Kate Revoredo

Revision of first-order Bayesian classifiers

New representation languages that integrate first order logic with Bayesian networks have been proposed in the literature. Probabilistic Relational models (PRM) and Bayesian Logic Programs (BLP) are examples. Algorithms to learn both the qualitative and the quantitative components of these languages have been developed. Recently, we have developed an algorithm to revise a BLP. In this paper, we discuss the relationship among these approaches, extend our revision algorithm to return the highest probabilistic scoring BLP and argue that for a classification task our approach, which uses techniques of theory revision and so searches a smaller hypotheses space, can be a more adequate choice.

Search-Based Class Discretization for Hidden Markov Model for Regression

The regression-by-discretization approach allows the use of classification algorithm in a regression task. It works as a pre-processing step in which the numeric target value is discretized into a set of intervals. We had applied this approach to the Hidden Markov Model for Regression (HMMR) which was successfully compared to the Naive Bayes for Regression and two traditional forecasting methods, Box-Jenkins and Winters. In this work, to further improve these results, we apply three discretization methods to HMMR using ten time series data sets. The experimental results showed that one of the discretization methods improved the results in most of the data sets, although each method improved the results in at least one data set. Therefore, it would be better to have a search algorithm to automatically find the optimal number and width of the intervals.

PFORTE: revising probabilistic FOL theories

There has been significant recent progress in the integration of probabilistic reasoning with first order logic representations (SRL). So far, the learning algorithms developed for these models all learn from scratch, assuming an invariant background knowledge. As an alternative, theory revision techniques have been shown to perform well on a variety of machine learning problems. These techniques start from an approximate initial theory and apply modifications in places that performed badly in classification. In this work we describe the first revision system for SRL classification, PFORTE, which addresses two problems: all examples must be classified, and they must be classified well. PFORTE uses a two step-approach. The completeness component uses generalization operators to address failed proofs and the classification component addresses classification problems using generalization and specialization operators. Experimental results show significant benefits from using theory revision techniques compared to learning from scratch.

Text mining as a transparency enabler to support decision making in a people management process

This paper discusses a case study that was performed using mining techniques to analyze data pertinent to a people management process of a Federal University. This process consists in observing documents containing data concerning the organizational environment, the duties required by the position, the course program carried out by the employee, and whether they have direct or indirect correlation. Currently, this correlation evaluation is performed subjectively and there are no instruments that can indicate the degree of similarity between the information. We use text mining techniques to automatically identify correlation through textual representation approaches and syntactic and semantic modeling, which retrieve terms and dimension their respective meanings. To obtain the degree of similarity between the respective documents, the measure of the cosines similarity was used. The results showed that the documents evaluated as correlated by the domain expert presented a degree of similarity consistent with the automatic evaluation. For the uncorrelated cases, it was perceived that the degree of high similarity was influenced by the comprehensiveness of the organizational environment common to all documents. After investigation and identification of the appropriate environment specification, the grades obtained represented the evaluation correctly. The proposed approach contributes to the speed of process judgment, as well as to promote formulations of criticism about the content of political qualifications. In addition, it enhanced processes and information transparency by tracking and publicizing all steps. Lastly, we present a simulation for a course recommendation task, considering position profiles and organizational environment.

Improvement of transparency through mining techniques for reclassification of texts: the case of brazilian transparency portal

Several countries passed transparency laws requiring that governments make data about its fiscal and financial expenditures publicly available in Internet portals. Nevertheless, available data is not always synonymous with transparent data. This is the case of the Transparency Portal of Brazilian Federal Government, since key data is presented as unstructured text hindering the control of purchased items. This article describes the application of text mining techniques with the objective of reclassifying descriptive texts of unit of measurement related to the products and services procured by Federal Government in Brazil. The results of the efficacy of this model are presented, including the production of analysis based on the transformed dataset, identifying probable input errors, suspicious companies and purchasers and factors affecting procurement prices as well as presenting suggestions for future research and improvements for the way the data is inputted and made available

Speech Acts Featuring Decisions in Knowledge-Intensive Processes

A Knowledge-Intensive Process (KiP) is specified as a composition of a set of prospective activities (events) whose execution contributes to achieving a goal and whose control-flow, at the instance level, typically presents a high degree of variability among its several past executions. Variability is a consequence of a combination of decision points and informal interactions among participants on collaborative and innovative activities. These interactions may occur through message exchange, thus understanding the interplay of illocutionary acts within messages may bring insights on how participants make decisions. In this paper, we propose mechanisms that identify speech acts in the set of messages that mostly lead to decision points in a KiP providing an understanding of conversational patterns. We empirically evaluate our proposal considering data from a company that provides IT services to several customers.

Modeling and Using Context in Business Process Management: A Research Agenda

Business Process Management (BPM) is the art and science of monitoring how the work is performed within an organization to ensure consistent results and opportunities for improvement. One important research topic on BPM relates to the issue of flexibility. Unplanned conditions may occur at any time during process execution. In dynamic environments, changes must be performed more frequently and systematically and considering not only pre-established rules, but also contextual information. The literature indicates context as a source of information that should be considered in the modeling of business processes in order to contribute to their flexibility when in the execution phase. In this paper, we present the results from our studies about context in the whole cycle of BPM. The goal is to describe an approach that comprises solutions for modeling, gathering, using and evolving context in a BPM oriented environment.

Comparação de Funções de Avaliação em Revisão de Teorias Probabilísticas de Primeira-Ordem

Recentemente tem sido grande o interesse em integrar raciocı́nio probabil ı́stico com representaç̃ oes ĺ ogicas de primeira-ordem. Os modelos propostos na literatura aprendem considerando como espaço de busca modi£caç ões em toda a teoria. Em um trabalho anterior argumentamos que quando a teoria é aproximadamente correta a utilizaç̃ao de t́ecnicas de revis̃ ao para apenas alterar a teoria nos pontos em que ela falha na cobertura dos exemplos seria mais vantajoso. Para avaliar estas modi£caç̃oes e escolher a melhor foi utilizado log verossimilhança. Entretanto, em tarefas de classi£caç̃ ao Bayesiana proposicionais foi mostrado que esta ñao se mostra adequada e por esse motivo log verossimilhança condicional deve ser utilizada. Neste artigo, comparamos os resultados experimentais da utilizaç ̃ o de quatro funções de avaliaç̃ ao probabilı́sticas, incluindo log verossimilhança condicional, ao revisar uma teoria probabiĺ ıstica de primeira-ordem.

Further results of probabilistic first-order revision of theories from examples

Recently, there has been great interest in integrating first-order logic based formalisms with mechanisms for probabilistic reasoning, thus defining probabilistic first-order theories (PFOT).Several algorithms for learning PFOTs have been proposed in the literature. They all learn the model from scratch. Consider a PFOT approximately correct, i.e., such that only a few points of its structure prevent it from reflecting the database correctly. It is much more efficient to identify these points and then propose modifications only to them than to use an algorithm that learns the theory from scratch. Therefore, in [3] we proposed a Bayesian Logic Programs Revision system (RBLP), which receives an initial BLP and through the examples discovers points that fail in covering some of them, similarly to FORTE [4] for the logical approach. These points are called logical revision points. RBLP then considers modifications only for those points choosing the best one through a scoring function. It is required that the implemented modification improves examples covering. It is expected that the returned BLP is consistent with the database.When learning or revising probabilistic first-order theories negative examples are incorporated into the set of positive examples, since the distributions of probabilities will reflect this difference in accordance with the domain of the predicates. At first, this would suggest only using generalization operators. The probabilistic learning algorithms also considers specialization operators where specialization is guided by the scoring function. The question arises of whether using specialization operators when revising a PFOT can improve classification and the result of the scoring function.In [2], besides experimentally comparing scoring functions, we extended RBLP, presenting PFORTE, arguing for the use of specialization operators even when there are no negative examples. We defined then probabilistic revision points, which are the places in theory that result in inaccurate classification of examples (the example was proved, but the value infered for the class is not the one given in the example). Modifications in these points are proposed by specialization operators and the best one is chosen through a scoring function. It is required that these modifications improve the score while not allowing any example to be unproved.Although the ideas presented here can be used for most kinds of PFOTs, we used BLP [1] to implement our system and experimentally compare the results.In the present work, we further study the benefits of considering specialization operators. We compare PFORTE, RBLP, and RBLP modified to allow specialization when a rule is being created by the add rule operator, using four datasets and considering conditional log likelihood as scoring function (since it obtained in [2] the best results). The resultant probabilistic accuracy for PFORTE was the best one considering p < 0.01 as significance level.In another experiment, using the family domain, we provide to PFORTE an approximately correct PFOT (85% of covering and 78% of classification) and an empty PFOT. The running time average for theory revision was 2,92 times faster than for learning from scratch. The experiment suggests that it may be more efficient revising than learning from scratch when the PFOT is approximately correct.