No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
How Much Is an Image Worth? Airbnb Property Demand Estimation Leveraging Large Scale Image Analytics
... Methods for text and image mining As noted, in business research, the use of text mining has increased notably in the past 10 years, and recent contributions suggest ways for specific business disciplines to implement it, across consumer (Humphreys and Wang, 2017), marketing (Hartmann et al., 2019) and management (McKenney et al., 2018) research. In contrast, developments of image mining techniques are less advanced, and most implementations appear in working papers (Zhang et al., 2017;Liu et al., 2018;Zhang and Luo, 2018;Dzyabura et al., 2018). Due to this state, the conceptualization of both methods is guided mostly by text mining research, following a method engineering approach (Brinkkemper, 1996;Iivari et al., 2000) that prioritizes the design, construction, and evaluation of methods to assess how experiences with one method implementation (e.g. ...
... Predefined attributes (e.g. aesthetic quality of a photo) can be extracted or predicted using machine learning methods, which quantify the impact of certain attributes on a variable of interest (Zhang et al., 2017). Latent topics similarly might be explored with image mining, such as a technique that identifies brand clusters according to brand attributes that appear in product images (Liu et al., 2018). ...
... Methods such as convolutional neural networks (CNN) provide a form of supervised machine learning with superior predictive accuracy, due to their use of a large amount of training data (manually annotated) and many predictors (features). As a specific type of deep learning model, CNN has been widely applied to marketing and information systems efforts, involving text and image mining Liu et al., 2018;Zhang et al., 2017;Zhang and Luo, 2018). ...
Purpose:
Describe and position the state-of-the-art of text and image mining methods in business research. By providing a detailed conceptual and technical review of both methods, it aims to increase their utilization in service research.
Design/methodology/approach:
On a first stage, the authors review business literature in marketing, operations and management concerning the use of text and image mining methods. On a second stage, the authors identify and analyze empirical papers that used text and image mining methods in services journals and premier business. Finally, avenues for further research in services are provided.
Findings:
The manuscript identifies seven text mining methods and describes their approaches, processes, techniques and algorithms, involved in their implementation. Four of these methods are positioned similarly for image mining. There are 39 papers using text mining in service research, with a focus on measuring consumer sentiment, experiences, and service quality. Due to the nonexistent use of image mining service journals, the authors review their application in marketing and management, and suggest ideas for further research in services.
Research limitations/implications:
This manuscript focuses on the different methods and their implementation in service research, but it does not offer a complete review of business literature using text and image mining methods.
Practical implications:
The results have a number of implications for the discipline that are presented and discussed. The authors provide research directions using text and image mining methods in service priority areas such as artificial intelligence, frontline employees, transformative consumer research and customer experience.
Originality/value:
The manuscript provides an introduction to text and image mining methods to service researchers and practitioners interested in the analysis of unstructured data. This paper provides several suggestions concerning the use of new sources of data (e.g. customer reviews, social media images, employee reviews and emails), measurement of new constructs (beyond sentiment and valence) and the use of more recent methods (e.g. deep learning).
... Even though prior research has examined the appeal of visual attributes in improving service sales, empirical inquiries into these attributes have been constrained to low-level elements such as brightness, colors, and image structure (Zhang et al. 2017). Furthermore, there is also a paucity of studies that attempt to articulate the theoretical mechanisms underlying how visual attributes would affect viewers' evaluation of the quality of services being presented. ...
... Drawing on the visual rhetoric theory and extant literature on service tangibility, we construct a research model (see Figure 1) to articulate the role of two distinct visual cues, namely facial cue and verbal anchoring in affecting online service sales and explore how consumers' perceived importance of service tangibility would govern the persuasiveness the two visual cues. Photographical attributes of images such as brightness, color, saturation, and layout have been reported to have an effect on service sales (Zhang et al. 2017), and are thus included in the model as control variables. ...
... Prior research on advertisement and digital imagery has demonstrated the economic impact of image attributes on ads' Click Through Rate (CTR) and service sales Cheng et al. 2012;Zhang et al. 2017). Specifically, low-level image features, such as brightness, saturation, contrast, color, and clarity, were found to not only determine the aesthetics and quality of an image (Jiang et al. 2016), but it also directly impacts the demand of services (Zhang et al. 2017). ...
While past studies have acknowledged that both tangible and intangible features reside in all goods and services, a persistent challenge for online service retailing resides in the difficultly of portraying services online due to the latter's abstract nature. Drawing on visual rhetoric theory, we advance human facial cues and verbal anchoring as visual cues that can bolster the appeal of tangible elements embedded in the portal image of a given service, which in turn culminates in increased sales. We further examine how the effects of facial cue and verbal anchoring are moderated by consumers' reliance on the tangible cues of the service. By employing computer vision and deep learning techniques to parse out portal images of over 299,000 local service offerings on a popular Chinese group buying site, we discovered that the presence of facial cues and the richness of verbal anchoring embedded in the portal image of a service offering significantly increase sales. Our results further illustrate that these effects are reinforced by consumers' perceived importance of service tangibility.
... The available basic features that can be extracted with AIA are: height and width of the image, file size in kB, RGB [4], HSV [5] and greyscale values each as a mean, shannon entropy [6], blur effect (greyscale and RGB) [7], number of unique segmentations (greyscale and RGB) using the Felzenszwalb segmentation algorithm [8], visual complexity as kB, visual balance symmetrical and diagonal [10], image clarity [1], sharpness, brisque IQA [16], warm and hold hues as percentages [1,5], RMS contrast [9], and bounding boxes of detected objects [17]. For details on the variables, please refer to the original citations. ...
... The available basic features that can be extracted with AIA are: height and width of the image, file size in kB, RGB [4], HSV [5] and greyscale values each as a mean, shannon entropy [6], blur effect (greyscale and RGB) [7], number of unique segmentations (greyscale and RGB) using the Felzenszwalb segmentation algorithm [8], visual complexity as kB, visual balance symmetrical and diagonal [10], image clarity [1], sharpness, brisque IQA [16], warm and hold hues as percentages [1,5], RMS contrast [9], and bounding boxes of detected objects [17]. For details on the variables, please refer to the original citations. ...
Images are rapidly increasing in importance for business and social science research. This is evidenced by an increasing number of publications leveraging visual information. However, the scale of these data often exceed human annotation capabilities and extracting features from images automatically is a complex task. While open-source code for automated image analysis exists, implementations are often scattered across different repositories, making their use burdensome for applied researchers. To address this issue, we provide a bundling of automated image analysis tools that enables researchers to easily extract 47 visual features from images, including low-level features such as image quality and brightness as well as high-level features such as face presence and prominence. All code that we consolidate is open-source and applying our pipeline requires no programming skills. Consequently, we hope our tool contributes to making images more accessible to address meaningful, substantive research problems.
... Currently, nearly 1,000 photos are posted on Instagram each second, and more than 50 billion photos have been posted to date (Aslam 2021). As a result, many marketing scholars have begun to address important marketing problems using either image (e.g., Burnap et al. 2020, Dew et al. 2019, Dzyabura and Peres 2021, Guan et al. 2020, Hartmann et al. 2021, Li and Xie 2020, Li et al. 2019, Liu et al. 2017b, Liu et al. 2020, Malik et al. 2019, Peng et al. 2020, Shin et al. 2020, Troncoso and Luo 2020, van der Lans et al. 2021, Xiao and Ding 2014, Zhang et al. 2014, Zhang et al. 2017, Zhang et al. 2021 or video (e.g., Liu et al. 2018, Lu et al. 2016 Berger 2019, Teixeira et al. 2012, Teixeira et al. 2014, Tellis et al. 2019 Jedidi 2020, Zhang et al. 2020) data. ...
... Even if generating theoretical insights is not the goal, the inability to understand how they work reduces confidence and trust in these models, further limiting the adoption of otherwise fruitful applications. As a result, few publications in marketing journals report deep learning models as the primary methodology (Burnap et al. 2020, Dew et al. 2019, Gabel and Timoshenko 2021, Gabel et al. 2019, Guan et al. 2020, Hartmann et al. 2021, Hu et al. 2019, Li et al. 2019, Liu et al. 2019, Liu et al. 2020, Malik et al. 2019, Shin et al. 2020, Timoshenko and Hauser 2019, Tkachenko and Jedidi 2020, Troncoso and Luo 2020, Xia et al. 2019, Zhang and Luo 2018, Zhang et al. 2021, Zhang et al., 2017 Here, we propose the transparent model of unabridged data (TMUD), which resolves this dilemma by integrating several existing analytic tools. A model of unabridged data (MUD) uses raw data as input which is typically rich and unstructured. ...
Recent advancements in computational power and algorithms have enabled unabridged data (e.g., raw images or audio) to be used as input in some models (e.g., deep learning). However, the black box nature of such models reduces their likelihood of adoption by marketing scholars. Our paradigm of analysis, the Transparent Model of Unabridged Data (TMUD), enables researchers to investigate the inner workings of such black box models by incorporating an ex ante filtration module and an ex post experimentation module. We empirically demonstrate the TMUD by investigating the role of facial components and sexual dimorphism in face perceptions, which have implications for four marketing contexts: advertisement (perceptions of approachability, trustworthiness, and competence), brand (perceptions of whether a face represents a brand's typical customer), category (perceptions of whether a face represents a category's typical customer), and customer persona (perceptions of whether a face represents the persona of a brand's customer segment). Our results reveal new and useful findings that enrich the existing literature on face perception, most of which is based on abridged attributes (e.g., width of mouth). The TMUD has great potential to be a useful paradigm for generating theoretical insights and may encourage more marketing researchers and practitioners to use unabridged data.
... Pictures embedded in social media brand content (Villarroel Ordenes et al. 2019), user-generated content in social media (Klostermann et al. 2018), customer reviews of product and services (Zhang and Luo 2018), and product and service offerings on online platforms (e.g., pictures of Airbnb houses; Zhang et al. 2017) all feature image data, which are unstructured, with no predefined numeric representation, multiple facets, and high media richness. State-of-the-art applications such as Google Vision AI and Amazon Rekognition can identify several objects or actions within an image with high accuracy (e.g., emotions, presence of brands). ...
... At a conceptual level, previous research in multimodality faces two main challenges: (1) finding a theoretical lens to fit the focal mode (e.g., text, images, audio, video) and research facet (e.g., syntax, semantics, semiotics) and (2) articulating a good theoretical frame for studying two or more modes (and/or facets) jointly. Studies that deal with only text data tend to draw on linguistic and psycholinguistic theories (Humphreys and Wang 2017); those dealing with image data mostly rely on semiotics of photographic schemes (Farace et al. 2020;Zhang et al. 2017); and research that solely includes video data mainly draws on movement theory (Jia, Kim, and Ge 2020), consistent with the definition of videos as series of images. Few studies use audio data alone, and they mainly pertain to atmospherics (e.g., music, Huang and Labroo 2020). ...
Digital communication, the electronic transmission of information, reflects and influences consumers’ perceptions, attitudes, behaviors, and shopping journeys. Thus, data stemming from digital communication is an important source of insights for retailers, manufacturers, and service firms alike. This article discusses emerging trends and recent advances in digital communication research, as well as its future opportunities for retail practice and research. The authors outline four consumer–retailer domains relevant to digital communication, which in turn frame their discussion of the properties of communication dynamics (e.g., trends, variations) within messages, communicators, and their interaction, as well as communication multimodality (i.e., numeric heuristics, text, audio, image, and video). These factors are critical for understanding and predicting consumers’ behaviors and market developments. Furthermore, this article delineates conceptual and methodological challenges for researchers working in contexts that feature dynamics and multimodality. Finally, this article proposes an agenda for continued research, with the goal of stimulating further efforts to unlock the “black boxes” of digital communication and gain insights into both consumers and markets.
... Human images contained in print ads also increase the effectiveness of ads and website design (Cyr et al. 2009;Xiao and Ding 2014). High-level image content and style, such as image quality and the presence of visual art, has also been shown to affect viewers' evaluation of products and product sales (Hagtvedt and Patrick 2008;Zhang et al. 2017). Following these studies, we consider picture colorfulness, the presence of human face and facial expressions, and picture quality as image characteristics that may affect social media engagement in our analysis. ...
... We also find that tweets containing screenshots receive a significantly lower number of retweets than the ones with professionally taken pictures, probably because screenshots are less interesting and informative than other pictures. Finally, consistent with extant research (Hagtvedt and Patrick 2008;Zhang et al. 2017), we find that high-quality images can improve user engagement on social media posts, as reflected by their significant and positive effect on the number of retweets. ...
Are social media posts with pictures more popular than those without? Why do pictures with certain characteristics induce higher engagement than some other pictures? Using data sets of social media posts about major airlines and sport utility vehicle brands collected from Twitter and Instagram, the authors empirically examine the influence of image content on social media engagement. After accounting for selection bias on the inclusion of image content, the authors find a significant and robust positive mere presence effect of image content on user engagement in both product categories on Twitter. They also find that high-quality and professionally shot pictures consistently lead to higher engagement on both platforms for both product categories. However, the effect of colorfulness varies by product category, while the presence of human face and image–text fit can induce higher user engagement on Twitter but not on Instagram. These findings shed light on how to improve social media engagement using image content.
... Culotta and Cutler (2016) extract brands' social connections on Twitter. Other works analyze visual content (Zhang and Luo, 2023;Zhang et al., 2017;Pavlov and Mizik, 2019) . For instance, Liu et al. (2020) analyze consumergenerated images on social media to study consumer brand perceptions. ...
The development of Generative AI enables large-scale automation of product design. However, this automated process usually does not incorporate consumer preference information from a company's internal dataset. Meanwhile, external sources such as social media and user-generated content (UGC) websites often contain rich product design and consumer preference information, but such information is not utilized by companies in design generation. We propose a semi-supervised deep generative framework that integrates consumer preferences and external data into product design, allowing companies to generate consumer-preferred designs in a cost-effective and scalable way. We train a predictor model to learn consumer preferences and use predicted popularity levels as additional input labels to guide the training of a Continuous Conditional Generative Adversarial Network (CcGAN). The CcGAN can be instructed to generate new designs of a certain popularity level, enabling companies to efficiently create consumer-preferred designs and save resources by avoiding developing and testing unpopular designs. The framework also incorporates existing product designs and consumer preference information from external sources, which is particularly helpful for small or start-up companies who have limited internal data and face the "cold-start" problem. We apply the proposed framework to a real business setting by helping a large self-aided photography chain in China design new photo templates. We show that our proposed model performs well in generating appealing template designs for the company.
... For all marketing communication, the presentation of information is a critical factor in drawing viewers' attentions (Zhang et al., 2017). One aspect that has received tremendous interest is the length of information ). ...
Listing titles significantly impact consumer behaviors on shared-economy platforms (e.g., Airbnb). Due to limited research on the length, informativeness, and creativity of titles in this context, this research investigates consumers’ responses and perceptions towards various types of listing titles by exploring their cognitive processes and conscious preferences. In a three-phase study design, this research analyzed Airbnb listing titles in China and America, conducted eye-tracking experiments, and collected surveys to evaluate the effects of title length and informativeness. The results revealed cultural differences in titles, but similar behaviors among Chinese and English-speaking consumers. Cultural assumptions, processing, and informativeness were discussed.
... Ma, et al analyzed over 75,000 marketplace images of mixed quality and showed that higher quality images were indeed associated with a higher likelihood of purchase intent. Zhand, Lee, Singh, and Srinivasan (2017) found significant improvement in revenue in Airbnb listings that resulted from optimizing low-level image factors. They used computer vision algorithms to classify the quality of images of more than half a million photos across a dozen attributes, including composition, diagonal dominance, visual balance intensity and color, color, hue, saturation, brightness, contrast of brightness, image clarity, figure ground relationships, figure from ground, area difference, color difference, and texture difference. ...
This paper investigates the impact of user generated image content on marketing outcomes in online marketplaces. The study found that image content influences consumer purchase decisions in online marketplaces. Consumers tend to make purchase decisions based on the visual information presented that helps reduce information asymmetry. By utilizing AI techniques such as convolutional neural networks (CNNs) used in computer vision models, marketers can now address previously intractable problems that improve marketing outcomes. This study contributes to the field of marketing science by demonstrating the practical application of AI in enhancing marketing strategies.
... Ma, et al analyzed over 75,000 marketplace images of mixed quality and showed that higher quality images were indeed associated with a higher likelihood of purchase intent. Zhand, Lee, Singh, and Srinivasan (2017) found significant improvement in revenue in Airbnb listings that resulted from optimizing low-level image factors. They used computer vision algorithms to classify the quality of images of more than half a million photos across a dozen attributes, including composition, diagonal dominance, visual balance intensity and color, color, hue, saturation, brightness, contrast of brightness, image clarity, figure ground relationships, figure from ground, area difference, color difference, and texture difference. ...
This paper investigates the impact of user generated image content on marketing outcomes in online marketplaces. The study found that image content influences consumer purchase decisions in online marketplaces. Consumers tend to make purchase decisions based on the visual information presented that helps reduce information asymmetry. By utilizing AI techniques such as convolutional neural networks (CNNs) used in computer vision models, marketers can now address previously intractable problems that improve marketing outcomes. This study contributes to the field of marketing science by demonstrating the practical application of AI in enhancing marketing strategies.
... In addition, Chakraborty et al. [49] developed a Hybrid CNN-LSTM model to extract emotional characteristics from text data, and showed that it solves difficult emotion classification problems well for Yelp reviews. Zhang et al. [50] analyzed the effect of images on Airbnb's accommodation demand by using deep learning. The authors classified the quality of images using a convolution neural network (CNN), and found that high-quality photos increase the demand for the accommodation. ...
Machine learning technology is recently being applied to various fields. However, in the field of online consumer conversion, research is limited despite the high possibility of machine learning application due to the availability of big data. In this context, we investigate the following three research questions. First, what is the suitable machine learning model for predicting online consumer behavior? Second, what is the good data sampling method for predicting online con-sumer behavior? Third, can we interpret machine learning’s online consumer behavior prediction results? We analyze 374,749 online consumer behavior data from Google Merchandise Store, an online shopping mall, and explore research questions. As a result of the empirical analysis, the performance of the ensemble model eXtreme Gradient Boosting model is most suitable for pre-dicting purchase conversion of online consumers, and oversampling is the best method to mitigate data imbalance bias. In addition, by applying explainable artificial intelligence methods to the context of retargeting advertisements, we investigate which consumers are effective in retargeting advertisements. This study theoretically contributes to the marketing and machine learning lit-erature by exploring and answering the problems that arise when applying machine learning models to predicting online consumer conversion. It also contributes to the online advertising literature by exploring consumer characteristics that are effective for retargeting advertisements.
... Peck and Childers [4] reveal that product images can make up for the lack of haptic information of products, thereby increasing consumers' willingness to buy. Other research further shows that the product or property's image has a great influence on consumer purchase decision-making [20,21]. ...
Home sharing is a new industry that was born with the development of sharing economy. The online short-term rental platform is an important carrier for home sharing. On the short-term rental platform, house images are an essential way to display the overall situation of the house, and one of the main channels for tenants to obtain house information. This paper studies the relationship between house images' colors and contents and tenant booking decision-making and satisfaction. By utilizing image mining techniques on the data obtained from a popular short-term rental platform in China, this research reveals that the color richness of house images has a significant negative relationship with both tenant booking decision-making and tenant satisfaction. What's more, both household and leisure content displayed in house images has significant positive relationships with tenant booking decision-making. Our work supplements the research on the impact of house images on home sharing and provides meaningful guidance for both the short-term rental platforms and the landlords.
... While previous research has documented multiple factors that affect the financial performance of peer-to-peer (P2P) properties, including photos of a property, hosts' characteristics, size of the space, and wordings of property descriptions (Ert et al., 2016;Guttentag, 2015;S. Zhang et al., 2017), only a few studies have been conducted on how the wordings of listing titles influence the property pricing (Falk et al., 2019) and performance . Thus, this study aims to extend prior studies by investigating the most effective titles from both customers' perspectives and hosts' perspectives. In the first study, approaching from custom ...
Drawing on selective attention theory and language expectancy theory, and using a mixed method of text analysis and spatial analysis, this study examined the impacts of listing titles and locations on the financial performance of Airbnb properties. Guests’ preferred words and expected informative cues about property, location, and environment in Airbnb titles were first captured by a qualitative study. The results of the second study, which controlled for the spatial dependency based on the Hotspot analysis and geographically weighted regression for 14,938 property-level data in Phuket and Bangkok in Thailand, revealed that the linguistic styles and characteristics affecting the properties’ financial performance were significantly different between the two cities. For hosts in hot spots, affective and perceptive linguistic styles on the titles are recommended, while function-oriented information and photos should be highlighted for the cold spot areas.
... Visual and auditory modalities may convey the same information as text, or something different. Tools like Praat (Boersma & Weenink, 2018) can be used to extract pitch and tone from audio files (e.g., Van Zant & Berger, 2020) and research has started to use computer vision to extract features from images (Li & Xie, 2020;Zhang et al., 2017). While there has been less work in these areas than in text analysis, emerging approaches will hopefully enable better analysis of these important information channels. ...
Language can provide important insights into people, and culture more generally. Further, the digitization of information has made more and more textual data available. But by itself, all that data are just that: data. Realizing its potential requires turning that data into insight. We suggest that automated text analysis can help. Recent advances have provided novel and increasingly accessible ways to extract insight from text. While some psychologists may be familiar with dictionary methods, fewer may be aware of approaches like topic modeling, word embeddings, and more advanced neural network language models. This article provides an overview of natural language processing and how it can be used to deepen understanding of people and culture. We outline the dual role of language (i.e., reflecting things about producers and impacting audiences), review some useful text analysis methods, and discuss how these approaches can help unlock a range of interesting questions. (PsycInfo Database Record (c) 2021 APA, all rights reserved).
... Airbnb is a leading home-sharing platform. As a result, its economic, societal, and spillover effect has attracted much attention from scholars (Zervas et al. 2017, Zhang et al. 2017, Benjaafar et al. 2019, Chen et al. 2021b, Mayya et al. 2021). This space-sharing platform can drastically change local communities (Sundararajan 2016), which may be related to change in criminal patterns. ...
The rise of the sharing economy has disrupted traditional industries and has had many unforeseen societal impacts. This has sparked policy debates on whether and how the sharing economy should be regulated to promote the healthy growth of such markets. In this research, we examine the impact of platform self-regulations in the context of the home-sharing market. Using policy changes that reduce the number of Airbnb listings, we empirically test the impact of platform self-regulations on crime rates. Our results suggest that a reduction in Airbnb listings resulting from platform self-regulations leads to a reduction in crime. We further study the impact of these policy changes on different types of crime and find that these self-regulations lead to a reduction in incidents of crime such as assault, robbery, and burglary but an increase in theft incidents. In addition, we find that the impact of these policies varies based on the neighborhood’s characteristics, such as income, housing price, and population. This research contributes to our understanding of the societal impacts of the sharing economy and the impact of platform self-regulation. Our findings also provide empirical evidence to inform policy making.
... , whether Amazon reviews are informative and sentiment in restaurant reviews (Chakraborty et al., 2019). Research on image data has also used CNNs but within more complex architectures such as VGG-16 to predict image quality (S. Zhang et al., 2017) or classify brand images (Hartmann et al., 2020), Caffe framework to predict brand personality (Liu et al., 2018) and ResNet152 to predict product return rates . Past research on both text and image data has found that deep-learning models that self-generate features have better predictive ability than those that use exante hand-crafted features Liu et al., 2018;X. ...
The video streaming industry is growing rapidly, and consumers are increasingly using ad-supported streaming services. There are important questions related to the effect of ad schedules and video elements on viewer behavior that have not been adequately studied in the marketing literature. In my dissertation, I study these topics by applying causal and/or interpretable machine learning methods on behavioral data. In the first essay, “Finding the Sweet Spot: Ad Scheduling on Streaming Media”, I design an “optimal” ad schedule that balances the interest of the viewer (watching content) with that of the streaming platform (ad exposure). This is accomplished using a three-stage approach applied on a dataset of Hulu customers. In the first stage, I develop two metrics – Bingeability and Ad Tolerance – to capture the interplay between content consumption and ad exposure in a viewing session. Bingeability represents the number of completely viewed unique episodes of a show, while Ad Tolerance represents the willingness of a viewer to watch ads and subsequent content. In the second stage, I predict the value of the metrics for the next viewing session using the machine learning method – Extreme Gradient Boosting – while controlling for the non-randomness in ad delivery to a focal viewer using “instrumental variables” based on ad delivery patterns to other viewers. Using “feature importance analyses” and “partial dependence plots” I shed light on the importance and nature of the non-linear relationship with various feature sets, going beyond a purely black-box approach. Finally, in the third stage, I implement a novel constrained optimization procedure built around the causal predictions to provide an “optimal” ad-schedule for a viewer, while ensuring the level of ad exposure does not exceed her predicted Ad Tolerance. Under the optimized schedule, I find that “win-win” schedules are possible that allow for both an increase in content consumption and ad exposure. In the second essay, “Video Influencers: Unboxing the Mystique”, I study the relationship between advertising content in YouTube influencer videos (across text, audio and images) and marketing outcomes (views, interaction rates and sentiment). This is accomplished with the help of novel interpretable deep-learning architectures that avoid making a trade-off between predictive ability and interpretability. Specifically, I achieve high predictive performance by avoiding ex-ante feature engineering and achieve better interpretability by eliminating spurious relationships confounded by factors unassociated with “attention” paid to video elements. The attention mechanism in the Text and Audio models along with gradient maps in the Image model allow identification of video elements on which attention is paid while forming an association with an outcome. Such an ex-post analysis allows me to find statistically significant relationships between video elements and marketing outcomes that are supplemented by a significant increase in attention to video elements. By eliminating spurious relationships, I generate hypotheses that are more likely to have causal effects when tested in a field setting. For example, I find that mentioning a brand in the first 30 seconds of a video is on average associated with a significant increase in attention to the brand but a significant decrease in sentiment expressed towards the video. Overall, my dissertation provides solutions and identifies strategies that can improve the welfare of viewers, platform owners, influencers and brand partners. Policy makers also stand to gain from understanding the power exerted by different stakeholders over viewer behavior.
... Deep learning methods are used in more recent studies. Zhang et al. (2017) show how the quality and specific attributes of property images on Airbnb can affect the demand. The quality and attributes are human-labeled, and a convolutional neural network is used to train the images based on the labels. ...
Many differentiated products have key attributes that are unstructured and thus high-dimensional (e.g., design, text). Instead of treating unstructured attributes as unobservables in economic models, quantifying them can be important to answer interesting economic questions. To propose an analytical framework for this type of products, this paper considers one of the simplest design products -- fonts -- and investigates merger and product differentiation using an original dataset from the world's largest online marketplace for fonts. We quantify font shapes by constructing embeddings from a deep convolutional neural network. Each embedding maps a font's shape onto a low-dimensional vector. In the resulting product space, designers are assumed to engage in Hotelling-type spatial competition. From the image embeddings, we construct two alternative measures that capture the degree of design differentiation. We then study the causal effects of a merger on the merging firm's creative decisions using the constructed measures in a synthetic control method. We find that the merger causes the merging firm to increase the visual variety of font design. Notably, such effects are not captured when using traditional measures for product offerings (e.g., specifications and the number of products) constructed from structured data.
... Within videos, another important yet distinct form of contextual visual cue comes from the brightness of the picture (Zhang et al. 2019). Considered a simple and useful metric of visual information, brightness of a video is the average of the brightness/illumination of all its pixels . ...
Persuasion success is often related to hard-to-measure characteristics, such as the way the persuader speaks. To examine how vocal tones impact persuasion in an online appeal, this research measures persuaders’ vocal tones in Kickstarter video pitches using novel audio mining technology. Connecting vocal tone dimensions with real-world funding outcomes offers insight into the impact of vocal tones on receivers’ actions. The core hypothesis of this paper is that a successful persuasion attempt is associated with vocal tones denoting (1) focus, (2) low stress, and (3) stable emotions. These three vocal tone dimensions—which are in line with the stereotype content model—matter because they allow receivers to make inferences about a persuader’s competence. The hypotheses are tested with a large-scale empirical study using Kickstarter data, which is then replicated in a different category. In addition, two controlled experiments provide evidence that perceptions of competence mediate the impact of the three vocal tones on persuasion attempt success. The results identify key indicators of persuasion attempt success and suggest a greater role for audio mining in academic consumer research.
... With the development of Deep Learning, it became motivating to use such approaches to classify hotel images. Zhang et al. state that there is significant value in optimizing images in e-commerce settings [11]. They perform a CNN model on 16-month Airbnb panel dataset to classify the aesthetic quality for each image in the training sample. ...
... Liu et al., 2019), whether Amazon reviews are informative and sentiment in restaurant reviews (Chakraborty et al., 2019). Research on image data has also used CNNs but within more complex architectures such as VGG-16 to predict image quality (Zhang et al., 2017) or classify brand images (Hartmann et al., 2020), Caffe framework to predict brand personality (L. Liu et al., 2018) and ResNet152 to predict product return rates . ...
Influencer marketing is being used increasingly as a tool to reach customers because of the growing popularity of social media stars who primarily reach their audience(s) via custom videos. Despite the rapid growth in influencer marketing, there has been little research on the design and effectiveness of influencer videos. Using publicly available data on YouTube influencer videos, we implement novel interpretable deep learning architectures, supported by transfer learning, to identify significant relationships between advertising content in videos (across text, audio, and images) and video views, interaction rates and sentiment. By avoiding ex-ante feature engineering and instead using ex-post interpretation, our approach avoids making a trade-off between interpretability and predictive ability. We filter out relationships that are affected by confounding factors unassociated with an increase in attention to video elements, thus facilitating the generation of plausible causal relationships between video elements and marketing outcomes which can be tested in the field. A key finding is that brand mentions in the first 30 seconds of a video are on average associated with a significant increase in attention to the brand but a significant decrease in sentiment expressed towards the video. We illustrate the learnings from our approach for both influencers and brands.
... Quality. The quality of images is an important part of the visual presentation and can lead to major differences in demand (Zhang et al., 2017). As a measure for quality, we evaluate the focus in the frames. ...
Video content has become a major component of total internet traffic. Growing bandwidth and computational power conspire with an increasing number of video editing tools, smartphones, and online platforms that have facilitated video production, distribution, and consumption by businesses and consumers alike. This makes video content relevant across business research disciplines. However, analyzing videos can be a cumbersome manual task. Automated techniques are scattered across technical publications and are often not directly accessible to business researchers. This article synthesizes the current state of the art and provides a consolidated tool to efficiently extract 109 video-based variables, requiring no programming knowledge. The variables include structural video characteristics such as colorfulness as well as advanced content-related features such as scene cuts or human face detection. The authors discuss the research potential of video mining, the types of video features of likely interest, and illustrate application using a practical example.
... works directly on the raw images. Analyzing a flickr dataset, they showed the brand portrayals generated from the consumer supplied images are consistent with the corresponding firm-generated images as well as the result of a national survey, showing that their methods are able to extract accurate brand information from visual data on social media.Zhang et al. (2017) investigated the impact of images on property demand at Airbnb. They used a deep convolutional neural network classified to label images as high or low quality, and showed the verified photos with high image quality significantly increase property demand.Machine learning and AI methods have also been used in other marketing studies to g ...
Artificial intelligence (AI) agents driven by machine learning algorithms are rapidly transforming the business world, generating heightened interest from researchers. In this paper, we review and call for marketing research to leverage machine learning methods. We provide an overview of common machine learning tasks and methods, and compare them with statistical and econometric methods that marketing researchers traditionally use. We argue that machine learning methods can process large-scale and unstructured data, and have flexible model structures that yield strong predictive performance. Meanwhile, such methods may lack model transparency and interpretability. We discuss salient AI-driven industry trends and practices, and review the still nascent academic marketing literature which uses machine learning methods. More importantly, we present a unified conceptual framework and a multi-faceted research agenda. From five key aspects of empirical marketing research: method, data, usage, issue, and theory, we propose a number of research priorities, including extending machine learning methods and using them as core components in marketing research, using the methods to extract insights from large-scale unstructured, tracking, and network data, using them in transparent fashions for descriptive, causal, and prescriptive analyses, using them to map out customer purchase journeys and develop decision-support capabilities, and connecting the methods to human insights and marketing theories. Opportunities abound for machine learning methods in marketing, and we hope our multi-faceted research agenda will inspire more work in this exciting area.
... Quality. The quality of images is an important part of the visual presentation and can lead to major differences in demand (Zhang et al. 2017). As a measure for quality, we evaluate the focus in the frames. ...
Video content has become a major component of total internet traffic. Growing
bandwidth and computational power conspire with an increasing number of video editing tools, smartphones, and online platforms that have facilitated video production, distribution, and consumption by businesses and consumers alike. This makes video content relevant across business research disciplines, including commercials, product presentations, and manager interactions. However, analyzing videos can be a cumbersome manual task. Automated techniques are scattered across technical publications and are often not directly accessible to business researchers. This article synthesizes the current state of the art and provides a consolidated tool to efficiently extract 109 video-based variables, requiring no programming knowledge for standard use cases. The variables include structural video characteristics such as colorfulness as well as advanced content-related features such as scene cuts or human face detection. The authors discuss the research potential of video mining, the types of video features of likely interest and illustrate application using a practical example.
... In recent years, scholars have been keen to explore the impact of landlord photos on users' decisions on shared short-term rentals: Ert et al. used the experimental method to study Airbnb and found that compared with the landlord's reputation, the landlord's photos can influence consumers' shared short-term rental behavior [14]. Zhang et al. and others studied the impact of the source and quality of landlord's photos on the use of housing resources [15]. They found that the platform professionals took more convincing pictures, and high-quality pictures would also increase landlord's income. ...
With the development of social networks and the Internet-based sharing economy, shared short-term rentals are emerging as a new kind of service that provides a convenient way for people to buy short-term rental services in cities through social-network-enabled platforms. However, like other social-network-based services, shared short-term rental is also likely to be impacted by online reviews. This paper aims to investigate the impact of online reviews on users’ purchase decisions toward shared short-term rentals, and further to provide optimization suggestions for the future advance of shared short-term rentals. The contributions of this paper are many-fold. First, we introduce the Stimuli-Organism-Response (SOR) model into the study and propose new variables for the model, including stimulus variables, organism variables, response variable, and moderating variables. Second, we propose eight hypotheses to evaluate the impact of online reviews on users’ purchase decisions toward shared short-term rentals. Finally, we collect data through a questionnaire survey and present comprehensive results on many aspects. Based on the data analysis, we find out that the quality of online reviews impacts users’ perceived value and perceived risk, which in turn impacts users’ purchase decisions toward shared short-term rentals. In addition, the cognitive needs of users can adjust the impact of online reviews on the perceived risk of users but have no explicit adjusting effect for users’ perceived value. Further, we present some research implications as well as suggestions for rental platforms to advance shared short-term rentals in the Internet age.
... Wang and Li use image processing algorithms that allow extracting color-, complexity-, and symmetry-related information in their aesthetic analysis of app icons (Wang and Li, 2017). Zhang et al. use techniques from deep learning and computer vision to build a dedicated supervised learning classifier that classifies Airbnb property images into high or low quality (Zhang et al., 2017). ...
While user-generated images represent important information sources in IS in general and in social media in particular, there is little research that analyzes image design and its effects on image popu-larity. We introduce an innovative computational approach to extract image design characteristics that includes convolutional neural network-based image classification, a dimensionality reduction via principal component analysis, manual measurement validation, and a regression analysis. An analysis of 790,775 car images from 17 brands posted in 68 car model communities on a social me-dia platform reveals several effects of product presentation on image popularity that relate to the levels of utility reference, experience reference, and visual detail. A comparison of economy cars and premium cars shows that car class moderates these image design effects. Our results contribute to the extant literature on brand communities and content popularity in social media. The proposed computational visual analysis methodology may inform the study of other image-based IS.
In the emerging business model of short video commerce, consumers can engage in short sales videos by liking, commenting, and sharing. Compared with traditional shopping contexts, consumer engagement behavior in short video commerce settings is affected quickly and simultaneously by text, audio, and visual information, but the extant literature lacks sufficient exploration of the effects of such multimodal information. Based on the multimodal theory and signaling theory, this study aims to investigate the role of multimodal information including textual sentiment score, audio spectrum and visual effects of short sales videos on consumer engagement behavior. We collect 4292 veridical short sales videos from short video platform Douyin, while based on a multi-method approach of multiple regression analysis (MRA) and fuzzy set qualitative comparative analysis (fsQCA), it is found that the multimodal information features of short sales videos contribute to the promotion of consumer engagement behavior, and the cross-modal configurational solutions to enhance engagement is not unique. This study bridges the gap in the existing literature by capturing the impact of multimodal information features on consumer engagement behaviors and their asymmetric effects, which also provides practical implications for sellers and marketers on attracting consumers to engage in short video commerce.
Customer Generated Image (CGI) on e-commerce platforms has been widely recognized as a marketing tool to persuade customers into purchases. Despite its persuasive power, the effect of CGI on post purchase satisfaction has seldom been examined. This study draws upon Elaboration Likelihood Model and proposes that the affective cues in CGI could distract consumer's cognitive information processing and lead to unsatisfactory purchases with a larger probability. To empirically test our hypothesis, we employed a difference-indifferences model with propensity score weighting method and deep learning based face detection algorithm and found that CGI could cause subsequent review ratings 0.12 stars lower compared with those not exposed to CGI. Additional analysis indicated that this negative effect could be attenuated if the CGI contains human faces or the image review has a low rating. These findings have important implications for online platforms to better leverage user generated rich media content.
In recent years, machine learning (ML) and artificial intelligence (AI) have attracted considerable attention in different industry sectors, including marketing. ML and AI hold great promise for making marketing intelligent and efficient. In this study, we conduct a literature review of academic journal studies on ML in marketing applications and propose a conceptual framework highlighting the main ML tools and technologies that serve as the foundation of ML applications in marketing. We use the 7Ps marketing mix, that is, product, price, promotion, place, people, process, and physical evidence, to analyze these applications from 140 selected articles. The applications are supported by various ML tools (text, voice, image, and video analytics) and techniques such as supervised, unsupervised, and reinforcement learning algorithms. We propose a two-layer conceptual framework for ML applications in marketing development. This framework can serve future research and provide an illustration of the development of ML applications in marketing.
Social media platforms are becoming increasingly important marketing channels, and recently these channels are becoming dominated by content that is not textual, but visual in nature. In this paper, we explore the relationship between the visual complexity of firm-generated imagery (FGI) and consumer liking on social media. We use previously validated image mining methods, to automatically extract interpretable visual complexity measures from images. We construct a set of six interpretable measures that are categorized as either (1) feature complexity measures (i.e., unstructured pixel-level variation; color, luminance, and edges) or (2) design complexity measures (i.e., structured design-level variation; number of objects, irregularity of object arrangement, and asymmetry of object arrangement). These measures and their interpretability are validated using a human subject experiment. Subsequently, we relate these visual complexity measures to the number of likes. The results show an inverted u-shape between feature complexity and consumer liking and a regular u-shape relationship between design complexity and consumer liking. In addition, we demonstrate that using the six individual measures that constitute feature- and design complexity provides a more nuanced view of the relationship between the unique aspects of visual complexity and consumer liking of FGI on social media than observed in previous studies that used a more aggregated measure. Overall, the automated framework presented in this paper opens up a wide range of possibilities for studying the role of visual complexity in online content.
Revolutionary advances in machine and deep learning techniques within the field of computer field have dramatically expanded our opportunities to decipher the merits of digital imagery in the business world. Although extant literature on computer vision has yielded a myriad of approaches for extracting core attributes from images, the esotericism of the advocated techniques hinders scholars from delving into the role of visual rhetoric in driving business performance. Consequently, this tutorial aims to consolidate resources for extracting visual features via conventional machine and/or deep learning techniques. We describe resources and techniques based on three visual feature extraction methods, namely calculation-, recognition-, and simulation-based. Additionally, we offer practical examples to illustrate how image features can be accessed via open-sourced python packages such as OpenCV and TensorFlow.
Video is one of the fastest growing online services offered to consumers. The rapid growth of online video consumption brings new opportunities for marketing executives and researchers to analyze consumer behavior. However, video also introduces new challenges. Specifically, analyzing unstructured video data presents formidable methodological challenges that limit the use of multimedia data to generate marketing insights. To address this challenge, the authors propose a novel video feature framework based on machine learning and computer vision techniques, which helps marketers predict and understand the consumption of online video from a content-based perspective. The authors apply this framework to two unique data sets: one provided by MasterClass, consisting of 771 online videos and more than 2.6 million viewing records from 225,580 consumers, and another from Crash Course, consisting of 1,127 videos focusing on more traditional education disciplines. The analyses show that the framework proposed in this article can be used to accurately predict both individual-level consumer behavior and aggregate video popularity in these two very different contexts. The authors discuss how their findings and methods can be used to advance management and marketing research with unstructured video data in other contexts such as video marketing and entertainment analytics.
The new sharing economy model has introduced a dramatic, disruptive impact on the traditional industries by matching the demand and supply in real‐time. In this paper, we examine how the entry of Uber, a ride‐sharing services digital platform, brings new disruptive changes in public transportation operations. Significant debate has surrounded whether the new ride‐sharing model siphoned riders from public transit or made public transit feasible for more riders, but no consensus has been reached. One reason could be that the commonly used difference‐in‐differences empirical strategy fails to account for time‐varying unobserved confounders. To address this issue, we introduce a class of counterfactual estimators (CEs) to strengthen our causal identification and perform diagnostic tests to validate model assumptions for each CE. A significant drop in passenger trips with buses is found after Uber entry from both the conventional two‐way fixed effects model and the counterfactual estimators. Moreover, we provide empirical evidence that the entry of Uber has not significantly affected public demand response transportation, which indicates that Uber is not directly competing with other transportation services that aim to solve the “last mile problem.” Last, our additional analyses suggest that the effect of Uber entry is not uniform for different urban areas. All these empirical findings can be coherently explained in our framework of the substitution effect of Uber.
Given the increasing popularity of peer-to-peer (P2P) accommodation-sharing platforms (i.e., Airbnb), this paper aims to explore the effects of both picture color cues and textual cues related to color (TCC) on rental decisions made on these platforms in China. Based on cue utilization theory, this paper distinguishes picture color cues signaled from photos listed by hosts and TCC signaled from online guest reviews. The paper conducts an empirical study using data crawled from Xiaozhu.com. The research model is tested with poisson regression using 2520 valid listings collected in Guangzhou and is verified with additional data collected in Shanghai. Results show that both picture color cues and TCC affect rental transactions. Properties displayed with warm-colored pictures are more likely to be rented than those with cold-colored pictures. In addition, the consistency of picture color cues and TCC is found to have a significant impact on rental decisions. Interestingly, we also found an invert-U effect of brightness on rental purchase.
Online commerce, increasingly running on digital platforms, is subject to higher uncertainty and risk in transactions (e.g., fraud, opportunism) than offline commerce. To cope with this challenge, online commerce companies often use information systems to promote interactive communication between sellers and buyers. This study examines how such seller-buyer interactivity affects buyers’ purchase intention in an online commerce context of accommodation sharing. In particular, we use text mining techniques to analyze guest reviews and host responses on Airbnb. Our analysis suggests that not only the quantity but also the quality of the response messages matter: potential guests’ purchase intention increases with the relevance and richness of the host’s responses as well as their volume. Further, we find a notable nonlinearity in the quantity effect that the volume of host responses significantly affects guest purchase intention only when it is large enough. Making sufficient responses, the hosts could enjoy complementarity between the quantity and quality effects of their responses; otherwise, the relative richness of their responses becomes more important.
The sharing and gig economy are growing socio-economic phenomena. Increasing numbers of users are joining an ever-widening variety of sharing and gig platforms. In this paper, we explore the academic discussion on the sharing economy ecosystem from a cross-disciplinary perspective. We capture and focus on the early stages of the literature on these nascent research fields to identify and explain the themes that have emerged in the academic discussion. In doing so, we conduct a systematic and interdisciplinary review of the emerging literature on the sharing and gig economy and apply a horizon scanning approach to discuss the current state and future role of the sharing economy ecosystem. Our paper emphasizes that multi- and inter-disciplinary research efforts are becoming ever more important to (i) analytically address the growing diversity and complexity within the sharing economy ecosystem and (ii) gain a comprehensive understanding of the economics of sharing platforms and the competitive dynamics between different platform ecosystems.
Understanding consumers’ associations with brands is at the core of brand management. This is challenging since consumers can associate a brand with any number of objects, emotions, activities, sceneries, and concepts. This paper presents an elicitation platform, analysis methodology and results on consumer associations of US national brands. Our elicitation is direct, unaided, scalable, quantitative, and uses the power of visuals to depict a detailed representation of respondents’ relationships with a brand.
The proposed platform, Brand Visual Elicitation Platform (B-VEP), allows firms to collect online brand collages created by respondents and analyze them quantitatively to elicit brand associations. The authors use the platform to collect 4,743 collages for 303 large US brands from 1,851 respondents. Using unsupervised machine-learning and image-processing approaches, they analyze the collages and obtain a detailed set of associations for each brand, including objects (animals, food, people), constructs (abstract-art, horror, delicious, famous, fantasy), occupations (musician, bodybuilding, baking), nature (beach, misty, snowscape, wildlife), and institutions (corporate, army, school).
The authors demonstrate applications for brand management: obtaining prototypical brand visuals; relating associations to brand personality and equity; identifying favorable associations per category; exploring brand uniqueness through differentiating associations; and identifying commonalities between brands across categories for potential collaborations.
This research presents the use of machine learning analytics and metrics in the retailing context. We first discuss what is machine learning and explain the field’s origins. We then demonstrate the strengths of machine learning methods using an online retailing dataset, noting key areas of divergence from the traditional explanatory approach to data analysis. We then provide a review of the current state of machine learning in top-level retailing and marketing research, integrating ideas for future research and showcasing potential applications for practitioners. We propose that the explanatory and machine learning approaches need not be mutually exclusive. Particularly, we discuss four key areas in the general scientific research process that can benefit from machine learning: data exploration/theory building, variable creation, estimation, and predicting an outcome metric. Due to the customer-facing nature of retailing, we anticipate several challenges researchers and practitioners might face in the adoption and implementation of machine learning, such as ethical prediction and customer privacy issues. Overall, our belief is that machine learning can enhance customer experience and, accordingly, we advance opportunities for future research.
ResearchGate has not been able to resolve any references for this publication.