The growing popularity of online social networking has made it increasingly important to develop group recommender systems (RS) for delivering personalized services to the members of user groups. However, owing to the sparsity of data on group–item interactions (G–I interactions), existing group recommendation methods have concentrated on modeling user–item interactions (U–I interactions), which has limited the validity of the extracted group preferences. We propose a novel inter- and intra-view contrastive learning (I2VC) method for group recommendation, focusing on combining the direct view concerning group–item records and the indirect view concerning user–item records. The proposed method features a contrastive learning mechanism that incorporates two strategies (i.e., inter-view learning and intra-view learning) to overcome challenges in achieving the cross-view matching of the same group and the within-view discrimination among different groups. We empirically evaluate the proposed method using two real-world datasets. The results show that our method is more effective than other group recommendation methods. In addition, our findings show that the I2VC method is capable of boosting the alignment of strongly correlated group embeddings and the dispersion of weakly correlated ones, further demonstrating its effectiveness in view collaboration.

Livestream selling is an innovative form of online shopping that supports real-time interactions between streamers and consumers. However, a key challenge remains: Streamers have limited capacity to answer individual inquiries, whereas shoppers expect fast, personalized responses. This study investigates whether an AI-powered streaming assistant can address this tension by providing interactive, chat-based support to help consumers access and process information. Through a large-scale randomized field experiment on a leading livestream selling platform, we find that the AI assistant increases sales by 3.00% and reduces product return rates by 12.55%. Our analysis suggests that the AI assistant helps consumers feel more informed and confident in their purchases, thereby reducing uncertainty. At the same time, the AI assistant can occasionally disrupt the consumers’ livestream experience. Overall, the benefits of uncertainty reduction outweigh the negative influence of interruptions. For platform managers and policymakers, these findings evidence the potential of AI technology to enhance online commerce. The AI assistant is particularly effective for high-uncertainty products and for streamers with large audiences, offering implications for strategic deployment. Our research provides actionable insights for integrating AI into livestream selling and other digital commerce scenarios where real-time, AI-powered support can facilitate both consumer satisfaction and business growth.

Investors are increasingly relying on social media to seek insights into corporate prospects. However, it remains unclear whether social executives-those engaging with stakeholders through social media-provide valuable information that shapes investors' investment decisions, thereby influencing firm value. Drawing on emotions as social information theory, this study explores the impact of social executives' emotions, derived from social media posts, on firm value. Moreover, we consider variances in effects across different post types and firm sizes. Applying advanced cognitive analytics and deep learning techniques, our analysis reveals a significant association between the emotions of fear and anger expressed in posts related to firm events or routine work and firm value, with more pronounced effects observed in small firms. Additionally, our machine learning experiments demonstrate that social executives' emotions contribute to more accurate predictions of firm value than sentiments alone. These findings have important implications for both theory and practice.

Online learners often experience a lack of sustained motivation given the self-paced nature of online learning, resulting in inefficiency and a high dropout rate. It is important to explore options that help users optimize their learning behavior and improve their learning performance. Using a multimethod approach, we show that (a) starting learning sessions at on-the-hour time points activates users’ implemental mindset, which supports them in building greater learning persistence and achieving better learning performance, and (b) social presence significantly attenuates the effects of on-the-hour time points in online learning. Based on our findings, we suggest that both learners and instructors on online learning platforms can leverage common temporal cues, such as on-the-hour time points, to schedule learning activities in order to motivate online learners, enhance their learning persistence, and improve their learning performance. Additionally, online learning platforms can also adopt designs that facilitate virtual connections among geographically separated users to enhance their learning productivity.

Recommender systems are widely used by platforms/merchants to find the products that are likely to interest consumers. However, existing dynamic methods still face challenges with regard to diverse behaviors, variability in interest shifts, and the identification of psychological dynamics. Premised on the marketing funnel perspective to analyze consumer shopping journeys, this study proposes a novel and effective machine learning approach for product recommendation, namely, multi-stage dynamic Bayesian network (MS-DBN), which models the generative processes of consumers’ interactive behaviors with products in light of their stage transitions and interest shifts. In this way, consumers’ stage-interest-behavior dynamics can be learnt, especially the variability in interest shifts. This provides managerial implications for practice. MS-DBN demonstrates significant performance advantage with general applicability by extracting the generalizable regularity during shopping journeys, which compensates the diversity and sparsity frequently observed in consumer behaviors. In addition, aided by the identification strategies integrated into the learning process, the latent variables in the model can be detected such that consumers’ invisible psychological stages and interests in products can be identified from their observed behaviors, shedding light on the targeted marketing of platforms/merchants and thus enriching the practical value of the approach.

Purpose Entrepreneurs and individual sellers heavily leverage their social ties embedded in social media, expressive or instrumental, to penetrate the market and achieve business success. However, the extant social commerce literature offers limited understanding on how different forms of buyer−seller social ties embedded in social media affect buyers' purchase behaviors. The study draws on the theoretical lens of social ties and proposes an integrative theoretical framework to understand the direct and indirect influences of expressive and instrumental ties (ExTSM and InTSM) between buyers and sellers on buyers' purchase intention (PI) in social commerce. Design/methodology/approach The authors first validated the measures of ExTSM and InTSM with survey data from 166 Weibo commerce buyers. They then tested their theoretical framework and hypotheses with survey data from 246 buyer−seller dyads in WeChat commerce. Findings With a buyer-centric view, (1) ExTSM and InTSM, respectively, had a direct negative and a positive influence on PI; (2) both trust and perceived product value displayed inconsistent mediation effects on the negative relationship between ExTSM and PI; and (3) only perceived product value mediated the positive influence of InTSM on PI. From sellers' viewpoint, (1) their ExTSM and InTSM with buyers were mixed up, and (2) the mingled social ties negatively impacted buyers' purchase intention. Originality/value The findings of the study advance the theoretical understanding of social commerce and offer practical guidance for small and medium-sized enterprises to effectively utilize social media for business purposes.

Purpose It has become increasingly clear that the objectives of privacy and competition policy are in conflict with one another with regard to platform data. While privacy policies aim at limiting the use of platform data for purposes other than those for which the data were collected in order to protect the privacy of platform users, competition policy aims at making such data widely available in order to curb the power of platforms. Design/methodology/approach We draw on Commons' Institutional Economics to contrast the current control-based approaches to ensuring the protection as well as the sharing of platform data with an ownership approach. We also propose the novel category of platform use data and contrast this with the dichotomy of personal/non-personal data which underlies current regulatory initiatives. Findings We find that current control- and ownership-based approaches are ineffective with regard to their capacity to balance these conflicting objectives and propose an alternative approach which makes platform data saleable. We discuss this approach in view of its capacity to balance the conflicting objectives of privacy and competition policy and its effectiveness in supporting each separately. Originality/value Our approach clarifies the fundamental difference between data markets and other concepts such as data exchanges.

Preference prediction is the building block of personalized services, and its implementation at the group level helps enterprises identify their target customers effectively. Existing methods for preference prediction mainly focus on behavioral interactions to extract the associations between groups and products, ignoring the importance of other auxiliary records (e.g., online reviews and social tags) in association detection. This paper proposes a novel method named GMAT for group preference prediction, aiming to collectively detect the sophisticated association patterns from user generated content (UGC) and behavioral interactions. In doing so, we construct a tripartite graph to collaborate these two types of data, and design a deep-learning algorithm with mutual attention module for generating the contextualized representations of groups and products. Extensive experiments on two real-world datasets show that GMAT is superior to other baselines in terms of group preference prediction. Additionally, GMAT is able to improve prediction accuracy compared with its different variants, further verifying the proposed method’s effectiveness on association pattern detection.

The abundance of multiple types of consumer digital footprints recorded on e-commerce platforms has fueled the design of personalized recommender systems for decision support. However, capturing consumers' inherent preferences for effective recommendations based on consumer digital footprints can be challenging because of the multitude of factors driving consumer behaviors. Model training and recommendation outcomes may become biased if other factors are inappropriately recognized as consumers' inherent preferences in the learning process. Drawing on consumer behavior theories, we tease out various factors that drive consumers' digital footprints at different consumption stages. We develop a novel recommendation approach, namely, DISC (Disentangling consumers' Inherent preferences, item Salience effect, and Conformity effect), which leverages disentangled representation learning with a causal graph to derive the effect of each factor driving consumer behaviors. This approach provides personalized and interpretable recommendations based on the inference of consumers' normative inherent preferences. The DISC model's identifiability is demonstrated through theoretical analysis, enabling rigorous causal inference based on observational data. To evaluate DISC's performance, extensive experiments are conducted on real-world data sets with a carefully designed protocol. The results reveal that DISC outperforms state-of-the-art baselines significantly and possesses good interpretability. Moreover, we illustrate the potential impact of different marketing strategies' by intervening on the disentangled causes through follow-up counterfactual analyses based on the causal graph. Our study contributes to the literature and practice by causally unpacking the behavioral mechanism behind consumers' digital footprints and designing an interpretable personalized recommendation approach anchored in their inherent preferences.

It becomes increasingly clear that the objectives of privacy and competition policy are in conflict with one another with regard to platform data. While privacy policies aim at limiting the use of platform data for purposes other than those for which the data were collected in order to protect the privacy of platform users, competition policy aims at making such data widely available in order to curb the power of platforms. We find that current controland ownership-based approaches are ineffective with regard to their capacity to balance these conflicting objectives and propose an institutional approach which makes platform data saleable. We discuss this approach in view of its capacity to balance the conflicting objectives of privacy and competition policy and with regard to its effectiveness in supporting each separately. Our approach also clarifies the fundamental difference between data markets and other concepts such as data spaces.

Livestream technology is increasingly transforming consumers’ online shopping experience, as it enables streamers to perform real-time product presentation while interacting with a large number of consumers to sell those products. In this research, we address how livestream selling platforms can potentially mitigate the tension between streamers’ constrained service capacity and individual service demands with algorithm-based assistants (termed “AI assistants”). In partnership with the world’s largest online shopping platform, we report a large-scale field experiment wherein the consumers in the treatment group have access to an AI assistant that predicts consumers’ potential needs and provides individualized services, while the consumers in the control group do not have access to such an AI assistant. By analyzing data from 132,199 consumers, we find that the introduction of the AI assistant increases sales by 2.61% and reduces product returns by 62.86%. More interestingly, with granular clickstream data, our models on multiple-stage purchase decision-making reveal that an AI assistant increases the duration of the awareness and consideration stages, improves the probability of placing an order in the evaluation stage and, more importantly, reduces the likelihood of product returns (given that the order is placed) at the post-purchase stage. Further, our analyses of consumer comments reveal that interacting with the AI assistant also reduces consumers’ expressions of affection and positive emotions. We further indicate that interaction with an AI assistant triggers consumers to make fewer emotion-driven decisions, which likely leads to high-quality purchase decisions. Moreover, our heterogeneous treatment effects analyses reveal that an AI assistant only exhibits significant effects in selling products that have a higher level of product uncertainty. Overall, our findings provide actionable implications for online shopping platforms in designing and implementing AI artifacts.

Voice‐based artificial intelligence (AI) systems have been recently deployed to replace traditional interactive voice response (IVR) systems in call center customer service. However, there is little evidence that sheds light on how the implementation of AI systems impacts customer behavior, as well as AI systems’ effects on call center customer service performance. By leveraging the proprietary data obtained from a natural field experiment in a large telecommunication company, we examine how the introduction of a voice‐based AI system affects call length, customers’ demand for human service, and customer complaints in call center customer service. We find that the implementation of the AI system temporarily increases the duration of machine service and customers’ demand for human service; however, it persistently reduces customer complaints. Furthermore, our results reveal interesting heterogeneity in the effectiveness of the voice‐based AI system. For relatively simple service requests, the AI system reduces customer complaints for both experienced and inexperienced customers. However, for complex requests, customers appear to learn from the prior experience of interacting with the AI system, which leads to fewer complaints. Moreover, the AI‐based system has a significantly larger effect on reducing customer complaints for older and female customers as well as for customers who have had extensive experience using the IVR system. Finally, we find that speech‐recognition failures in customer‐AI interactions lead to increases in customers’ demand for human service and customer complaints. The results from this study provide implications for the implementation of an AI system in call center operations.

Research on recommender systems has noted that the ranking of recommended items may play an important role in the performance of recommendation algorithms. To advance recommender systems research beyond the traditional approach that ranks recommended products in descending, it is crucial to understand the cognitive processes that online consumers experience when they evaluate products in a sequence. Drawing on evaluability theory and the order effects perspective, we formulate a scenario in which two products are presented sequentially and each product has two attributes, one of which can be evaluated independently while the other is difficult to evaluate without comparison. Analyses show that in two out of the three cases examined, presenting the most recommended product in the second place will result in stronger consumer purchase intentions and willingness to pay. Research hypotheses are proposed based on the results of the scenario analyses and are empirically tested through three laboratory experiments. In Study 1, evidence for the hypothesized order effects is found for the settings with randomly assigned product recommendations. In Study 2, the same effects are observed for the settings with personalized recommendations generated by a collaborative filtering algorithm. In Study 3, it is shown that such order effects also exist in terms of the recommendation strength of recommender systems. These findings provide novel insights into the behavioral implications of using recommender systems in e-commerce, shedding light on additional means of improving the design of such systems.

Current online review systems widely suffer from rating biases. Biased ratings can lead to violations of customer trust and failures of business intelligence. Hence, both practitioners and researchers have directed massive efforts toward curbing rating biases. In this paper, we investigate bandwagon bias, the rating distortion resulting from individuals posting ratings shifted toward the displayed average rating, and propose a bias warning approach to mitigate this bias. Drawing on the flexible correction model, the theory of valuation in behavioral economics, and previous warning research, we design an effective warning strategy in two steps. First, we start with the risk-alert warning strategy, which prior research has widely employed, and rationalize its deficiencies by synthesizing theoretical analysis and extant empirical evidence. Second, considering the deficiencies, we identify a supplementary content design factor—the ranking task—and construct a risk-alert-with-ranking-task warning strategy. We then empirically test the effects of the two warning strategies on individual ratings in cases in which bandwagon bias either occurs or does not occur in individuals’ initial assessments. The results of four controlled experiments indicate that (1) the risk-alert strategy can reduce bandwagon bias in individual ratings but will elicit unwanted rating distortions when bandwagon bias does not occur in individuals’ initial assessments, and (2) the risk-alert-with-ranking-task strategy can mitigate bandwagon bias while avoiding the unwanted rating distortions above and can thus function as an effective warning strategy. Our research contributes to the literature by proposing an effective debiasing solution for bandwagon bias and a bias warning approach for online rating debiasing, which can help increase rating informativeness on online platforms.