,验证了方法的有效性。     

采用改进全卷积网络的“高分一号”影像居民地提取

针对现有的高分辨率遥感图像居民地信息提取精度和效率不够高的问题，提出了一种基于改进全卷积网络的“高分一号”（GF-1）遥感影像居民地提取方法。首先，通过专业的目视解译制备大量居民地训练样本；然后，将预训练过的深度卷积神经网络进行全卷积网络的改造，并以具有多尺度卷积核的Inception模块代替由全连接层改造的卷积层，达到减小网络模型参数量、增加特征表达能力的目的；最后，用制作好的高分辨率遥感图像居民地数据集进行训练和验证，生成可直接进行居民地信息提取的全卷积网络。实验结果表明，基于改进全卷积网络的方法可以实现精确有效的居民地信息提取，Kappa系数超过94%。     

What makes you click?—Mate preferences in online dating

We estimate mate preferences using a novel data set from an online dating service. The data set contains detailed information on user attributes and the decision to contact a potential mate after viewing his or her profile. This decision provides the basis for our preference estimation approach. A potential problem arises if the site users strategically shade their true preferences. We provide a simple test and a bias correction method for strategic behavior. The main findings are (i) There is no evidence for strategic behavior. (ii) Men and women have a strong preference for similarity along many (but not all) attributes. (iii) In particular, the site users display strong same-race preferences. Race preferences do not differ across users with different age, income, or education levels in the case of women, and differ only slightly in the case of men. For men, but not for women, the revealed same-race preferences correspond to the same-race preference stated in the users’ profile. (iv) There are gender differences in mate preferences; in particular, women have a stronger preference than men for income over physical attributes.     

Analysing customers' reviews and ratings for online food deliveries: A text mining approach

The purpose of this study was to explore the relationship between online reviews and ratings through text mining and empirical techniques. An Indian food delivery portal ( Zomato.com ) was used, where 50 restaurants on Presence Across Nation (PAN) basis were selected through stratified random sampling. A total of 2530 reviews were collected, scrutinized, and analysed. Using the NVivo software for qualitative analysis, seven themes were identified from collected reviews, out of which, the ‘delivery’ theme was explored further for identifying sub-themes. Linear regression modelling was used to identify the variables affecting delivery ratings and sentiment analysis was also performed on the identified sub-themes. Regression results revealed that hygiene and pricing (delivery subthemes) demonstrated lower delivery ratings. These variables can be established as indicators for restaurants and related online food delivery services to build their business model around them. Similarly, negative sentiments were observed in pricing and hygiene sub-themes. Restaurants and online food services can enhance hygiene levels of their food delivery process in order to receive higher delivery ratings. Similarly, pricing of food items can be modified such that customers are not deterred from ordering the items—food and ordering service do not become cost-prohibitive. This study devised a standardized methodology for analysing vast amounts of online user-generated content (UGC). Findings from this study can be extrapolated to other sectors and service industries such as, tourism, cleaning, transportation, hospitals and engineering especially during the pandemic.     

Avatar-mediated experience in the metaverse: The impact of avatar realism on user-avatar relationship

This study aims to explore the role of avatars in mediating users with the metaverse environment. This study applies the construal level theory and source attractiveness model to understand the impact of avatar realism on the interaction between users and their avatars. Computer-simulated experiences were tested in realistic and abstract avatar conditions using serial mediation analysis. It was found that the user-avatar relationship and the intention to use the avatar were both increased by avatar realism. Specifically, psychological distance and social presence serially mediated these effects. Furthermore, the mediation of psychological distance and social presence was moderated by user familiarity with avatars. This study identifies the role of psychological distance to the user-and-avatar dyad in the metaverse by examining the multi-layered psychological mechanism behind the virtual interaction and relationship. It also provides implications for developing metaverse platforms by considering avatar characteristics and user familiarity.     

一种基于孪生神经网络的深度异构网络嵌入算法

针对现有异构网络嵌入方法导致的捕获关系冗余和模糊的问题,提出了一种基于孪生神经网络的深度异构网络嵌入模型。首先,基于面向关系的深度嵌入(Relation-Oriented Deep Embedding,RODE)框架构建了异构网络嵌入模型,以区分同型节点和异型节点之间的关系;其次,将同型节点与异类节点之间的相似性近似到低维空间,通过构建多任务的孪生神经网络来实现节点之间结构和语义关系的深度嵌入;最后,选取四个数据集执行典型网络挖掘任务,并与其他六种算法进行实验对比分析。实验结果表明,保持相同类型节点之间的相似性有助于提高节点分类效率,且损失函数在提高异构网络嵌入质量方面具有良好的优越性;RODE模型能够有效提高稀疏网络的嵌入质量,且具有良好的稳定性和鲁棒性。     

An empirical comparison of machine learning methods for text-based sentiment analysis of online consumer reviews

The amount of digital text-based consumer review data has increased dramatically and there exist many machine learning approaches for automated text-based sentiment analysis. Marketing researchers have employed various methods for analyzing text reviews but lack a comprehensive comparison of their performance to guide method selection in future applications. We focus on the fundamental relationship between a consumer’s overall empirical evaluation, and the text-based explanation of their evaluation. We study the empirical tradeoff between predictive and diagnostic abilities, in applying various methods to estimate this fundamental relationship. We incorporate methods previously employed in the marketing literature, and methods that are so far less common in the marketing literature. For generalizability, we analyze 25,241 products in nine product categories, and 260,489 reviews across five review platforms. We find that neural network-based machine learning methods, in particular pre-trained versions, offer the most accurate predictions, while topic models such as Latent Dirichlet Allocation offer deeper diagnostics. However, neural network models are not suited for diagnostic purposes and topic models are ill equipped for making predictions. Consequently, future selection of methods to process text reviews is likely to be based on analysts’ goals of prediction versus diagnostics.     

A hybrid data mining model of feature selection algorithms and ensemble learning classifiers for credit scoring

Data mining techniques have numerous applications in credit scoring of customers in the banking field. One of the most popular data mining techniques is the classification method. Previous researches have demonstrated that using the feature selection (FS) algorithms and ensemble classifiers can improve the banks' performance in credit scoring problems. In this domain, the main issue is the simultaneous and the hybrid utilization of several FS and ensemble learning classification algorithms with respect to their parameters setting, in order to achieve a higher performance in the proposed model. As a result, the present paper has developed a hybrid data mining model of feature selection and ensemble learning classification algorithms on the basis of three stages. The first stage, as expected, deals with the data gathering and pre-processing. In the second stage, four FS algorithms are employed, including principal component analysis (PCA), genetic algorithm (GA), information gain ratio, and relief attribute evaluation function. In here, parameters setting of FS methods is based on the classification accuracy resulted from the implementation of the support vector machine (SVM) classification algorithm. After choosing the appropriate model for each selected feature, they are applied to the base and ensemble classification algorithms. In this stage, the best FS algorithm with its parameters setting is indicated for the modeling stage of the proposed model. In the third stage, the classification algorithms are employed for the dataset prepared from each FS algorithm. The results exhibited that in the second stage, PCA algorithm is the best FS algorithm. In the third stage, the classification results showed that the artificial neural network (ANN) adaptive boosting (AdaBoost) method has higher classification accuracy. Ultimately, the paper verified and proposed the hybrid model as an operative and strong model for performing credit scoring.     

基于网络编码的用户协作分集系统性能分析

针对非协作通信和传统协作通信系统不能达到高天线分集增益的问题,提出基于网络 编码的用户协作系统模型。研究了基于网络编码不同协作用户的天线分集增益,并分析非协 作机制、传统用户协作机制和基于网络编码用户协作机制的中断概率。理论分析和系统仿真 表明,基于网络编码的用户协作系统能显著降低系统中断概率,获得更大的分集增益,且协 作用户数越多,系统分集度越好。     

产业互联网平台价值共创机理——基于宏图智能物流的案例研究

文章通过宏图智能物流的案例分析,揭示产业互联网平台价值共创的机理。研究发现,产业互联网平台价值共创过程包括平台启动、平台成长和平台拓展三个阶段。产业互联网平台价值共创机理主要体现为平台企业基于瓶颈问题甄别,实施分阶段差异化的平台架构设计,推进平台与产业环节上多边用户的价值共创。在平台启动阶段体现为平台架构归核,通过平台架构精炼和用户关系连接解决用户体验梗阻问题;在平台成长阶段体现为平台架构扩核,通过平台垂直整合和用户关系嵌入解决平台优势弱化问题;在平台拓展阶段体现为平台架构嵌套,通过平台水平包络和用户关系锁定突破平台拓展困境。与消费互联网平台强调用户规模的快速增长不同,产业互联网平台更强调用户深度参与共创实现效率提升。