Preference Elicitation

Recommendation systems play an invisible but powerful role in our daily lives, helping us discover music, movies, courses, and even products we might enjoy. Yet two stubborn challenges remain. The first is the cold-start problem, where new users receive generic suggestions because the system knows nothing about them. The second is the lack of explicit feedback, since most people rarely bother to leave ratings or reviews. This conceptual paper explores a potential solution: adding a moodbased conversational chatbot as a feedback layer for recommendation systems. Instead of asking users to rate items with stars or numbers, the chatbot engages in simple, natural dialogue: "How are you feeling today?" or "What are you in the mood for?" This conversational design serves two purposes. It helps reduce cold-start limitations by gathering useful mood input right away, and it encourages richer, more sincere feedback without forcing users into mechanical rating tasks. The result is a system that feels more personal, more engaging, and more aligned with the way people actually make choices. By blending mood signals with existing recommendation algorithms, such systems could open new possibilities in entertainment, e-commerce, education, and even healthcare.

We study the transmission of risk attitudes in a unique survey of mothers and children in which both participated in an incentivized risk preference elicitation task. We document that risk preferences are correlated between mothers and children when the children are just 7-8 years old. This correlation is only present for daughters. We further show that a measure of maternal involvement is a strong moderator of the association between mothers' and daughters' risk tolerance. This is consistent with a role for socialization and parental investment in the intergenerational transmission of risk preferences.

Collaborative filtering systems rely heavily on matrix factorization techniques, which often face scalability issues when handling large datasets. This paper presents an efficient parallel algorithm that leverages distributed computing to perform largescale matrix factorization. Experimental results show that our algorithm significantly reduces computation time while maintaining high accuracy. The approach has practical implications for recommendation systems, particularly in ecommerce and social media platforms.

Constrained CP-nets Steve Prestwich 1 , Francesca Rossi 2 , Kristen Brent Venable 2 , Toby Walsh 1 1: Cork Constraint Computation Centre, University College Cork, Ireland. Email: s. prestwich@ cs. ucc. ie, tw@ 4c. ucc. ie 2: Department of Pure and Applied Mathematics, ...

This research proposes the development and implementation of a novel method for multi-criteria decision analysis (MCDA). The proposed method (MCGL: Multi Criteria Gain Loss) is based on tenets of prospect theory and the complexity theory. The simple preference elicitation in MCGL improves the performance of the decision maker (DM). The usage of descriptive theories of decision making to process these preferences increases the accuracy of MCGL in modeling the choice behavior of the DM. Through a pilot study, the effectiveness of MCGL is tested by collecting responses from 23 university students using the MCGL and the Analytic Hierarch Process (AHP) on a hypothetical mobile selection problem. When the ranking of mobile phones obtained from the AHP and from the MCGL are analyzed, the rankings obtained from the MCGL are more similar to the direct rankings given by the respondents. The results indicate that the MCGL captures the human decision making process more accurately than the AHP....

In the presence of incomplete risk attitudes, choices between noncomparable risky prospects are random. A random choice model advanced by Karni (Incomplete preferences and random choice (unpublished manuscript), 2021) includes the hypothesis that choices among noncomparable risky prospects are prompted by signals drawn from personal distributions. This paper introduces a scheme designed to elicit subjects' assessments of their personal likelihoods of choices among noncomparable risky prospects and describes experiments designed to test the aforementioned hypothesis.

We study the aggregation of a couple's preferences over their respective jobs when they enter a centralized labor market jointly, such as the market for assigning hospital residencies to medical students. Usually in such markets couples need to submit joint preferences over pairs of jobs. Starting from two individual preference orderings over jobs, we first study the Lexicographic and the Rank-Based Leximin aggregation rules, and then propose a family of aggregation rules, the k-Lexi-Pairing rules, and provide an axiomatic characterization of these rules. The parameter k indicates the degree of selfishness for one partner (and altruism for the other partner), with the least selfish Rank-Based Leximin rule at one extreme and the most selfish Lexicographic rule at the other extreme. Since couples care about geographic proximity, we also identify a simple parametric family of preference aggregation rules which build on the k-Lexi-Pairing rules and reflect the couple's preference for togetherness.

While content-based recommendation has been applied successfully in many different domains, it has not seen the same level of attention as collaborative filtering techniques have. However, there are many recommendation domains and applications where content and metadata play a key role, either in addition to or instead of ratings and implicit usage data. For some domains, such as movies, the relationship between content and usage data has seen thorough investigation already, but for many other domains, such as books, news, scientific articles, and Web pages we still do not know if and how these data sources should be combined to provided the best recommendation performance. The CBRecSys workshop provides a dedicated venue for papers dedicated to all aspects of content-based recommendation.

While content-based recommendation has been applied successfully in many different domains, it has not seen the same level of attention as collaborative filtering techniques have. However, there are many recommendation domains and applications where content and metadata play a key role, either in addition to or instead of ratings and implicit usage data. For some domains, such as movies, the relationship between content and usage data has seen thorough investigation already, but for many other domains, such as books, news, scientific articles, and Web pages we still do not know if and how these data sources should be combined to provided the best recommendation performance. The CBRecSys 2014 workshop aims to address this by providing a dedicated venue for papers dedicated to all aspects of content-based recommendation.

Recommender systems (RSs) have become increasingly vital in the modern information era and connected economy. They play a key role in business operations by generating personalized suggestions and minimizing information overload. However, the performance of traditional RSs is limited by data sparseness and cold-start issues. Though deep learning-based recommender systems (DLRSs) are very popular, they underperform when considering rating matrices with sparse entries. Despite their performance improvements, DLRSs also suffer from data sparsity, cold start, serendipity, and generalizability issues. We propose a multistage model that uses multimodal data embedding and deep transfer learning for effective and personalized product recommendations, and is designed to overcome data sparsity and cold-start issues. The proposed model includes two phases. In the first-offlinephase, a deep learning technique is implemented to learn hidden features from a large image dataset (targeting new item cold start), and a multimodal data embedding is used to produce dense user feature and item feature vectors (targeting user cold start). This phase produces three different similarity matrices that are used as inputs for the second-onlinephase to generate a list of top-n relevant items for a target user. We analyzed the accuracy and effectiveness of the proposed model against the existing baseline RSs using a Brazilian E-commerce dataset. The results show that our model scored 0.5882 for MAE and 0.4011 for RMSE which is lower than baseline RSs which indicates that the model achieved an improved accuracy and was able to minimize the typical cold start and data sparseness issues during the recommendation process.

Fifty-four percent of Benin's population in rural areas keep indigenous chickens for subsistence livelihoods. Maintaining diversity in chicken gene pools can provide several benefits for farmers, including risk pooling, diversification of income and maintaining availability of chickens suitable for cultural ceremonies. Despite the potential to alleviate poverty by improving indigenous chicken breeds, smallholders' participation in the implementation of breeding programmes is weak. Participation can be improved with greater understanding of the many functions of chickens to smallholders, particularly their economic contribution. The objectives of this study are (1) to evaluate chicken traits including market and non-market values, and (2) to assess factors that influence the conservation of indigenous breeds. Choice modelling, a multi-attribute preference elicitation technique, was applied across 300 households in two districts in Benin (Dassa and Toffo). The results revealed that adaptive and performance traits in chicken breeds are highly valued by farmers and that preferences differed greatly between farmers in the two districts. Many of the preferred traits are expressed in indigenous chickens, whose conservation should be supported through village chicken breeding programmes. However, a preference for white plumage, most common among exotic breeds, could hinder conservation of indigenous breeds, which are mostly brown or black. From an economic point of view, the aim of conserving culturally significant and disease resistant indigenous breeds is contrary to the objective of increasing chicken productivity. The lack of knowledge about chicken characterization and a lack of flock management were identified as further severe constraints to village breeding and conservation programmes.

We describe a web-based system to support groups in elaborating participatory budgets. Rather than using physical meetings with voting mechanisms, we promote virtual meetings with explicit preference elicitation, guided negotiations and, only if consensus is not reached, voting. Participatory budgets constitute an attempt to allow citizens to have a word and aid in deciding and approving how public budgets, mainly in municipalities, are spent, whether (and how much) in transportation, culture, education, urban development, health,... They constitute a budget allocation model based on dialogue and citizen participation, which diverges from the current predominant representative model in which citizens choose representatives for four years, with practically no other direct opportunity to influence council policies. In a sense, participatory budgets are transforming the idea of a representative democracy, in which the preferences of the citizens are considered just at the moment of elections, through voting, to move closer to a participatory democracy, based on direct participation and discussion of issues. In words of Souza, see [10]: Participatory budgets constitute a universal, direct and volunteer kind of democracy which gives population the opportunity of discussing about budgets and public policy, making relevant decisions. Citizens not only vote, but also get involved. Among many potential benefits of participatory budgets, we would include: -Legitimation -Approaching decisions to citizens -Public decisions made publicly -Mitigation of alienation and apathy -Transparency

The ultimate goal of any software developer seeking a competitive edge is to meet stakeholders' needs and expectations. To achieve this, it is necessary to effectively and accurately manage stakeholders' system requirements. The paper proposes a systematic way of classifying stakeholders and then describes a novel method for calculating stakeholder priority taking into consideration the fact that different stakeholders will have different importance level and different requirement preference. Finally the requirement preference calculation is done where stakeholders choose the best requirements based on two factors, value and urgency of the requirement. The proposed method actively involves stakeholders in the requirement elicitation process (Abstract)

Background There is growing evidence of a reluctance to allocate health care solely on the basis of maximizing quality-adjusted life years (QALYs). Stated preference methods can be used to elicit preferences for efficiency vs. equity in the allocation of health-care resources. Objective To compare discrete choice experiment (DCE) and constant-sum paired comparison (CSPC) methods for eliciting societal preferences. Methods Over a series of choice pairs, DCE respondents allocated a fixed budget to one preferred group and CSPC respondents allocated budget percentages between the groups. Questionnaires were compared in terms of completion rates, preference consistency, dominant preferences and derived attribute importance. There was no significant difference in the proportions that rated the questionnaires somewhat or extremely difficult, but a significantly greater proportion completed the DCE compared to the CSPC. Preference consistency was also higher in the DCE. The incidence of dominant preferences, including for aggregate QALYs, was low and not significantly different between questionnaires. Similarly, no CSCP respondents equalized budgets or outcomes in every task. Final health state was the most important attribute in both questionnaires, but the rankings diverged for the other attributes. Notably, the total patients' treated attribute was important in the CSPC but insignificant in the DCE, perhaps reflecting a 'prominence effect'. Conclusions Despite lower completion rates and preference consistency, CSPC may offer advantages over DCE in eliciting preferences over the distribution of resources and/or outcomes as well as attribute levels, avoiding extreme 'all-or-nothing' distributions and possibly aligning respondent attention more closely with a societal perspective.

This study further examines the failure of game form recognition in preference elicitation by making elicitation more cognitively demanding through a cognitive load manipulation. We hypothesized that if subjects misperceive one game for another game, then by depleting their cognitive resources, subjects would misconceive the morecognitively demanding task for the less-cognitively demanding task at a higher rate. We find no evidence that subjects suffer from a first-price-auction game-form misconception but rather that once cognitive resources are depleted, subjects' choices are better explained by random choice. More cognitively able subjects are more immune to deviations from sub-optimal play than lower cognitively able subjects.

The rapid growth of Global Positioning System (GPS) and availability of real-time Geo-located data allow the mobile devices to provide information which leads towards the Location Based Services (LBS). The need for providing suggestions to personals about the activities of their interests, the LBS contributing more effectively to this purpose. Recommendation system (RS) is one of the most effective and efficient features that has been initiated by the LBS. Our proposed system is intended to design a recommendation system that will provide suggestions to the user and also find a suitable place for a group of users and it is according to their preferred type of places. In our work, we propose the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm for clustering the check-in spots of the user’s and user-based Collaborative Filtering (CF) to find similar users as we are considering constructing an interest profile for each user. We also introduced a grid-base...

IntroductionApproximately one million undiagnosed persons living with HIV in Southern and Eastern Africa need to test for HIV. Novel approaches are necessary to identify HIV testing options that match the heterogeneous testing preferences of high-risk populations. This pragmatic randomised controlled trial (PRCT) will evaluate the efficacy of a preference-informed, heterogeneity-focused HIV counselling and testing (HCT) offer, for improving rates of HIV testing in two high-risk populations.Methods and analysisThe study will be conducted in Moshi, Tanzania. The PRCT will randomise 600 female barworkers and 600 male Kilimanjaro mountain porters across three study arms. All participants will receive an HIV testing offer comprised of four preference-informed testing options, including one ‘common’ option—comprising features that are commonly available in the area and, on average, most preferred among study participants—and three options that are specific to the study arm. Options will b...

We explore risk preference elicitation via direct choice over lotteries. The choice task attributes vary incrementally, and range from choosing between two lotteries to selecting a portfolio from a continuous set of bundled Arrow securities. Each subject completes multiple instances of five different tasks, and responses for each task are summarized in parametric (CRRA) and non-parametric (normalized risk premium) measures of risk preference. We find that the distributions of elicited preferences vary widely across tasks, as does within-subject correlation across pairs of tasks. Observed variation in elicited preferences across tasks, and in correlations, is explained in part by variation in design attributes, such as graphical vs text representation, that have no role in standard decision theory.

1 ABSTRACT The goal of the research presented in this paper is to propose, test, document and evaluate the application of preference models in the field of urban planning and design. The preference models simulate the human decision making in various hypothetical situations. The preference model used in the research is created in experimental way by using the conjoint analysis. It is demonstrated how the complex characteristics of the environment can be transformed into operational form that can be consequently used for preference analysis and modelling. The preference model assembly phases are demonstrated. Once the preference model is created, it provides with the predictions of preferences for alternative scenarios (combinations of environment characteristics) and other information that supports the interpretation of the factors influencing the identified preferences. The paper presents concrete outputs that can be directly applied to everyday practice of urban planners and desig...

This paper is an application of the contingent valuation method on community plantations in the highlands of Ethiopia. A discrete-continuous elicitation format was applied. It was found that there is a problem in applying a closed ended elicitation format in this context with a community resource since a community resource typically implies a community based scenario and such a scenario invites to yea-saying. The wellknown problem of compliance bias is also difficult to avoid in such settings. Application of a closed-ended format under such circumstances would exaggerate the willingness to pay for the good in question. The study asked both spouses in a household for their willingness to pay for a new plantation. The analysis of the bid function shows that there are gender variations in the factors that affect the bids. The common preference model was thus rejected in this application. The analysis also indicates that it might be a good idea to concentrate plantation efforts since there seem to be specialization going on in collection behavior. Women in villages without any existing community plantation are, however, significantly more interested in the establishment of a plantation than men. The aggregate willingness to pay vary dramatically between villages pointing at the need for good selection methods in targeting such interventions.

A travers cette thèse nous apportons un éclairage sociologique au débat engagé depuis des décennies sur la problématique de la mise en vente des bois publics. A ce titre, nous étudions, dans la première partie de la thèse, le système d’enchères qui reste encore le principal mode de vente utilisé par l’Office National des Forêts. Nous identifions les conditions d’encastrement de ces ventes, considérant que les comportements des participants sont déterminés par l’interaction complexe de facteurs économiques, socio-structuraux, politiques, cognitifs, institutionnels formels et informels. Nous montrons à cette occasion que l’intérêt accordé à l’enchère réside dans sa capacité à trouver collectivement un consensus sur la valeur et l’allocation des lots de bois. Dans la deuxième partie de la thèse, nous analysons l’évolution récente du cadre institutionnel appliqué aux ventes de bois publics en nous appuyant sur les notions de construction sociale des institutions et de dépendance au chem...

Ce travail n'aurait sans doute pas vu le jour sans la participation d'un grand nombre de personnes que je tiens, ici, à remercier. J'adresse tout d'abord mes remerciements les plus sincères à mon directeur de thèse, Pierre Demeulenaere, pour m'avoir fait confiance et accompagné durant toutes ces années de DEA et de thèse. Je voudrais aussi remercier Raphaële Préget, pour avoir accepté tout de suite de travailler avec moi sur la thématique des enchères de bois publics. Ses commentaires et son optimisme constant à l'égard de mes travaux ont été pour moi un facteur de motivation. Je remercie également Claude Didry, Franck Cochoy, Marie-France Garcia Parpet et Philippe Steiner pour l'honneur, qu'ils me font de participer à ce jury. Ce travail doit également beaucoup à la collaboration de tous les membres du Laboratoire d'Economie Forestière de Nancy, encore présents ou partis dans d'autres centres de recherche. Merci, tout particulièrement, à Anne Stenger et Franck Lecoq pour m'avoir ouvert les portes du laboratoire dès notre premier rencontre, considérant qu'une approche sociologique au côté de celle des économistes ne pourrait qu'améliorer la compréhension des problématiques forestières. Merci à Serge et Claire pour leur soutien (en tant qu'exilés du Sud, on se comprend !), à Marielle pour avoir accepté de partager depuis tant d'années le même bureau, à Alexandra pour m'avoir toujours trouvé, dans la minute, les données sur la forêt dont j'avais besoin, à Max pour son expertise forestière avisée, à Evelyne pour toute l'aide administrative apportée, et enfin à Ahmed, Philippe, Olivier, Jens, Arnaud, Julien, Marieke, Jean-Pierre mais aussi Sandrine, Jean-Marc, Lisette et Lise, et bien sûr, à tous mes collègues doctorants ou ex-doctorants pour toute la sympathie qu'ils ont su témoigner en permanence à l'égard de « leur sociologue ».

We first show that the optimal and undominated outcomes of an unconstrained (and possibly cyclic) CP-net are the solutions of a set of hard constraints. We then propose a new algorithm for finding the optimal outcomes of a constrained CPnet which makes use of hard constraint solving. Unlike previous algorithms, this new algorithm works even with cyclic CP-nets. In addition, the algorithm is not tied to CP-nets, but can work with any preference formalism which produces a preorder over the outcomes. We also propose an approximation method which weakens the preference ordering induced by the CP-net, returning a larger set of outcomes, but provides a significant computational advantage. Finally, we describe a weighted constraint approach that allows to find good solutions even when optimals do not exist.

Abstract: User preferences for automated assistance often vary widely, depending on the situation, and quality or presentation of help. Developing effectivemodels to learn individual preferences online requires domain models that associate observations of user behavior with their utility functions, which in turn can be constructed using utility elicitation techniques. However, most elicitation methods ask for users' predicted utilities based on hypothetical scenarios rather than more realistic experienced utilities. This is especially true in ...

We investigate the solution of constraint-based configuration problems in which the preference function over outcomes is unknown or incompletely specified. The aim is to configure a system, such as a personal computer, so that it will be optimal for a given user. The goal of this project is to develop algorithms that generate the most preferred feasible configuration by posing preference queries to the user. In order to minimize the number and the complexity of preference queries posed to the user, the algorithm reasons about the ...

Many methods to elicit preference models in multi-attribute decision making rely on evaluations of a set of sample alternatives by decision makers. Using orthogonal design methods to create this set of alternatives might require respondents to evaluate unrealistic alternatives. In this paper, we perform an empirical study to analyze whether the presence of such implausible alternatives has an effect on the quality of utility elicitation. Using a new approach to measure consistency, we find that implausible alternatives in fact, have a positive effect on consistency of intra-attribute preference information and consistency with dominance, but do not affect inter-attribute preference information.

We introduce a suite of interlinked software tools for eliciting preferences, doing decision-theoretic planning, and displaying the plans. The software allows a user to walk through possible trajectories, adjust preferences, and compare potential trajectories. This paper focuses on the elicitation process, the plan display, and interactions between the two.

improving quality of life over extending life. ConClusions: Advanced NSCLC is known to impact many domains of patients' lives. This study demonstrates that emotional impact and time taken undergoing treatment may be undervalued by commonly employed HRQoL metrics in clinical trials. Future clinical trials of new lung cancer treatments should include assessment of these concepts. Ultimately, HRQoL instruments should be identified/developed that satisfactorily capture all factors deemed important by patients in order to fully reflect impact of new treatments on patients' lives.

In this paper, I present an axiomatic choice theory of Bayesian decision makers and define choice-based subjective probabilities that truly represent Bayesian decision makers' prior and posterior beliefs. I give an example that illustrates the potential consequences of ascribing to a decision maker subjective probabilities that do not represent his beliefs. I argue that no equivalent results may be obtained in theories that invoke Savage's (1954) idea of a state space.

In late 2000, the European Union adopted the Water Framework Directive (WFD) and funded a series of research and innovation projects to support its implementation. One of these was the MULINO project (MULti-sectoral, INtegrated and Operational Decision Support System for Sustainable Use of Water Resources at the Catchment Scale). Its main product was a decision support system (mDSS) tool designed to help water managers make choices related to WFD implementation in a participatory manner. After the end of MULINO, a long sequence of research projects allowed for the maintenance and continuous development of its tool, which has been applied for more than 20 years in various contexts related to environmental and integrated management. This experience and an analysis of the literature allow us to draw some general conclusions regarding DSS tools for water management and their role in our societies. Lessons learned are proposed, from the need to frame tools within sound methodological frameworks for the management of decision processes, supporting instead of substituting decision-makers in their roles, to the trade-offs that appear between ease of use and specificity on one side and flexibility and reusability on the other. The specific strengths attributed to mDSS include the provision of an interface based on a simplified and understandable conceptual framework that facilitates communication with interested parties, the flexibility and ability to approach a wide variety of decisional issues, the relatively simple and understandable decision rules provided by the tool, and the simplified connections with other software environments. This paper presents the current version of the software and reports on the experience of its development and use over more than two decades; it also identifies the way forward. Impact statement Decision support system tools are often delivered by research projects for translating scientific knowledge into practical applications, but they are rarely used outside or after the project. Limited reuse may derive from tools that incorporate case-specific data or that are not adequately documented, or that become obsolete in their application context or codes. The case we examine here is that of MULINO-DSS (mDSS), which was designed and released by a European project that ended more than 20 years ago and is still in use. It has thrived for over two decades for at least two reasons: first, it still provides useful functionality to support environmental and integrated decision-making, and second, its authors have maintained and renewed the code through a long series of project grants. The main strengths of mDSS can be found in its reference to a simplified conceptual framework that facilitates communication with interested parties, its flexibility and ability to approach a wide variety of decisional issues, and its relatively simple and understandable decision rules grounded in multi-criteria analysis methods. However, these strengths come with the cost of the effort and skills needed to tailor the system to new applications, in particular for interfacing the DSS with external models and data. The water policy background Water resources management (WRM) is a topic that captures the interests of both policy-makers and the public. Water is crucial for sustaining of life and has significant environmental, economic, and cultural implications. Nevertheless, the outcomes of the 2023 United Nations Water Conference (UN, 2023) showed that the world is not on track to achieve water-related Sustainable Development Goals and targets (UN, 2015), and the global water crisis is worsening.

Conjoint Preference Elicitation Methods in the Broader Context of Random Utility Theory Preference Elicitation Methods Hensher, Louviere & Swait 2 but some are not, some methods permit one to combine experimental with actual marketplace choice data but many do not, etc. Indeed, there are many ways to understand and model preferences and choices, some of which bear scant relation to one another and others that are incompatible, both theoretically and analytically. RUT offers a way to unify many seemingly disparate approaches to understand and model preference formation and choice. Figure represents an overview of the general problems covered by RUT, which can assist our understanding of the role of conjoint analysis methods within RUT. Figure should be regarded as a pedagogical vehicle to help explain why a more general view is required; it is not a theory per se.

The term "decision analysis" can be used more narrowly or more broadly, and this can be a source of confusion when our societies interact. For example, there are substantial differences between the editorial statements about Decision Analysis in the INFORMS Journals and the editorial statement for the Journal of Multicriteria Decision Analysis. For clarity, we discuss the terminology around decision analysis as DAS members would typically understand it. According to the editorial statement for the Decision Analysis area of the journal Operations Research: "Decision analysis methods have traditionally addressed modeling uncertainty (Bayesian inference, subjective probability elicitation, combining expert opinions, scoring rules for eliciting and evaluating probability assessments, sensitivity analysis, information value); structuring preferences (utility and risk attitude, utility assessment, stochastic dominance, structuring objectives and attributes, multiattribute utility and value); and representing and solving decision problems (decision trees, influence diagrams, Bayesian networks, alternative generation, value trees, fault trees, dynamic programming). Decision analysis methods also draw on related fields such as the psychology of judgment and choice (heuristics and biases, prospect theory) or methods for dealing with multiple stakeholders (cooperative and noncooperative game theory, negotiation)." The objective of much of this work is to assist individual decision makers in choosing among alternatives with uncertain outcomes.

https://scholar.google.com/citations?view_op=view_citation&hl=en&user=2GNKVYUAAAAJ&citation_for_view=2GNKVYUAAAAJ:YsMSGLbcyi4C

Recommender systems (RS) are increasingly important in so- cial media, entertainment, and e-commerce in the information explosion era. However, the collected data contains many bi- ases such as selection bias, as users are free to choose items to rate, making the collected data not representative of the target population. Recently, many methods such as relabeling-based and reweighting-based have been proposed to mitigate the selection bias. However, the effectiveness of these methods relies on strong assumptions, which are difficult to satisfy in real-world scenarios, leading to sub-optimal debiasing perfor- mance. In this paper, we propose a debiasing method from the machine unlearning perspective. Specifically, we first pro- pose a user unlearning rate network to determine which user needs to be unlearned. Then we generate the error-maximizing pseudo-labels for each user and fusion such pseudo-labels and the observed labels based on the learned user unlearning rate to mitigate the selection bias. In addition, we further propose an unlearning to debias training algorithm to achieve unbiased learning of the prediction model. Finally, we conduct exten- sive experiments on three real-world datasets to validate the effectiveness of our method.

Most models of utility elicitation in decision support and interactive optimization assume a predefined set of "catalog" features over which user preferences are expressed. However, users may differ in the features over which they are most comfortable expressing their preferences. In this work we consider the problem of feature elicitation: a user's utility function is expressed using features whose definitions (in terms of "catalog" features) are unknown. We cast this as a problem of concept learning, but whose goal is to identify only enough about the concept to enable a good decision to be recommended. We describe computational procedures for identifying optimal alternatives w.r.t. minimax regret in the presence of concept uncertainty; and describe several heuristic query strategies that focus on reduction of relevant concept uncertainty.

We propose new methods of preference elicitation for constraint-based optimization problems based on the use of minimax regret. Specifically, we assume a constraintbased optimization problem (e.g., product configuration) in which the objective function (e.g., consumer preferences) are unknown or imprecisely specified. Assuming a graphical utility model, we describe several elicitation strategies that require the user to answer only binary (bound) queries on the utility model parameters. While a theoretically motivated algorithm can provably reduce regret quickly (in terms of number of queries), we demonstrate that, in practice, heuristic strategies perform much better, and are able to find optimal (or near-optimal) configurations with far fewer queries.

Bayesian approaches to utility elicitation typically adopt (myopic) expected value of information (EVOI) as a natural criterion for selecting queries. However, EVOI-optimization is usually computationally prohibitive. In this paper, we examine EVOI optimization using choice queries, queries in which a user is ask to select her most preferred product from a set. We show that, under very general assumptions, the optimal choice query w.r.t. EVOI coincides with the optimal recommendation set, that is, a set maximizing the expected utility of the user selection. Since recommendation set optimization is a simpler, submodular problem, this can greatly reduce the complexity of both exact and approximate (greedy) computation of optimal choice queries. We also examine the case where user responses to choice queries are error-prone (using both constant and mixed multinomial logit noise models) and provide worst-case guarantees. Finally we present a local search technique for query optimization that works extremely well with large outcome spaces.

Utility elicitation is a critical function of any automated decision aid, allowing decisions to be tailored to the preferences of a specific user. However, the size and complexity of utility functions often precludes full elicitation, requiring that decisions be made without full utility information. Adopting the minimax regret criterion for decision making with incomplete utility information, we describe and empirically compare several new procedures for incremental elicitation of utility functions that attempt to reduce minimax regret with as few questions as possible. Specifically, using the (continuous) space of standard gamble queries, we show that myopically optimal queries can be computed effectively (in polynomial time) for several different improvement criteria. One such criterion, in particular, empirically outperforms the others we examine considerably, and has provable improvement guarantees.

In many situations, a set of hard constraints encodes the feasible configurations of some system or product over which multiple users have distinct preferences. However, making suitable decisions requires that the preferences of a specific user for different configurations be articulated or elicited, something generally acknowledged to be onerous. We address two problems associated with preference elicitation: computing a best feasible solution when the user's utilities are imprecisely specified; and developing useful elicitation procedures that reduce utility uncertainty, with minimal user interaction, to a point where (approximately) optimal decisions can be made. Our main contributions are threefold. First, we propose the use of minimax regret as a suitable decision criterion for decision making in the presence of such utility function uncertainty. Second, we devise several different procedures, all relying on mixed integer linear programs, that can be used to compute minimax regret and regret-optimizing solutions effectively. In particular, our methods exploit generalized additive structure in a user's utility function to ensure tractable computation. Third, we propose various elicitation methods that can be used to refine utility uncertainty in such a way as to quickly (i.e., with as few questions as possible) reduce minimax regret. Empirical study suggests that several of these methods are quite successful in minimizing the number of user queries, while remaining computationally practical so as to admit real-time user interaction.

Preference elicitation is a key problem facing the deployment of intelligent systems that make or recommend decisions on the behalf of users. Since not all aspects of a utility function have the same impact on object-level decision quality, determining which information to extract from a user is itself a sequential decision problem, balancing the amount of elicitation effort and time with decision quality. We formulate this problem as a partially-observable Markov decision process (POMDP). Because of the continuous nature of the state and action spaces of this POMDP, standard techniques cannot be used to solve it. We describe methods that exploit the special structure of preference elicitation to deal with parameterized belief states over the continuous state space, and gradient techniques for optimizing parameterized actions. These methods can be used with a number of different belief state representations, including mixture models.

The specification of a Markov decision process (MDP) can be difficult. Reward function specification is especially problematic; in practice, it is often cognitively complex and time-consuming for users to precisely specify rewards. This work casts the problem of specifying rewards as one of preference elicitation and aims to minimize the degree of precision with which a reward function must be specified while still allowing optimal or near-optimal policies to be produced. We first discuss how robust policies can be computed for MDPs given only partial reward information using the minimax regret criterion. We then demonstrate how regret can be reduced by efficiently eliciting reward information using bound queries, using regret-reduction as a means for choosing suitable queries. Empirical results demonstrate that regret-based reward elicitation offers an effective way to produce near-optimal policies without resorting to the precise specification of the entire reward function.

We investigate the application of classification tech niques to utility elicitation. In a decision problem, two sets of parameters must generally be elicited: the prob abilities and the utilities. While the prior and condi tional probabilities in the model do not change from user to user, the utility models do. Thus it is necessary to elicit a utility model separately for each new user. Elicitation is long and tedious, particularly if the out come space is large and not decomposable. There are two common approaches to utility function elicitation. The first is to base the determination of the user's util ity function solely on elicitation of qualitative prefer ences. The second makes assumptions about the form and decomposability of the utility function. Here we take a different approach: we attempt to identify the new user's utility function based on classification rel ative to a database of previously collected utility func tions. We do this by identifying clusters of utility func tions that minimize an appropriate distance measure. Having identified the clusters, we develop a classifi cation scheme that requires many fewer and simpler assessments than full utility elicitation and is more ro bust than utility elicitation based solely on preferences. We have tested our algorithm on a small database of utility functions in a prenatal diagnosis domain and the results are quite promising.

Prostate cancer is a slow progressing cancer that affects millions of men in the US. Due to uncertainties in outcomes and treatment complications, it is important that patients engage in informed decision making to choose the "optimal treatment". Patient centered care that encompasses informed decision-making can improve treatment choice and quality of care. Thus, assessing patient treatment preferences is critical for developing an effective decision support system. The objective of this patient-centered randomized clinical trial was to study the comparative effectiveness of a conjoint analysis intervention compared to usual care in improving subjective and objective outcomes in prostate cancer patients. We identified preferred attributes of alternative prostate cancer treatments that will aid in evaluating attributes of treatment options. In this two-phase study, in Phase 1 we used mixed methods to develop an adaptive conjoint task instrument. The conjoint task required ...

Conversational recommendation systems (CRS) effectively address information asymmetry by dynamically eliciting user preferences through multi-turn interactions. However, existing CRS methods commonly assume that users have clear, definite preferences for one or multiple target items. This assumption can lead to over-trusting user feedback, treating accepts/rejects as definitive signals to filter items and reduce the candidate space, potentially causing over-filtering and excluding relevant alternatives. In reality, users often exhibit vague preferences, lacking well-defined inclinations for certain attribute types (e.g., color, pattern), and their decision-making process during interactions is rarely binary. Instead, users' choices are relative, reflecting a range of preferences rather than strict likes or dislikes. To address this issue, we introduce a novel scenario called Vague Preference Multi-round Conversational Recommendation (VPMCR), which employs a soft estimation mechanism to assign non-zero confidence scores to all candidate items, accommodating users' vague and dynamic preferences while mitigating over-filtering. In the VPMCR setting, we introduce a solution called Vague Preference Policy Learning (VPPL), which consists of two main components: Ambiguity-aware Soft Estimation (ASE) and Dynamism-aware Policy Learning (DPL). ASE aims to capture the vagueness of user preferences by estimating preference scores for both clicked and non-clicked options, employing a choice-based approach and a time-aware preference decay strategy. DPL implements a policy learning framework, leveraging the preference distribution from ASE, to guide the conversation and adapt to changes in users' preferences for making recommendations or querying attributes. Extensive experiments conducted on diverse datasets demonstrate the effectiveness of VPPL within the VPMCR framework, outperforming existing methods and setting a new benchmark for CRS research. Our work represents a significant advancement in accommodating the inherent ambiguity and relative decision-making processes exhibited by users, improving the overall performance and applicability of CRS in real-world settings.

We consider the problem of minimizing preference elicitation in efficient multiattribute auctions, that support dynamic negotiation over non-price based attributes such as quality, time-of-delivery, and processor speed. We introduce asynchronous price-based multiattribute auctions, with proxy bidding agents that sit between the auctioneer and the participants. Empirical results demonstrate the preference elicitation savings that are provided with minimal price spaces, asynchronous updates, and proxy agents.