Improving Collaborative Filtering Based Recommender Systems Using Pareto Dominance

Los sistemas de recomendación son un tipo de solución al problema de sobrecarga de información que sufren los usuarios de los sitios web en los que se pueden votar ciertos artículos. El sistema de recomendación de filtrado colaborativo es considerado como el método con más éxito debido a que sus recomendaciones se hacen basándose en los votos de usuarios similares a un usuario activo. Sin embargo, el método de filtrado de colaboración tradicional selecciona usuarios insuficientemente representativos como vecinos de cada usuario activo. Esto significa que las recomendaciones hechas a posteriori no son lo suficientemente precisas. El método propuesto en esta tesis realiza un pre-filtrado del proceso, mediante el uso de dominancia de Pareto, que elimina los usuarios menos representativos del proceso de selección k-vecino y mantiene los más prometedores. Los resultados de los experimentos realizados en MovieLens y Netflix muestran una mejora significativa en todas las medidas de calidad estudiadas en la aplicación del método propuesto. ABSTRACTRecommender systems are a type of solution to the information overload problem suffered by users of websites on which they can rate certain items. The Collaborative Filtering Recommender System is considered to be the most successful approach as it make its recommendations based on votes of users similar to an active user. Nevertheless, the traditional collaborative filtering method selects insufficiently representative users as neighbors of each active user. This means that the recommendations made a posteriori are not precise enough. The method proposed in this thesis performs a pre-filtering process, by using Pareto dominance, which eliminates the less representative users from the k-neighbor selection process and keeps the most promising ones. The results from the experiments performed on Movielens and Netflix show a significant improvement in all the quality measures studied on applying the proposed method.

Regularized model learning in EDAs for continuous and multi-objective optimization

Probabilistic modeling is the de�ning characteristic of estimation of distribution algorithms (EDAs) which determines their behavior and performance in optimization. Regularization is a well-known statistical technique used for obtaining an improved model by reducing the generalization error of estimation, especially in high-dimensional problems. `1-regularization is a type of this technique with the appealing variable selection property which results in sparse model estimations. In this thesis, we study the use of regularization techniques for model learning in EDAs. Several methods for regularized model estimation in continuous domains based on a Gaussian distribution assumption are presented, and analyzed from di�erent aspects when used for optimization in a high-dimensional setting, where the population size of EDA has a logarithmic scale with respect to the number of variables. The optimization results obtained for a number of continuous problems with an increasing number of variables show that the proposed EDA based on regularized model estimation performs a more robust optimization, and is able to achieve signi�cantly better results for larger dimensions than other Gaussian-based EDAs. We also propose a method for learning a marginally factorized Gaussian Markov random �eld model using regularization techniques and a clustering algorithm. The experimental results show notable optimization performance on continuous additively decomposable problems when using this model estimation method. Our study also covers multi-objective optimization and we propose joint probabilistic modeling of variables and objectives in EDAs based on Bayesian networks, speci�cally models inspired from multi-dimensional Bayesian network classi�ers. It is shown that with this approach to modeling, two new types of relationships are encoded in the estimated models in addition to the variable relationships captured in other EDAs: objectivevariable and objective-objective relationships. An extensive experimental study shows the e�ectiveness of this approach for multi- and many-objective optimization. With the proposed joint variable-objective modeling, in addition to the Pareto set approximation, the algorithm is also able to obtain an estimation of the multi-objective problem structure. Finally, the study of multi-objective optimization based on joint probabilistic modeling is extended to noisy domains, where the noise in objective values is represented by intervals. A new version of the Pareto dominance relation for ordering the solutions in these problems, namely �-degree Pareto dominance, is introduced and its properties are analyzed. We show that the ranking methods based on this dominance relation can result in competitive performance of EDAs with respect to the quality of the approximated Pareto sets. This dominance relation is then used together with a method for joint probabilistic modeling based on `1-regularization for multi-objective feature subset selection in classi�cation, where six di�erent measures of accuracy are considered as objectives with interval values. The individual assessment of the proposed joint probabilistic modeling and solution ranking methods on datasets with small-medium dimensionality, when using two di�erent Bayesian classi�ers, shows that comparable or better Pareto sets of feature subsets are approximated in comparison to standard methods.

Interval-based ranking in noisy evolutionary multiobjective optimization

As one of the most competitive approaches to multi-objective optimization, evolutionary algorithms have been shown to obtain very good results for many realworld multi-objective problems. One of the issues that can affect the performance of these algorithms is the uncertainty in the quality of the solutions which is usually represented with the noise in the objective values. Therefore, handling noisy objectives in evolutionary multi-objective optimization algorithms becomes very important and is gaining more attention in recent years. In this paper we present ?-degree Pareto dominance relation for ordering the solutions in multi-objective optimization when the values of the objective functions are given as intervals. Based on this dominance relation, we propose an adaptation of the non-dominated sorting algorithm for ranking the solutions. This ranking method is then used in a standardmulti-objective evolutionary algorithm and a recently proposed novel multi-objective estimation of distribution algorithm based on joint variable-objective probabilistic modeling, and applied to a set of multi-objective problems with different levels of independent noise. The experimental results show that the use of the proposed method for solution ranking allows to approximate Pareto sets which are considerably better than those obtained when using the dominance probability-based ranking method, which is one of the main methods for noise handling in multi-objective optimization.

Using quantitative descriptive analysis and temporal dominance of sensations analysis as complementary methods for profiling commercial blackcurrant squashes

Quantitative descriptive analysis (QDA) is used to describe the nature and the intensity of sensory properties from a single evaluation of a product, whereas temporal dominance of sensation (TDS) is primarily used to identify dominant sensory properties over time. Previous studies with TDS have focused on model systems, but this is the first study to use a sequential approach, i.e. QDA then TDS in measuring sensory properties of a commercial product category, using the same set of trained assessors (n = 11). The main objectives of this study were to: (1) investigate the benefits of using a sequential approach of QDA and TDS and (2) to explore the impact of the sample composition on taste and flavour perceptions in blackcurrant squashes. The present study has proposed an alternative way of determining the choice of attributes for TDS measurement based on data obtained from previous QDA studies, where available. Both methods indicated that the flavour profile was primarily influenced by the level of dilution and complexity of sample composition combined with blackcurrant juice content. In addition, artificial sweeteners were found to modify the quality of sweetness and could also contribute to bitter notes. Using QDA and TDS in tandem was shown to be more beneficial than each just on its own enabling a more complete sensory profile of the products.

Dominance Measuring Method Performance under Incomplete Information about Weights.

In multi-attribute utility theory, it is often not easy to elicit precise values for the scaling weights representing the relative importance of criteria. A very widespread approach is to gather incomplete information. A recent approach for dealing with such situations is to use information about each alternative?s intensity of dominance, known as dominance measuring methods. Different dominancemeasuring methods have been proposed, and simulation studies have been carried out to compare these methods with each other and with other approaches but only when ordinal information about weights is available. In this paper, we useMonte Carlo simulation techniques to analyse the performance of and adapt such methods to deal with weight intervals, weights fitting independent normal probability distributions orweights represented by fuzzy numbers.Moreover, dominance measuringmethod performance is also compared with a widely used methodology dealing with incomplete information on weights, the stochastic multicriteria acceptability analysis (SMAA). SMAA is based on exploring the weight space to describe the evaluations that would make each alternative the preferred one.

Gestión óptima de embalses en avenidas incorporando al concepto de incertidumbre : aplicación a embalses con central hidroeléctrica

Esta tesis realiza una contribución metodológica al problema de la gestión óptima de embalses hidroeléctricos durante eventos de avenidas, considerando un enfoque estocástico y multiobjetivo. Para ello se propone una metodología de evaluación de estrategias de laminación en un contexto probabilístico y multiobjetivo. Además se desarrolla un entorno dinámico de laminación en tiempo real con pronósticos que combina un modelo de optimización y algoritmos de simulación. Estas herramientas asisten a los gestores de las presas en la toma de decisión respecto de cuál es la operación más adecuada del embalse. Luego de una detallada revisión de la bibliografía, se observó que los trabajos en el ámbito de la gestión óptima de embalses en avenidas utilizan, en general, un número reducido de series de caudales o hidrogramas para caracterizar los posibles escenarios. Limitando el funcionamiento satisfactorio de un modelo determinado a situaciones hidrológicas similares. Por otra parte, la mayoría de estudios disponibles en este ámbito abordan el problema de la laminación en embalses multipropósito durante la temporada de avenidas, con varios meses de duración. Estas características difieren de la realidad de la gestión de embalses en España. Con los avances computacionales en materia de gestión de información en tiempo real, se observó una tendencia a la implementación de herramientas de operación en tiempo real con pronósticos para determinar la operación a corto plazo (involucrando el control de avenidas). La metodología de evaluación de estrategias propuesta en esta tesis se basa en determinar el comportamiento de éstas frente a un espectro de avenidas características de la solicitación hidrológica. Con ese fin, se combina un sistema de evaluación mediante indicadores y un entorno de generación estocástica de avenidas, obteniéndose un sistema implícitamente estocástico. El sistema de evaluación consta de tres etapas: caracterización, síntesis y comparación, a fin de poder manejar la compleja estructura de datos resultante y realizar la evaluación. En la primera etapa se definen variables de caracterización, vinculadas a los aspectos que se quieren evaluar (seguridad de la presa, control de inundaciones, generación de energía, etc.). Estas variables caracterizan el comportamiento del modelo para un aspecto y evento determinado. En la segunda etapa, la información de estas variables se sintetiza en un conjunto de indicadores, lo más reducido posible. Finalmente, la comparación se lleva a cabo a partir de la comparación de esos indicadores, bien sea mediante la agregación de dichos objetivos en un indicador único, o bien mediante la aplicación del criterio de dominancia de Pareto obteniéndose un conjunto de soluciones aptas. Esta metodología se aplicó para calibrar los parámetros de un modelo de optimización de embalse en laminación y su comparación con otra regla de operación, mediante el enfoque por agregación. Luego se amplió la metodología para evaluar y comparar reglas de operación existentes para el control de avenidas en embalses hidroeléctricos, utilizando el criterio de dominancia. La versatilidad de la metodología permite otras aplicaciones, tales como la determinación de niveles o volúmenes de seguridad, o la selección de las dimensiones del aliviadero entre varias alternativas. Por su parte, el entorno dinámico de laminación al presentar un enfoque combinado de optimización-simulación, permite aprovechar las ventajas de ambos tipos de modelos, facilitando la interacción con los operadores de las presas. Se mejoran los resultados respecto de los obtenidos con una regla de operación reactiva, aun cuando los pronósticos se desvían considerablemente del hidrograma real. Esto contribuye a reducir la tan mencionada brecha entre el desarrollo teórico y la aplicación práctica asociada a los modelos de gestión óptima de embalses. This thesis presents a methodological contribution to address the problem about how to operate a hydropower reservoir during floods in order to achieve an optimal management considering a multiobjective and stochastic approach. A methodology is proposed to assess the flood control strategies in a multiobjective and probabilistic framework. Additionally, a dynamic flood control environ was developed for real-time operation, including forecasts. This dynamic platform combines simulation and optimization models. These tools may assist to dam managers in the decision making process, regarding the most appropriate reservoir operation to be implemented. After a detailed review of the bibliography, it was observed that most of the existing studies in the sphere of flood control reservoir operation consider a reduce number of hydrographs to characterize the reservoir inflows. Consequently, the adequate functioning of a certain strategy may be limited to similar hydrologic scenarios. In the other hand, most of the works in this context tackle the problem of multipurpose flood control operation considering the entire flood season, lasting some months. These considerations differ from the real necessity in the Spanish context. The implementation of real-time reservoir operation is gaining popularity due to computational advances and improvements in real-time data management. The methodology proposed in this thesis for assessing the strategies is based on determining their behavior for a wide range of floods, which are representative of the hydrological forcing of the dam. An evaluation algorithm is combined with a stochastic flood generation system to obtain an implicit stochastic analysis framework. The evaluation system consists in three stages: characterizing, synthesizing and comparing, in order to handle the complex structure of results and, finally, conduct the evaluation process. In the first stage some characterization variables are defined. These variables should be related to the different aspects to be evaluated (such as dam safety, flood protection, hydropower, etc.). Each of these variables characterizes the behavior of a certain operating strategy for a given aspect and event. In the second stage this information is synthesized obtaining a reduced group of indicators or objective functions. Finally, the indicators are compared by means of an aggregated approach or by a dominance criterion approach. In the first case, a single optimum solution may be achieved. However in the second case, a set of good solutions is obtained. This methodology was applied for calibrating the parameters of a flood control model and to compare it with other operating policy, using an aggregated method. After that, the methodology was extent to assess and compared some existing hydropower reservoir flood control operation, considering the Pareto approach. The versatility of the method allows many other applications, such as determining the safety levels, defining the spillways characteristics, among others. The dynamic framework for flood control combines optimization and simulation models, exploiting the advantages of both techniques. This facilitates the interaction between dam operators and the model. Improvements are obtained applying this system when compared with a reactive operating policy, even if the forecasts deviate significantly from the observed hydrograph. This approach contributes to reduce the gap between the theoretical development in the field of reservoir management and its practical applications.

Relating dominance of dialogue participants with their verbal intelligence scores

In this work we investigated whether there is a relationship between dominant behaviour of dialogue participants and their verbal intelligence. The analysis is based on a corpus containing 56 dialogues and verbal intelligence scores of the test persons. All the dialogues were divided into three groups: H-H is a group of dialogues between higher verbal intelligence participants, L-L is a group of dialogues between lower verbal intelligence participant and L-H is a group of all the other dialogues. The dominance scores of the dialogue partners from each group were analysed. The analysis showed that differences between dominance scores and verbal intelligence coefficients for L-L were positively correlated. Verbal intelligence scores of the test persons were compared to other features that may reflect dominant behaviour. The analysis showed that number of interruptions, long utterances, times grabbed the floor, influence diffusion model, number of agreements and several acoustic features may be related to verbal intelligence. These features were used for the automatic classification of the dialogue partners into two groups (lower and higher verbal intelligence participants); the achieved accuracy was 89.36%.

Dominance intensity measure within fuzzy weight oriented MAUT: an application

We introduce a dominance intensity measuring method to derive a ranking of alternatives to deal with incomplete information in multi-criteria decision-making problems on the basis of multi-attribute utility theory (MAUT) and fuzzy sets theory. We consider the situation where there is imprecision concerning decision-makers’ preferences, and imprecise weights are represented by trapezoidal fuzzy weights.The proposed method is based on the dominance values between pairs of alternatives. These values can be computed by linear programming, as an additive multi-attribute utility model is used to rate the alternatives. Dominance values are then transformed into dominance intensity measures, used to rank the alternatives under consideration. Distances between fuzzy numbers based on the generalization of the left and right fuzzy numbers are utilized to account for fuzzy weights. An example concerning the selection of intervention strategies to restore an aquatic ecosystem contaminated by radionuclides illustrates the approach. Monte Carlo simulation techniques have been used to show that the proposed method performs well for different imprecision levels in terms of a hit ratio and a rank-order correlation measure.

A new dominance intensity method to deal with ordinal information about a DM's preferences within MAVT

Dominance measuring methods are a new approach to deal with complex decision-making problems with imprecise information. These methods are based on the computation of pairwise dominance values and exploit the information in the dominance matrix in dirent ways to derive measures of dominance intensity and rank the alternatives under consideration. In this paper we propose a new dominance measuring method to deal with ordinal information about decision-maker preferences in both weights and component utilities. It takes advantage of the centroid of the polytope delimited by ordinal information and builds triangular fuzzy numbers whose distances to the crisp value 0 constitute the basis for the de?nition of a dominance intensity measure. Monte Carlo simulation techniques have been used to compare the performance of this method with other existing approaches.

Dominance intensity methods for ranking of alternatives in mcdm with imprecise information

Los decisores cada vez se enfrentan a problemas más complejos en los que tomar una decisión implica tener que considerar simultáneamente muchos criterios que normalmente son conflictivos entre sí. En la mayoría de los problemas de decisión es necesario considerar criterios económicos, sociales y medioambientales. La Teoría de la Decisión proporciona el marco adecuado para poder ayudar a los decisores a resolver estos problemas de decisión complejos, al permitir considerar conjuntamente la incertidumbre existente sobre las consecuencias de cada alternativa en los diferentes atributos y la imprecisión sobre las preferencias de los decisores. En esta tesis doctoral nos centramos en la imprecisión de las preferencias de los decisores cuando éstas pueden ser representadas mediante una función de utilidad multiatributo aditiva. Por lo tanto, consideramos imprecisión tanto en los pesos como en las funciones de utilidad componentes de cada atributo. Se ha considerado el caso en que la imprecisión puede ser representada por intervalos de valores o bien mediante información ordinal, en lugar de proporcionar valores concretos. En este sentido, hemos propuesto métodos que permiten ordenar las diferentes alternativas basados en los conceptos de intensidad de dominación o intensidad de preferencia, los cuales intentan medir la fuerza con la que cada alternativa es preferida al resto. Para todos los métodos propuestos se ha analizado su comportamiento y se ha comparado con los más relevantes existentes en la literatura científica que pueden ser aplicados para resolver este tipo de problemas. Para ello, se ha realizado un estudio de simulación en el que se han usado dos medidas de eficiencia (hit ratio y coeficiente de correlación de Kendall) para comparar los diferentes métodos. ABSTRACT Decision makers increasingly face complex decision-making problems where they have to simultaneously consider many often conflicting criteria. In most decision-making problems it is necessary to consider economic, social and environmental criteria. Decision making theory provides an adequate framework for helping decision makers to make complex decisions where they can jointly consider the uncertainty about the performance of each alternative for each attribute, and the imprecision of the decision maker's preferences. In this PhD thesis we focus on the imprecision of the decision maker's preferences represented by an additive multiattribute utility function. Therefore, we consider the imprecision of weights, as well as of component utility functions for each attribute. We consider the case in which the imprecision is represented by ranges of values or by ordinal information rather than precise values. In this respect, we propose methods for ranking alternatives based on notions of dominance intensity, also known as preference intensity, which attempt to measure how much more preferred each alternative is to the others. The performance of the propose methods has been analyzed and compared against the leading existing methods that are applicable to this type of problem. For this purpose, we conducted a simulation study using two efficiency measures (hit ratio and Kendall correlation coefficient) to compare the different methods.

Inteligencia artificial para la predicción y control del acabado superficial en procesos de fresado a alta velocidad

El objetivo principal de esta tesis doctoral es profundizar en el análisis y diseño de un sistema inteligente para la predicción y control del acabado superficial en un proceso de fresado a alta velocidad, basado fundamentalmente en clasificadores Bayesianos, con el prop´osito de desarrollar una metodolog´ıa que facilite el diseño de este tipo de sistemas. El sistema, cuyo propósito es posibilitar la predicción y control de la rugosidad superficial, se compone de un modelo aprendido a partir de datos experimentales con redes Bayesianas, que ayudar´a a comprender los procesos dinámicos involucrados en el mecanizado y las interacciones entre las variables relevantes. Dado que las redes neuronales artificiales son modelos ampliamente utilizados en procesos de corte de materiales, también se incluye un modelo para fresado usándolas, donde se introdujo la geometría y la dureza del material como variables novedosas hasta ahora no estudiadas en este contexto. Por lo tanto, una importante contribución en esta tesis son estos dos modelos para la predicción de la rugosidad superficial, que se comparan con respecto a diferentes aspectos: la influencia de las nuevas variables, los indicadores de evaluación del desempeño, interpretabilidad. Uno de los principales problemas en la modelización con clasificadores Bayesianos es la comprensión de las enormes tablas de probabilidad a posteriori producidas. Introducimos un m´etodo de explicación que genera un conjunto de reglas obtenidas de árboles de decisión. Estos árboles son inducidos a partir de un conjunto de datos simulados generados de las probabilidades a posteriori de la variable clase, calculadas con la red Bayesiana aprendida a partir de un conjunto de datos de entrenamiento. Por último, contribuimos en el campo multiobjetivo en el caso de que algunos de los objetivos no se puedan cuantificar en números reales, sino como funciones en intervalo de valores. Esto ocurre a menudo en aplicaciones de aprendizaje automático, especialmente las basadas en clasificación supervisada. En concreto, se extienden las ideas de dominancia y frontera de Pareto a esta situación. Su aplicación a los estudios de predicción de la rugosidad superficial en el caso de maximizar al mismo tiempo la sensibilidad y la especificidad del clasificador inducido de la red Bayesiana, y no solo maximizar la tasa de clasificación correcta. Los intervalos de estos dos objetivos provienen de un m´etodo de estimación honesta de ambos objetivos, como e.g. validación cruzada en k rodajas o bootstrap.---ABSTRACT---The main objective of this PhD Thesis is to go more deeply into the analysis and design of an intelligent system for surface roughness prediction and control in the end-milling machining process, based fundamentally on Bayesian network classifiers, with the aim of developing a methodology that makes easier the design of this type of systems. The system, whose purpose is to make possible the surface roughness prediction and control, consists of a model learnt from experimental data with the aid of Bayesian networks, that will help to understand the dynamic processes involved in the machining and the interactions among the relevant variables. Since artificial neural networks are models widely used in material cutting proceses, we include also an end-milling model using them, where the geometry and hardness of the piecework are introduced as novel variables not studied so far within this context. Thus, an important contribution in this thesis is these two models for surface roughness prediction, that are then compared with respecto to different aspects: influence of the new variables, performance evaluation metrics, interpretability. One of the main problems with Bayesian classifier-based modelling is the understanding of the enormous posterior probabilitiy tables produced. We introduce an explanation method that generates a set of rules obtained from decision trees. Such trees are induced from a simulated data set generated from the posterior probabilities of the class variable, calculated with the Bayesian network learned from a training data set. Finally, we contribute in the multi-objective field in the case that some of the objectives cannot be quantified as real numbers but as interval-valued functions. This often occurs in machine learning applications, especially those based on supervised classification. Specifically, the dominance and Pareto front ideas are extended to this setting. Its application to the surface roughness prediction studies the case of maximizing simultaneously the sensitivity and specificity of the induced Bayesian network classifier, rather than only maximizing the correct classification rate. Intervals in these two objectives come from a honest estimation method of both objectives, like e.g. k-fold cross-validation or bootstrap.

Dominance measuring methods within MAVT/MAUT with imprecise information concerning decision-makers'preferences

Dominance measuring methods are an approach for dealing with complex decision-making problems with imprecise information within multi-attribute value/utility theory. These methods are based on the computation of pairwise dominance values and exploit the information in the dominance matrix in different ways to derive measures of dominance intensity and rank the alternatives under consideration. In this paper we review dominance measuring methods proposed in the literature for dealing with imprecise information (intervals, ordinal information or fuzzy numbers) about decision-makers? preferences and their performance in comparison with other existing approaches, like SMAA and SMAA-II or Sarabando and Dias? method.

Veto values in Group Decision Making within MAUT: aggregating complete rankings derived from dominance intensity measures

We consider a groupdecision-making problem within multi-attribute utility theory, in which the relative importance of decisionmakers (DMs) is known and their preferences are represented by means of an additive function. We allow DMs to provide veto values for the attribute under consideration and build veto and adjust functions that are incorporated into the additive model. Veto functions check whether alternative performances are within the respective veto intervals, making the overall utility of the alternative equal to 0, where as adjust functions reduce the utilty of the alternative performance to match the preferences of other DMs. Dominance measuring methods are used to account for imprecise information in the decision-making scenario and to derive a ranking of alternatives for each DM. Specifically, ordinal information about the relative importance of criteria is provided by each DM. Finally, an extension of Kemeny's method is used to aggregate the alternative rankings from the DMs accounting for the irrelative importance.