Modeling decision-making outside of controlled environments requires accounting for asynchronous, exogenous signals, such as notifications or algorithmic feeds, that dynamically alter user response times. Standard Drift-Diffusion Models (DDM) become analytically intractable when drift rates vary continuously with time. In this paper, we derive a closed-form analytical approximation for the first-passage time distribution of a single-boundary DDM with time-dependent drift, valid in the high-threshold regime. The main result allows us to analytically study the optimal timing of external signals to maximize the probability of a user response within our approximation framework. To evaluate our response time model, we conduct an extensive empirical comparison with state-of-the-art methods for user watch-time prediction and evaluation in simulated environments.

Modeling the dynamic evolution of user interests from sequential interaction data is a fundamental challenge in knowledge discovery and recommender systems. While exploring user preferences is crucial for improving experiences in recommender systems, it requires systematically modeling the sequential evolution of knowledge regarding individual user preferences. To that end, it is important to provide sufficient support for temporal dynamics and personalization when selecting and recommending appropriate exploratory content. In addition, it is beneficial to capture the long-term impact, besides immediate benefits, of exploratory items on future user behaviors. However, traditional exploration-oriented models struggle to capture the continuity of these interactions, suffering from feedback-loop biases that limit the system's ability to discover new knowledge. To address this research gap, we formulate recommendations as a user trajectory learning task in this paper, and propose Forward-looking Dynamical System Recommender (FDSR), a new exploratory recommendation method that determines these trajectory points based on dynamical system theory, where we utilize Ordinary Differential Equations (ODEs) to model the latent state transitions of the user preference. By numerically solving dynamical system equations designed to capture the user's desire for exploration, we produce the next point on the evolution trajectory and recommend items based on this predicted point. Theoretically, this formulation allows us to analyze the stability of interest evolution, providing a robust framework for handling sparse temporal data. Extensive offline experiments on three real-world datasets demonstrate significant performance improvements of FDSR over state-of-the-art baselines for both exploration and exploitation metrics. A simulation study further illustrates its long-term benefits across multiple rounds of interactions, and a large-scale online A/B test at a leading video streaming platform shows that the FDSR method significantly outperforms the latest production model. As a result, it has since been deployed in production.

The present work proposes a Deep Learning architecture for the prediction of various consumer choice behaviors from time series of raw gaze or eye fixations on images of the decision environment, for which currently no foundational models are available. The architecture, called STARE (Spatio-Temporal Attention Representation for Eye Tracking), uses a new tokenization strategy, which involves mapping the x- and y- pixel coordinates of eye-movement time series on predefined, contiguous Regions of Interest. That tokenization makes the spatio-temporal eye-movement data available to the Chronos, a time-series foundation model based on the T5 architecture, to which co-attention and/or cross-attention is added to capture directional and/or interocular influences of eye movements. We compare STARE with several state-of-the art alternatives on multiple datasets with the purpose of predicting consumer choice behaviors from eye movements. We thus make a first step towards developing and testing DL architectures that represent visual attention dynamics rooted in the neurophysiology of eye movements.

LLMs are becoming increasingly capable and widespread. Consequently, the potential and reality of their misuse is also growing. In this work, we address the problem of detecting LLM-generated text that is not explicitly declared as such. We present a novel, general-purpose, and supervised LLM text detector, SElected-Next-Token tRAnsformer (SENTRA). SENTRA is a Transformer-based encoder leveraging selected-next-token-probability sequences and utilizing contrastive pre-training on large amounts of unlabeled data. Our experiments on three popular public datasets across 24 domains of text demonstrate SENTRA is a general-purpose classifier that significantly outperforms popular baselines in the out-of-domain setting.

Context has been an important topic in recommender systems over the past two decades. Most of the prior CARS papers manually selected and considered only a few crucial contextual variables in an application, such as time, location, and company of a person. This prior work demonstrated significant recommendation performance improvements when various CARS-based methods have been deployed in numerous applications. In this paper, we study “context-rich” applications dealing with a large variety of different types of contexts. We demonstrate that supporting only a few of the most important contextual variables that could be manually identified, although useful, is not sufficient. In particular, we develop an approach to extract a large number of contextual variables for the dialogue-based recommender systems. In our study, we processed dialogues of bank managers with their clients and managed to identify over two hundred types of contextual variables forming the Long Tail of Context (LTC). We empirically demonstrate that LTC matters, and using all these contextual variables from the Long Tail leads to better recommendation performance.

Static user and item features constitute important information to be taken into account in the recommendation process. However, as these features are usually sparse and of large-vocabulary, existing deep learning-based methods typically construct large tables of high-dimensional feature embeddings, which is inefficient in terms of memory storage and is computationally problematic. On the other hand, while product quantization-based methods have been proposed to compress latent embeddings, they usually come at the cost of compromising recommendation performance due to the restrictive expressive power, as feature correlations and user-item interactions are not properly captured in the compression process. To address these issues, we propose a novel Dual Contrastive Learning method to generate low-dimensional discrete static feature representations that significantly reduce memory storage and computational complexity, while simultaneously producing superior recommendation performance. Extensive offline experiments on three large-scale industrial datasets demonstrate that our proposed model significantly outperforms the selected baselines. In addition, we conducted an online A/B test at Alibaba and show that the proposed model significantly improves the average video streaming time, while reducing the size of the feature embedding table by 90% over the currently deployed system.

In this paper, we study the consumers’ variety-seeking behavior in recommender system applications and propose a comprehensive framework to measure such behavior based on past consumption records. The effectiveness of the proposed framework is validated through user questionnaire studies conducted at Alibaba, where our constructed variety-seeking measures match well with consumers’ self-reported levels of their variety-seeking behaviors. We subsequently present a recommendation framework that combines the identified variety-seeking levels with unexpected recommender systems in the data mining literature to address consumers’ heterogenous desire for product variety, in which we provide more unexpected product recommendations to variety-seeking consumers and vice versa. Through off-line experiments on three different recommendation scenarios and a large-scale online controlled experiment at a major video-streaming platform, we demonstrate that those models following our recommendation framework significantly increase various business performance metrics and generate tangible economic impact for the company. Our findings lead to important managerial implications to better understand consumers’ variety-seeking behaviors and design recommender systems. As a result, the best performing model in our proposed frameworks is deployed by the company to serve all consumers on the video-streaming platform.

Optimizing multiple objectives simultaneously is an important task for recommendation platforms to improve their performance. However, this task is particularly challenging since the relationships between different objectives are heterogeneous across different consumers and dynamically fluctuating according to different contexts. Especially in those cases when objectives become conflicting with each other, the result of recommendations will form a pareto-frontier, where the improvements of any objective comes at the cost of a performance decrease of another objective. Existing multi-objective recommender systems do not systematically consider such dynamic relationships; instead, they balance between these objectives in a static and uniform manner, resulting in only suboptimal multi-objective recommendation performance. In this paper, we propose a Deep Pareto Reinforcement Learning (DeepPRL) approach, where we (1) comprehensively model the complex relationships between multiple objectives in recommendations; (2) effectively capture personalized and contextual consumer preference for each objective to provide better recommendations; (3) optimize both the short-term and the long-term performance of multi-objective recommendations. As a result, our method achieves significant pareto-dominance over the state-of-the-art baselines in the offline experiments. Furthermore, we conducted a controlled experiment at the video streaming platform of Alibaba, where our method simultaneously improved three conflicting business objectives over the latest production system significantly, demonstrating its tangible economic impact in practice.

Contextual situations, such as having dinner at a restaurant on Friday with the spouse, became a useful mechanism to represent context in context-aware recommender systems (CARS). Prior research has shown important advantages of using latent embedding representation approaches to model contextual information in the Euclidean space leading to better recommendations. However, these traditional approaches have major challenges with the construction of proper embeddings of hierarchical structures of contextual information, as well as with interpretations of the obtained representations. To address these problems, we propose the HyperCARS method that models hierarchical contextual situations in the latent hyperbolic space. HyperCARS combines hyperbolic embeddings with hierarchical clustering to construct contextual situations, which allows loose coupling of the contextual modeling component with recommendation algorithms and, therefore, provides flexibility to use a broad range of previously developed recommendation algorithms. We demonstrate empirically that HyperCARS better captures and interprets hierarchical contextual representations, leading to better context-aware recommendations. Because hyperbolic embeddings can also be used in many other applications besides CARS, we also propose the latent embeddings representation framework that systematically classifies prior work on embeddings and identifies novel research streams for hyperbolic embeddings across information systems applications.

Exploring consumer preferences is a crucial task for recommender systems to address the problems of popularity bias, filter bubbles, and consumer boredom, leading to significant improvements in consumer experiences as well as business performance. Meanwhile, when selecting the appropriate exploratory content for expanding consumers' horizons, existing approaches usually suffer from inadequate support of temporal dynamics and personalization in the exploration process, since they do not explicitly model the sequential evolution of preference of each consumer. In addition, they do not systematically capture the long-term impact of exploratory products on future user behaviors. To tackle these challenges, we propose a novel recommendation framework in this paper, where we formulate recommendations as a consumer trajectory learning task and determine trajectory points based on the concept of dynamical systems in mathematics. By doing so, we can simultaneously optimize both the magnitude and the trend of consumer exploration through the transition vectors connecting trajectory points, as well as capturing the sequential continuity and evolution of consumer interests. In addition, we design a novel forward-looking component into the dynamical system to model the impact of current consumer decisions on their future behaviors, leading to even more insightful and useful recommendations. We conduct extensive simulation and offline experiments to demonstrate significant improvements of our method over the state-of-the-art baselines in terms of both the exploration and exploitation performance. In addition, we conduct a large-scale online controlled experiment at a leading video streaming platform in Asia, where our method significantly outperforms the latest production model in the company across multiple business metrics and leads to better consumer experiences. These improvements potentially translate into an additional 20 million USD annual revenue for the company based on the estimate of the manager in charge of the described platform, demonstrating tangible economic impact.

Although building a 360-degree comprehensive view of a customer has been a long-standing goal in marketing, this challenge has not been successfully addressed in many marketing applications because fractured customer data stored across different “silos” are hard to integrate under “one roof” for several reasons. Instead of integrating customer data, in this article we propose to integrate several domain-specific partial customer views into one consolidated or composite customer profile using a Deep Learning-based method that is theoretically grounded in Kolmogorov’s Mapping Neural Network Existence Theorem. Furthermore, our method needs to securely access domain-specific or siloed customer data only once for building the initial customer embeddings. We conduct extensive studies on two industrial applications to demonstrate that our method effectively reconstructs stable composite customer embeddings that constitute strong approximations of the ground-truth composite embeddings obtained from integrating the siloed raw customer data. Moreover, we show that these data-security preserving reconstructed composite embeddings not only perform as well as the original ground-truth embeddings but significantly outperform partial embeddings and state-of-the-art baselines in recommendation and consumer preference prediction tasks.

Recommender systems (RSs) have become one of the major applications that aim to tailor items to the user’s preferences. Traditional recommendation algorithms capture users’ interests and their interactions with items without taking into account contextual information, such as time and location. However, user interests may change depending on the context [1] . In real-life applications, there is plenty of information regarding user’s circumstances and surroundings (e.g., the activity of the user, time, location, weather, etc.). Such contextual information can be high-dimensional and is gathered from multiple sources, such as web pages, mobile devices, and more. RSs taking context information into account are called context-aware recommender systems (CARSs) [1] .

Cross domain recommender systems have been increasingly valuable for helping consumers identify useful items in different applications. However, existing cross-domain models typically require large number of overlap users, which can be difficult to obtain in some applications. In addition, they did not consider the duality structure of cross-domain recommendation tasks, thus failing to take into account bidirectional latent relations between users and items and achieve optimal recommendation performance. To address these issues, in this paper we propose a novel cross-domain recommendation model based on dual learning that transfers information between two related domains in an iterative manner until the learning process stabilizes. We develop a novel latent orthogonal mapping to extract user preferences over multiple domains while preserving relations between users across different latent spaces. Furthermore, we combine the dual learning method with the metric learning approach, which allows us to significantly reduce the required common user overlap across the two domains and leads to even better cross-domain recommendation performance. We test the proposed model on two large-scale industrial datasets and six domain pairs, demonstrating that it consistently and significantly outperforms all the state-of-the-art baselines. We also show that the proposed model works well with very few overlap users to obtain satisfying recommendation performance comparable to the state-of-the-art baselines that use many overlap users.

Existing cross domain recommender systems typically assume homogeneous user preferences across multiple domains to capture similarities of user-item interactions and to provide cross domain recommendations accordingly. Meanwhile, the heterogeneity of user behaviors is usually not well studied and captured during the recommendation process, where users might have vastly different interests in different domains. In addition, previous models focus primarily on recommendation tasks between domain pairs, and cannot be naturally extended to serve for multiple domain recommendation applications. To address these challenges, we propose to utilize the idea of adversarial learning to intelligently incorporate global user preferences and domain-specific user preferences for providing satisfying cross domain recommendations. In particular, our proposed Adversarial Cross Domain Recommendation (ACDR) model first obtains the latent representations of global user preferences from their explicit feature information, and then transforms them into domain-specific user embeddings, where we take into account user behaviors and their heterogeneous preferences among different domains. By doing so, we address the differences among user representations in the domain-specific latent space while also preserving global user preferences, as we effectively segment the distributions of domain-specific user embeddings in the shared latent space. The convergence of our proposed model is theoretically guaranteed. The proposed ACDR model leads to significant and consistent improvements in cross domain recommendation performance over the state-of-the-art baseline models, which we demonstrate through extensive experiments on three real-world datasets. In addition, we show that the improvements are greater on those datasets that are smaller and more sparse, on those users that have fewer interaction records in the dataset, and when user interactions from more product domains are included in the cross domain recommendation model.

We propose the use of a deep learning architecture, called RETINA, to predict multi-alternative, multi-attribute consumer choice from eye movement data. RETINA directly uses the complete time series of raw eye-tracking data from both eyes as input to state-of-the art Transformer and Metric Learning Deep Learning methods. Using the raw data input eliminates the information loss that may result from first calculating fixations, deriving metrics from the fixations data and analysing those metrics, as has been often done in eye movement research, and allows us to apply Deep Learning to eye tracking data sets of the size commonly encountered in academic and applied research. Using a data set with 112 respondents who made choices among four laptops, we show that the proposed architecture outperforms other state-of-the-art machine learning methods (standard BERT, LSTM, AutoML, logistic regression) calibrated on raw data or fixation data. The analysis of partial time and partial data segments reveals the ability of RETINA to predict choice outcomes well before participants reach a decision. Specifically, we find that using a mere 5 s of data, the RETINA architecture achieves a predictive validation accuracy of over 0.7. We provide an assessment of which features of the eye movement data contribute to RETINA’s prediction accuracy. We make recommendations on how the proposed deep learning architecture can be used as a basis for future academic research, in particular its application to eye movements collected from front-facing video cameras.

Context has been an important topic in recommender systems over the past two decades. A standard representational approach to context assumes that contextual variables and their structures are known in an application. Most of the prior CARS papers following representational approach manually selected and considered only a few crucial contextual variables in an application, such as time, location, and company of a person. This prior work demonstrated significant recommendation performance improvements when various CARS-based methods have been deployed in numerous applications. However, some recommender systems applications deal with a much bigger and broader types of contexts, and manually identifying and capturing a few contextual variables is not sufficient in such cases. In this paper, we study such ``context-rich'' applications dealing with a large variety of different types of contexts. We demonstrate that supporting only a few most important contextual variables, although useful, is not sufficient. In our study, we focus on the application that recommends various banking products to commercial customers within the context of dialogues initiated by customer service representatives. In this application, we managed to identify over two hundred types of contextual variables. Sorting those variables by their importance forms the Long Tail of Context (LTC). In this paper, we empirically demonstrate that LTC matters and using all these contextual variables from the Long Tail leads to significant improvements in recommendation performance.

In this paper, we study an important problem of parsing contextual information from user reviews for recommendation purposes. First, we study the ways contextual information is expressed in user reviews and obtain novel insights about it. Among other things, we demonstrate that such type of information tends to appear at the beginning of the review, in longer sentences, in the sentences written in the past tense or using gerund form, and in the sentences referring to some points in time. Second, we propose a novel context parsing method for systematically extracting contextual information from user-generated reviews that rely on the insights obtained in our study. We apply the proposed method to three different Yelp applications (restaurants, hotels, and beauty & spas) and demonstrate that it works well and leads to better recommendation performance than the baseline approaches. Our method systematically extracts more comprehensive sets of relevant contextual variables and corresponding phrases than the baselines. Our analysis also shows the importance of the newly discovered contextual information for recommendation purposes. The obtained results and the proposed method can help to get more comprehensive knowledge about contextual variables in a given application that leads to better recommendations.

Contextual information has been widely recognized as an important modeling dimension in social sciences and in computing. In particular, the role of context has been recognized in enhancing recommendation results and retrieval performance. While a substantial amount of existing research has focused on context-aware recommender systems (CARS), many interesting problems remain under-explored. The CARS 2022 workshop provides a venue for presenting and discussing: the important features of the next generation of CARS; and application domains that may require the use of novel types of contextual information and cope with their dynamic properties in group recommendations and in online environments.

In this paper, we examine the effectiveness of various recommendation strategies in the mobile channel and their impact on consumers’ utility and demand levels for individual products. We find significant differences in effectiveness among various recommendation strategies. Interestingly, recommendation strategies that directly embed social proofs for the recommended alternatives outperform other recommendations. In addition, recommendation strategies combining social proofs with higher levels of induced awareness due to the prescribed temporal diversity have an even stronger effect on the mobile channel. We also examine the heterogeneity of the demand effect across items, users, and contextual settings, further verifying empirically the aforementioned information and persuasion mechanisms and generating rich insights. We also facilitate the estimation of causal effects in the presence of endogeneity using machine-learning methods. Specifically, we develop novel econometric instruments that capture product differentiation (isolation) based on deeplearning models of user-generated reviews. Our empirical findings extend the current knowledge regarding the heterogeneous impact of recommender systems, reconcile contradictory prior results in the related literature, and have significant business implications.