Recommendation Agents Research Articles

Towards multi-agent system for learning object recommendation

The rapid increase of online educational content has made it harder for students to find specific information. E-learning recommender systems help students easily find the learning objects they require, improving the learning experience. The effectiveness of these systems is further improved by integrating deep learning with multi-agent systems. Multi-agent systems facilitate adaptable interactions within the system's various parts, and deep learning processes extensive data to understand learners' preferences. This collaboration results in custom-made suggestions that cater to individual learners. Our research introduces a multi-agent system tailored for suggesting learning objects in line with learners' knowledge levels and learning styles. This system uniquely comprises four agents: the learner agent, the tutor agent, the learning object agent, and the recommendation agent. It applies the Felder and Silverman model to pinpoint various student learning styles and organizes educational content based on the newest IEEE Learning Object Metadata standard. The system uses advanced techniques, such as Convolutional Neural Networks (CNN) and Multilayer Perceptrons (MLP), to propose learning objects. In terms of creating personalized learning experiences, this system is a considerable step forward. It effectively suggests learning objects that closely match each learner's personal profile, greatly enhancing student engagement and making the learning process more efficient.

After opening the black box: Meta-dehumanization matters in algorithm recommendation aversion

Perceptions of algorithms as opaque, commonly referred to as the black box problem, can make people reluctant to accept a recommendation from an algorithm rather than a human. Interventions that enhance people's subjective understanding of algorithms have been shown to reduce this aversion. However, across four preregistered studies (N = 960), we found that in the online shopping context, after explaining the algorithm recommendation process (versus human recommendation), users felt dehumanized and thus averse to algorithms (Study 1). This effect persisted, regardless of the type of algorithm (i.e., conventional algorithms or large language models; Study 2) or recommended product (i.e., search or experience products; Study 3). Notably, considering large language models (versus conventional algorithms) as the recommendation agent (Study 2) and framing algorithm recommendation as consumer-serving (versus website-serving; Study 4) mitigated algorithm aversion caused by meta-dehumanization. Our findings contribute to ongoing discussions on algorithm transparency, enrich the literature on human–algorithm interaction, and provide practical insights for encouraging algorithm adoption.

Self-Supervised Bot Play for Transcript-Free Conversational Critiquing with Rationales

Conversational critiquing in recommender systems offers a way for users to engage in multi-turn conversations to find items they enjoy. For users to trust an agent and give effective feedback, the recommender system must be able to explain its suggestions and rationales. We develop a two-part framework for training multi-turn conversational critiquing in recommender systems that provide recommendation rationales that users can effectively interact with to receive better recommendations. First, we train a recommender system to jointly suggest items and explain its reasoning via subjective rationales. We then fine-tune this model to incorporate iterative user feedback via self-supervised bot-play. Experiments on three real-world datasets demonstrate that our system can be applied to different recommendation models across diverse domains to achieve state-of-the-art performance in multi-turn recommendation. Human studies show that systems trained with our framework provide more useful, helpful, and knowledgeable suggestions in warm- and cold-start settings. Our framework allows us to use only product reviews during training, avoiding the need for expensive dialog transcript datasets that limit the applicability of previous conversational recommender agents.

M 3 Rec: A Context-aware Offline Meta-level Model-based Reinforcement Learning Approach for Cold-Start Recommendation

Reinforcement learning (RL) has shown great promise in optimizing long-term user interest in recommender systems. However, existing RL-based recommendation methods need a large number of interactions for each user to learn the recommendation policy. The challenge becomes more critical when recommending to new users who have a limited number of interactions. To that end, in this paper, we address the cold-start challenge in the RL-based recommender systems by proposing a novel context-aware offline meta-level model-based reinforcement learning approach for user adaptation. Our proposed approach learns to infer each user's preference with a user context variable that enables recommendation systems to better adapt to new users with limited contextual information. To improve adaptation efficiency, our approach learns to recover the user choice function and reward from limited contextual information through an inverse reinforcement learning method, which is used to assist the training of a meta-level recommendation agent. To avoid the need for online interaction, the proposed method is trained using historically collected offline data. Moreover, to tackle the challenge of offline policy training, we introduce a mutual information constraint between the user model and recommendation agent. Evaluation results show the superiority of our developed offline policy learning method when adapting to new users with limited contextual information. In addition, we provide a theoretical analysis of the recommendation performance bound.

Integrating object detection and natural language processing models to build a personalized attraction recommendation agent in a smart product service system

Product Service System (PSS) is a new business model that integrates tangible products and intangible services to better meet customer needs and expectations. In recent years, scholars had some efforts to enhance the capability of PSS with the support of artificial intelligence (AI) which is known as Smart PSS(SPSS). So far, most previous studies adopted a single model which cannot handle multiple tasks simultaneously that results in unsatisfactory services to customers. In addition, customer preference cannot be fully addressed to achieve personalization. Therefore, this study proposes a method that integrates multiple models into an SPSS with three steps: (1) Collect data and construct an appropriate object detection model. (2) Develop smart PSS solutions. Then, (3) Optimize the system based on feedback through Natural Language Processing (NLP) to provide customers with personalized services. An attraction recommendation case study with experiment is designed to verify the proposed method. The results show that proposed SPSS can optimize the system in time according to the feedback of users, and provide better personalized services. This research is the first research that utilizes applied both text and image data to extract customer characteristics to better capture the voice of customers.

Consumer reactions to technology in retail: choice uncertainty and reduced perceived control in decisions assisted by recommendation agents

The emergence of artificial intelligence technologies, such as recommendation agents, presents new challenges and opportunities for marketing. Recommendation agents assist consumers in their online grocery shopping decisions by analyzing data on preferences and behaviors. This research highlights that while recommendation agents can reduce choice overload and make purchase decisions easier for consumers, they are also associated with higher uncertainty in decision-making. Three experimental studies confirmed that purchases aided by recommendation agents are perceived as more uncertain and reduced perceptions of control over the choices explain this outcome. Furthermore, lower choice satisfaction and purchase intentions are confirmed as consequences of perceived uncertainty. Personal characteristics such as risk aversion and maximization tendencies are considered boundary conditions for these effects.

Comparing the effectiveness of recommendation agents across devices

This research empirically studies whether recommendation agents are as effective on newer, mobile devices (i.e., tablets, smartphones) as they are on older, stationary ones (i.e., desktop computers). We analyze clickstream data from Airbnb with a novel econometric model based on beta and logit regressions and estimated within the Bayesian framework. The model controls for self-selectivity bias. Our empirical findings show that recommendation agents are less effective on mobile devices; the number of recommended alternatives clicked by smartphone (desktop computer) users is smaller (larger) than that of tablet users. This is managerially important as we also show that a consumer’s purchase likelihood is directly related to the number of recommended alternatives evaluated by them. Furthermore, we found that men and younger consumers rely less on recommendation agents. Our results highlight the importance of redesigning recommendation agents for mobile devices as well as identifying consumer segments that need stronger incentives to shop online.

Optimal delegation and information transmission under limited awareness

We study the delegation problem between a principal and an agent, who not only has better information about the performance of the available actions but also superior awareness of the set of actions that are actually feasible. We provide conditions under which the agent finds it optimal to leave the principal unaware of relevant options. By doing so, the agent increases the principal's cost of distorting the agent's choices and increases the principal's willingness to grant him higher information rents. We further show that the principal may use the option of renegotiation as a tool to implement actions that are not describable to her at the contracting stage. If the agent renegotiates, his proposal signals information about the payoff state. Due to her limited awareness, the principal makes a coarse inference from the agent's recommendations and, as a result, accepts a large number of the agent's proposals, which ultimately benefits both.

How Online Consumers Use Multiple Advice Sources: An Empirical Exploration Using Verbal Protocol Analysis

With more online stores providing recommendations and reviews simultaneously from multiple advice sources—such as recommendation agents, other consumers, and experts—consumers are facing the challenge of deciding how to use wide-ranging and possibly conflicting sets of information to improve their product selection. Although previous studies have investigated the classical decision-making strategies used in preferential choice problems, most of these are not directly applicable to multiple advice source environments. In addition, since most of these studies have mainly focused on a variance model rather than a process model, they cannot fully explain how decision makers reach their product selection decisions. To shed light on the processes that online consumers use in making product selections when using multiple advice sources, this study: (1) explores if, how, and when consumers use consistency as a part of their decision-making strategies through developing process models; (2) identifies the consistency strategies utilized when using multiple advice sources; and (3) proposes a new consistency-based decision-making model. Through concurrent verbal protocol analysis, we identified four consistency strategies and found that the use of consistency strategies increases decision quality more than traditional nonconsistency strategies. Our findings are triangulated through the theoretical lens of cognitive dissonance theory, information search process model, and confirmation bias for rigorous validation. We also describe the theoretical and practical implications of our findings.

Effect of Recommendation Agent on On-Line Consumer Unplanned Purchase Behavior through Loyalty

Effect of Recommendation Agent on On-Line Consumer Unplanned Purchase Behavior through Loyalty

When Post Hoc Explanation Knocks: Consumer Responses to Explainable AI Recommendations

Artificial intelligence (AI) recommendations are becoming increasingly prevalent, but consumers are often reluctant to trust them, in part due to the “black-box” nature of algorithm-facilitated recommendation agents. Despite the acknowledgment of the vital role of interpretability in consumer trust in AI recommendations, it remains unclear how to effectively increase interpretability perceptions and consequently enhance positive consumer responses. The current research addresses this issue by investigating the effects of the presence and type of post hoc explanations in boosting positive consumer responses to AI recommendations in different decision-making domains. Across four studies, the authors demonstrate that the presence of post hoc explanations increases interpretability perceptions, which in turn fosters positive consumer responses (e.g., trust, purchase intention, and click-through) to AI recommendations. Moreover, they show that the facilitating effect of post hoc explanations is stronger in the utilitarian (vs. hedonic) decision-making domain. Further, explanation type modulates the effectiveness of post hoc explanations such that attribute-based explanations are more effective in enhancing trust in the utilitarian decision-making domain, whereas user-based explanations are more effective in the hedonic decision-making domain.

Learning From Atypical Behavior: Temporary Interest Aware Recommendation Based on Reinforcement Learning

Traditional robust recommendation methods view atypical user-item interactions as noise and aim to reduce their impact with some kind of noise filtering technique, which often suffers from two challenges. First, in real world, atypical interactions may signal users' temporary interest different from their general preference. Therefore, simply filtering out the atypical interactions as noise may be inappropriate and degrade the personalization of recommendations. Second, it is hard to acquire the temporary interest since there are no explicit supervision signals to indicate whether an interaction is atypical or not. To address this challenges, we propose a novel model called Temporary Interest Aware Recommendation (TIARec), which can distinguish atypical interactions from normal ones without supervision and capture the temporary interest as well as the general preference of users. Particularly, we propose a reinforcement learning framework containing a recommender agent and an auxiliary classifier agent, which are jointly trained with the objective of maximizing the cumulative return of the recommendations made by the recommender agent. During the joint training process, the classifier agent can judge whether the interaction with an item recommended by the recommender agent is atypical, and the knowledge about learning temporary interest from atypical interactions can be transferred to the recommender agent, which makes the recommender agent able to alone make recommendations that balance the general preference and temporary interest of users. At last, the experiments conducted on real world datasets verify the effectiveness of TIARec.

AIRA: An Intelligent Recommendation Agent Application for Movies

An intelligent Recommendation App has been developed to assist caregivers. This project's primary objective is to assist parents in determining whether a particular movie/cartoon/drama is adequate for their children by providing ratings that will assist them in identifying age-appropriate content. This application will provide reliable evaluations, reviews, and recommendations to parents. Each rating and review are based on fundamental, essential child development principles. Intelligent Recommendation Agent aids families in making intelligent media selections. It provides the most extensive and reliable database of learning ratings, age recommendations, and content evaluations for films, television series, and dramas. In addition, there will be a list of abusive words from the content with its subtitles so that parents can identify appropriate content for children. By limiting their child's exposure to violent acts, parents can play a positive role in their child's life by using this application. Movies with positive role models can also have a positive effect on children.

Testing the performance of online recommendation agents: A meta-analysis

Many retailers (e.g., Amazon, Walmart) use various types of online recommendation agents (RAs) on their websites to suggest goods and services to consumers. These RAs screen millions of options to ease consumers’ information search and evaluation. To determine which RA types best support consumers’ efforts, the present research reports a meta-analysis of perceived recommendation quality research, a key performance metric that gauges RAs from consumers’ perspectives. To test the framework derived from this meta-analysis, the authors rely on data gathered from 32,172 consumers, reported in 122 samples. The results affirm that some RAs perform better than others in leveraging the effects of perceived recommendation quality on consumers’ decision-making satisfaction, RA satisfaction, and intention to use the RA in the future. The best performing RAs feature specific algorithms (i.e., collaborative filtering, interactive RAs, and self-serving recommendations), recommendation presentations (i.e., solicited recommendation), and data sources (i.e., location-based and social network–based RAs). Moreover, the results suggest that some RAs perform better than others in leveraging the effects of decision-making and RA satisfaction on future use intentions. These insights advance RA theory and provide guidance for managers, with regard to choosing the optimal RA.

When Do Humans Heed AI Agents' Advice? When Should They?

We manipulate the presence, skill, and display of artificial intelligence (AI) recommendations in a strategy game to measure their effect on users' performance. Many applications of AI require humans and AI agents to make decisions collaboratively. Success depends on how appropriately humans rely on the AI agent. We demonstrate an evaluation method for a platform that uses neural network agents of varying skill levels for the simple strategic game of Connect Four. We report results from a 2 × 3 between-subjects factorial experiment that varies the format of AI recommendations (categorical or probabilistic) and the AI agent's amount of training (low, medium, or high). On each round of 10 games, participants proposed a move, saw the AI agent's recommendations, and then moved. Participants' performance improved with a highly skilled agent, but quickly plateaued, as they relied uncritically on the agent. Participants relied too little on lower skilled agents. The display format had no effect on users' skill or choices. The value of these AI agents depended on their skill level and users' ability to extract lessons from their advice. Organizations employing AI decision support systems must consider behavioral aspects of the human-agent team. We demonstrate an approach to evaluating competing designs and assessing their performance.

Decisions with ChatGPT: Reexamining choice overload in ChatGPT recommendations

This research examines how individuals respond differently to recommendation options generated by ChatGPT, an AI-powered language model, in five studies. In contrast to previous research on choice overload, Studies 1 and 2 demonstrate that people tend to respond positively to a large number of recommendation options (60 options), revealing diverse consumer perceptions of AI-generated recommendations. Studies 3 and 4 further illustrate the moderating effect of recommendation agents and indicate that choice overload elicits distinct patterns of consumer reactions depending on whether the recommendations are from a human or AI agent. Lastly, Study 5 directly measures consumer preferences for recommendation agents, revealing a general preference for ChatGPT, particularly when a large number of options are available. These findings have significant implications for recommendation system design and user preferences regarding AI-powered recommendations.

Online Impulsive Buying Behavior: The Mediating Effect of Browsing on Egyptian Consumers

This paper aims to investigate the joint effects of three types of external stimuli (social networking, marketing, and website cues) on consumers' browsing type and online impulsive buying behavior in the Egyptian retailing industry. As will be mentioned in detail, it is observed that the scant preceding research on online impulse purchases emphasizes the critical function of social influence in Egypt. The proposed objectives were achieved using a quantitative research approach using an online questionnaire. The study's 384 sample cases included18-40-year-old old Egyptian online buyers living in Cairo. That was calculated utilizing formulas covered in chapters one and three (Population and sampling). Shoppers who exhibit utilitarian browsing have a bigger impact on the temptation to make impulsive purchases. Recommender agents, sales promotions, and customer reviews positively impact consumers' browsing types, which ultimately influence their impulse buying behavior. According to this study's findings, consumers are greatly influenced by recommendation agents while browsing, whether it’s utilitarian or hedonic browsing value, more than other independent variables. Comparing the two types of browsing, recommendation place a greater emphasis on the hedonic value of browsing. Consumers who consider customer reviews place a greater emphasis on hedonic browsing, while those with sales promotions place a greater emphasis on utilitarian value.

Improving Recommendation for Effective Personalization in Context-Aware Data Using Novel Neural Network

The digital technologies that run based on users’ content provide a platform for users to help air their opinions on various aspects of a particular subject or product. The recommendation agents play a crucial role in personalizing the needs of individual users. Therefore, it is essential to improve the user experience. The recommender system focuses on recommending a set of items to a user to help the decision-making process and is prevalent across e-commerce and media websites. In Context-Aware Recommender Systems (CARS), several influential and contextual variables are identified to provide an effective recommendation. A substantial trade-off is applied in context to achieve the proper accuracy and coverage required for a collaborative recommendation. The CARS will generate more recommendations utilizing adapting them to a certain contextual situation of users. However, the key issue is how contextual information is used to create good and intelligent recommender systems. This paper proposes an Artificial Neural Network (ANN) to achieve contextual recommendations based on user-generated reviews. The ability of ANNs to learn events and make decisions based on similar events makes it effective for personalized recommendations in CARS. Thus, the most appropriate contexts in which a user should choose an item or service are achieved. This work converts every label set into a Multi-Label Classification (MLC) problem to enhance recommendations. Experimental results show that the proposed ANN performs better in the Binary Relevance (BR) Instance-Based Classifier, the BR Decision Tree, and the Multi-label SVM for Trip Advisor and LDOS-CoMoDa Dataset. Furthermore, the accuracy of the proposed ANN achieves better results by 1.1% to 6.1% compared to other existing methods.

Recommendation agents: an analysis of consumers’ risk perceptions toward artificial intelligence

Recommendation agents: an analysis of consumers’ risk perceptions toward artificial intelligence

The impact of multi-criteria ratings in social networking sites on the performance of online recommendation agents

Recommender Systems (RSs) have played an important role in online retailing portals and customers’ decision-making processes. Recommender systems that are based on the conventional Collaborative Filtering (CF) approach rely on single customers’ ratings on retailing websites. Multi-criteria CF (MCCF) approaches that rely on multi-aspects of the products have provided more reliable and effective recommendations on retailing websites. However, these approaches should be improved in terms of accuracy by solving sparsity issues and incorporating criteria ratings. In addition, most of the recommendation agents that are based on MCCF cannot learn automatically from the features of the products to model customers’ preferences and generate accurate recommendations on retailing websites. Besides, although previous studies have utilized single and multi-criteria ratings in recommendation agents of tourism websites, still, if there is a lack of ratings of items, most of these systems will fail to generate accurate recommendations to users. In this research, we develop a new recommendation agent based on a MCCF approach to effectively improve the performance of previous recommendation systems for tourism websites. The results demonstrated that the method can predict the most relevant products to users, particularly when the dataset is sparse.