Kernel-based learning methods for preference aggregation

The mathematical representation of human preferences has been a subject of study for researchers in different fields. In multi-criteria decision making (MCDM) and fuzzy modeling, preference models are typically constructed by interacting with the human decision maker (DM). However, it is known that a DM often has difficulties to specify precise values for certain parameters of the model. He/she instead feels more comfortable to give holistic judgements for some of the alternatives. Inference and elicitation procedures then assist the DM to find a satisfactory model and to assess unjudged alternatives. In a related but more statistical way, machine learning algorithms can also infer preference models with similar setups and purposes, but here less interaction with the DM is required/allowed. In this article we discuss the main differences between both types of inference and, in particular, we present a hybrid approach that combines the best of both worlds. This approach consists of a very general kernel-based framework for constructing and inferring preference models. Additive models, for which interpretability is preserved, and utility models can be considered as special cases. Besides generality, important benefits of this approach are its robustness to noise and good scalability. We show in detail how this framework can be utilized to aggregate single-criterion outranking relations, resulting in a flexible class of preference models for which domain knowledge can be specified by a DM.      

Information aggregation in multicandidate elections under plurality rule and runoff voting

I consider a model in which imperfectly informed voters with common interests participate in a multicandidate election decided by either plurality rule or a runoff. Prior to the election, each voter receives a private signal corresponding to the candidate the voter thinks is best. Voters are relatively more likely to think a given candidate is best if the candidate is a relatively better candidate. I show that there is a sequence of equilibrium strategies for the voters such that, as the number of voters goes to infinity, the probability that the best candidate is elected goes to 1. I further show that all candidates receive significant vote shares in any equilibrium in which information fully aggregates under plurality rule and that voters do at least as well when the election is decided by a runoff as they do when the election is decided by plurality rule.     

Social utility functions for strategic decisions in probabilistic voting models

This paper studies societies which have probabilistic voting that is smooth, scalable and unbiased. Its results establish that, in such societies, the decisions of vote-seeking candidates who start at a common location (such as the status quo for the society's policies and/or the same allocation of campaign resources) contain implicit rationality properties. In particular, it shows that in every such society there exist social utility functions which simultaneously rationalize the directional Nash behavior of candidates, the stationary electoral equilibria, and the non-degenerate local electoral equilibria which can occur at these locations. This is shown to be true both for unconstrained and for constrained sets of possible candidate locations. An example of such a utility function (which occurs in every one of the societies under consideration) is also provided.     

Limited resoluteness and strategic voting: The case of linear sincere preference orderings

This note is concerned with the strategic manipulation of the aggregation procedures, which for every profile of individual preference orderings and every set of alternatives (presented for choice) specify a non-empty subset of alternatives. In the context of multi–valued aggregation procedures, the strategic behaviour of an individual crucially depends on the criterion of comparing different subsets of alternatives. It is shown that when the sincere (or true) preference orderings are restricted to be strict, the requirement of resoluteness in pairwise choice is the frontier between the manipulability and non-manipulability results in social choice as long as individuals adopt some versions of the sure-thing principle in comparing different subsets of alternatives.     

A new distance measure including the weak preference relation: Application to the multiple criteria aggregation procedure for mixed evaluations

We introduce a new distance measure between two preorders that captures indifference, strict preference, weak preference and incomparability relations. This measure is the first to capture weak preference relations. We illustrate how this distance measure affords decision makers greater modeling power to capture their preferences, or uncertainty and ambiguity around them, by using our proposed distance measure in a multiple criteria aggregation procedure for mixed evaluations.     

Normalizing rank aggregation method for priority of a fuzzy preference relation and its effectiveness

The aim of this paper is to show that the normalizing rank aggregation method can not only be used to derive the priority vector for a multiplicative preference relation, but also for the additive transitive fuzzy preference relation. To do so, a simple functional equation between fuzzy preference’s element and priority weight is derived firstly, then, based on the equation, three methods are proposed to prove that the normalizing rank aggregation method is simple and effective for deriving the priority vector. Finally, a numerical example is used to illustrate the proposed methods.     

Enabling e-transactions with multi-attribute preference models

This paper describes potential applications of multi-attribute preference models (MAPM) in e-commerce and offers some guidelines for their implementation. MAPM are methodologies for modeling complex preferences that depend on more than one attribute or criterion, and include multi-attribute utility theory, conjoint analysis, and the Analytic Hierarchy Process. There are numerous examples of applications in e-commerce that would benefit from the acquisition of information regarding the preferences of a consumer, a customer, an advice seeker, or a decision maker. Here, the focus is on applications of MAPM models in B2C and B2B websites, where preferences of consumers are assessed for the purpose of identifying products or services that closely match their needs.     

Multiattribute preference models with reference points

In the context of multiple attribute decision making, preference models making use of reference points in an ordinal way have recently been introduced in the literature. This text proposes an axiomatic analysis of such models, with a particular emphasis on the case in which there is only one reference point. Our analysis uses a general conjoint measurement model resting on the study of traces induced on attributes by the preference relation and using conditions guaranteeing that these traces are complete. Models using reference points are shown to be a particular case of this general model. The number of reference points is linked to the number of equivalence classes distinguished by the traces. When there is only one reference point, the induced traces are quite rough, distinguishing at most two distinct equivalence classes. We study the relation between the model using a single reference point and other preference models proposed in the literature, most notably models based on concordance and models based on a discrete Sugeno integral.     

Scaling limits for internal aggregation models with multiple sources

We study the scaling limits of three different aggregation models on ℤ ^d : internal DLA, in which particles perform random walks until reaching an unoccupied site; the rotor-router model, in which particles perform deterministic analogues of random walks; and the divisible sandpile, in which each site distributes its excess mass equally among its neighbors. As the lattice spacing tends to zero, all three models are found to have the same scaling limit, which we describe as the solution to a certain PDE free boundary problem in ℝ ^d . In particular, internal DLA has a deterministic scaling limit. We find that the scaling limits are quadrature domains, which have arisen independently in many fields such as potential theory and fluid dynamics. Our results apply both to the case of multiple point sources and to the Diaconis-Fulton smash sum of domains.     

A criterion for time aggregation in intertemporal dynamic models

Nonlinear intertemporal general equilibrium models are hard to solve because of the dimensionality of the optimization problem involved. The computation of intertemporal general equilibria therefore calls for time-aggregation assumptions. A question then immediately arises: what criterion should one use to choose a sequence of possibly unequal time intervals in order to reduce the dimensionality of the optimization problem, yet keep under control the errors resulting from the numerical approximation of a continuous time process by a discrete time process? We propose one such criterion based on the current value of capital, which exploits near steady-state optimal dynamics. We show, using a parameterized version of the standard Ramsey—Koopmans—Cass model of optimal growth, that it outperforms alternative criterions used in the literature.     

Non-classical preference models in combinatorial problems: Models and algorithms for graphs

This is a summary of the most important results presented in the authors PhD thesis (Spanjaard 2003). This thesis, written in French, was defended on 16 December 2003 and supervised by Patrice Perny. A copy is available from the author upon request. This thesis deals with the search for preferred solutions in combinatorial optimization problems (and more particularly graph problems). It aims at conciliating preference modelling and algorithmic concerns for decision aiding.Received: March 2004, MSC classification: 91B06, 90C27, 90B40, 16Y60     

Temporal aggregation in structural VAR models

This paper examines the effects of temporal aggregation on the estimated time series properties of economic data. Theory predicts that temporal aggregation loses information about the underlying data processes. We derive low frequency, quarterly and annual, models implied by high frequency, monthly, structural vector autoregressive (SVAR) models and we find that these losses in information are substantial. It is shown that the accuracy of both the estimates and the forecasts of this class of models improve substantially when monthly data are used. Moreover, the aggregated data show more long-run persistence than the underlying disaggregated data. © 1998 John Wiley & Sons, Ltd.     

A new perspective on implementation by voting trees

Voting trees describe an iterative procedure for selecting a single vertex from a tournament. They provide a very general abstract model of decision‐making among a group of individuals, and it has therefore been studied which voting rules have a tree that implements them, i.e., chooses according to the rule for every tournament. While partial results concerning implementable rules and necessary conditions for implementability have been obtained over the past 40 years, a complete characterization of voting rules implementable by trees has proven surprisingly hard to find. A prominent rule that cannot be implemented by trees is the Copeland rule, which singles out vertices with maximum degree. In this paper, we suggest a new angle of attack and re‐examine the implementability of the Copeland solution using paradigms and techniques that are at the core of theoretical computer science. We study the extent to which voting trees can approximate the maximum degree in a tournament, and give upper and lower bounds on the worst‐case ratio between the degree of the vertex chosen by a tree and the maximum degree, both for the deterministic model concerned with a single fixed tree, and for randomizations over arbitrary sets of trees. Our main positive result is a randomization over surjective trees of polynomial size that provides an approximation ratio of at least 1/2. The proof is based on a connection between a randomization over caterpillar trees and a rapidly mixing Markov chain. © 2010 Wiley Periodicals, Inc. Random Struct. Alg., 39, 59–82, 2011     

Fuzzy set approach to the utility,preference relations,and aggregation operators

Score x = (x₁, … , x_n) describing an alternative α is modelled by means of a continuous quasi-convex fuzzy quantity μ_α = μ_x, thus allowing to compare alternatives (scores) by means of fuzzy ordering (comparison) methods. Applying some defuzzification method leads to the introduction of operators acting on scores. A special stress is put on the Mean of Maxima defuzzification method allowing to introduce several averaging aggregation operators. Moreover, our approach allows to introduce weights into above mentioned aggregation, even in the non-anonymous (non-symmetric) case. Finally, Ordered Weighted Aggregation Operators (OWAO) are introduced, generalizing the standard OWA operators.     

Preferences for harmony and minority voting

We examine repeated collective decisions in which people care about how others are treated. We formulate the notion of preferences for harmony as the desire of a person that others should not suffer repeatedly from the tyranny of majorities. We suggest that minority voting is a suitable scheme for such problems. Under such a scheme, only members of the minority in one decision are allowed to vote in the next decision. We discuss a variety of applications and avenues for future research.     

Evaluating correct classification probability for weighted voting classifiers with plurality voting

Weighted voting classifiers (WVCs) consist of N units that each provide individual classification decisions. The entire system output is based on tallying the weighted votes for each decision and choosing the winning one (plurality voting) or one which has the total weight of supporting votes greater than some specified threshold (threshold voting). Each individual unit may abstain from voting. The entire system may also abstain from voting if no decision is ultimately winning. Existing methods of evaluating the correct classification probability (CCP) of WVCs can be applied to limited special cases of these systems (threshold voting) and impose some restrictions on their parameters. In this paper a method is suggested which allows the CCP of WVCs with both plurality and threshold voting to be exactly evaluated without imposing constraints on unit weights. The method is based on using the modified universal generating function technique.     

Some comments on preference order dynamic programming models

A simple deterministic dynamic programming model is used as a general framework for the analysis of stochastic versions of three classical optimization problems: knapsack, traveling salesperson, and assembly line balancing problems. It is shown that this model can provide an alternative to the preference order models proposed for these problems. Counterexample to the optimality of the preference order models are presented.     

Distance-based and ad hoc consensus models in ordinal preference ranking

This paper examines the problem of aggregating ordinal preferences on a set of alternatives into a consensus. This problem has been the subject of study for more than two centuries and many procedures have been developed to create a compromise or consensus.We examine a variety of structures for preference specification, and in each case review the related models for deriving a consensus. Two classes of consensus models are discussed, namely ad hoc methods, evolving primarily from parliamentary settings over the past 200 years, and distance or axiomatic-based methods. We demonstrate the levels of complexity of the various distance-based models by presenting the related mathematical programming formulations for them. We also present conditions for equivalence, that is, for yielding the same consensus ranking for some of the methods. Finally, we discuss various extensions of the basic ordinal ranking structures, paying specific attention to partial ranking, voting member weighted consensus, ranking with intensity of preference, and rank correlation methods, as alternative approaches to deriving a consensus. Suggestions for future research directions are given.