Practical Bayesian estimation of a finite beta mixture through gibbs sampling and its applications

Bouguila, Nizar; Ziou, Djemel; Monga, Ernest

doi:10.1007/s11222-006-8451-7

Cited by 104 publications

(65 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In fact, in contrast to the gamma distribution, the beta distribution is very versatile and it is capable of modelling a variety of uncertainties. Specifically, the gamma distribution may be L-shaped, skewed to the right, or symmetric, while the beta distribution may be Lshaped, U-shaped, J-shaped, skewed to the left, skewed to the right, or symmetric [Bouguila et al 2006]. The shapes of Gaussian, gamma and uniform distributions are special cases of the beta distribution [Boutemedjet et al 2011].…”

Section: Related Workmentioning

confidence: 99%

“…We have used the multivariate beta mixtures mainly because the beta distribution offers considerable flexibility and ease of use [Ma 2011], [Ma and Leijon 2009], [Bouguila et al 2006]. In fact, in contrast with other distributions such as the Gaussian, which permit only a symmetric shape, the beta distribution has a flexible shape; it can be symmetric, asymmetric, or convex.…”

Section: Introductionmentioning

confidence: 99%

“…In fact, in contrast with other distributions such as the Gaussian, which permit only a symmetric shape, the beta distribution has a flexible shape; it can be symmetric, asymmetric, or convex. The shapes of the beta distribution are variable enough to allow for an approximation of almost any arbitrary distribution [Bouguila et al 2006]. It is worth noting that, in contrast to the theoretical work that exists on the multivariate beta distribution [Olkin and Rubin 1964], very little work has been done on its practical applications [Ma 2011].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Identifying Authorities in Online Communities

Bouguessa

Romdhane

2015

ACM Trans. Intell. Syst. Technol.

View full text Add to dashboard Cite

Several approaches have been proposed for the problem of identifying authoritative actors in online communities. However, the majority of existing methods suffer from one or more of the following limitations: (1) there is a lack of an automatic mechanism to formally discriminate between authoritative and nonauthoritative users. In fact, a common approach to authoritative user identification is to provide a ranked list of users expecting authorities to come first. A major problem of such an approach is where to stop reading the ranked list of users? How many users should be chosen as authoritative? (2) supervised learning approaches for authoritative user identification suffer from their dependency on the training data. The problem here is that labeled samples are more difficult, expensive, and time consuming to obtain than unlabeled ones. (3) several approaches rely on some user parameters to estimate an authority score. Detection accuracy of authoritative users can be seriously affected if incorrect values are used. In this paper, we propose a parameterless mixture model-based approach which is capable of addressing the three aforementioned issues in a single framework. In our approach, we first represent each user with a feature vector composed of information related to its social behaviour and activity in an online community. Next, we propose a statistical framework, based on the multivariate beta mixtures, in order to model the estimated set of feature vectors. The probability density function is therefore estimated and the beta component that corresponds to the most authoritative users is identified. The suitability of the proposed approach is illustrated on real data extracted from Stack Exchange question-answering network and Twitter.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Identifying Authorities in Online Communities

Bouguessa

Romdhane

2015

ACM Trans. Intell. Syst. Technol.

View full text Add to dashboard Cite

show abstract

“…Bouguila et al [21] show that combining FCM and MM provides good initial parameters for the EM algorithm to converge.…”

Section: B Background: Finite Mixture Modelsmentioning

confidence: 99%

Behind the Curtain: Service Selection via Trust in Composite Services

Hang

Kalia

Singh

2012

2012 IEEE 19th International Conference on Web Services

View full text Add to dashboard Cite

Abstract-Service selection, where some of the services are accessed indirectly as constituents of composite services, is difficult for the following reasons: (1) the interpretation of service qualities is subjective; (2) evidence must be combined from multiple sources; (3) service profiles change dynamically; and (4) constituent services may be only partially observable behind composite services. We propose an approach where we map service qualities to a common probabilistic trust metric. Whereas current trust approaches estimate the trustworthiness of a composite service based on a fully observable and static setting, we propose a statistical approach built on expectation maximized over a finite mixture model. Our experiments show that our approach can dynamically punish or reward the constituents of composite services while making only partial observations.

show abstract

“…This new model includes the Beta regression model and the variable dispersion Beta regression model as particular cases. In addition, following a robust statistical modeling approach (Pinheiro et al 2001) we will show that Beta rectangular regression is more C. Bayes, J. Bazán and C. García 773 robust than the Beta regression model.The Beta rectangular distribution is just a mixture of a Beta distribution with a Uniform distribution and thus is a finite Beta mixture model as given by Bouguila et al (2006). It is well-known that mixture distributions are more robust to outliers (comparatively large or influential values) since by including an extra distributional component in the model, the variability is better accounted for and the estimation of the "true" mean parameter is less affected, as indicated by Markatou (2000).…”

mentioning

confidence: 99%

Contribuições ao estudo de variáveis dependentes limitadas contínuas e categóricas

Bazán¹

View full text Add to dashboard Cite

This text presents and analyzes a number of papers relating to works developed under the guidance of the author of this document. The papers are related to the project of development of new distributions to variables with limited response including discrete and continuous response as well as the modeling of such variables as function of covariates considering regression and latent variables models. Every paper to be presented is contextualized within three areas of the limited response: regression models for continuous limited response, regression models for categorical limited response and latent limited regression models (models of item response theory). The proposed developments are described as well as simulation studies and applications that illustrate the relevance the proposals in the corresponding areas. Moreover, are presented works that are in progress, but already they were submitted and are under review phase. However, the text summarizes the main contributions of the author to the area of modeling and analysis of regression performed acting as advisor of scientific research work, master's, doctoral and post-doctoral as well as the relationship with other associated researchers the past five years.Keywords: Modelling, regression analysis, inference, limited response, categorical data, bayesian estimation, item response theory.iii Sumário partido político por município em uma eleição, mas também o caso de resposta binária {0, 1} e o intervalo unitário aberto (0, 1). Quando tentamos modelar uma distribuição em que a variável resposta é assumida no intervalo aberto ou fechado, considerando covariáveis explicativas, estamos frente a um modelo de regressão para resposta limitada contínua. Quando focamos nas respostas tomando somente os valores zero e uns, estamos frente a uma variável binária e então o correspondente modelo é o de regressão biná-ria. As respostas binárias seguem um processo Bernoulli, assum-se a distribuição de uma variável correspondente a soma destes processos resulta em uma distribuição categórica.Este grupo de modelos faze parte dos modelos de regressão com resposta limitada categórica. Finalmente se temos um caso de respostas binárias multivariadas e assumimos como covariáveis ou fatores explicativos variáveis não observáveis ou latentes teremos uma regressão binária multivariada latente ou uma regressão limitada latente mais conhecida como modelos de teoria da resposta ao item. Um grupo importante de minhas pesqui-3 sas desenvolvidas nos últimos cinco anos fica classificada nessas três áreas ou classes de modelos que descrevemos a seguir em maiores detalhes. Modelos de regressão para resposta limitada contí-nuaEspecificamente, no caso do intervalo (0, 1). o principal modelo desenvolvido foi o modelo de regressão Beta o qual descrevemos a seguir.Considere que y = (y 1 , y 2 , . . . , y n ) denota um vetor n × 1 de n variáveis aleatórias independentes. Também seja x i = (x i1 , . . . , x in ) um vetor k × 1 de covariáveis, na qual x i1 pode ser 1, o que corresponde a um interc...

show abstract

Practical Bayesian estimation of a finite beta mixture through gibbs sampling and its applications

Cited by 104 publications

References 27 publications

Identifying Authorities in Online Communities

Identifying Authorities in Online Communities

Behind the Curtain: Service Selection via Trust in Composite Services

Contribuições ao estudo de variáveis dependentes limitadas contínuas e categóricas

Contact Info

Product

Resources

About