Célia Ghedini Ralha

A multi-agent data mining system for cartel detection in Brazilian government procurement

The main focus of this research project is the problem of extracting useful information from the Brazilian federal procurement process databases used by government auditors in the process of corruption detection and prevention to identify cartel formation among applicants. Extracting useful information to enhance cartel detection is a complex problem from many perspectives due to the large volume of data used to correlate information and the dynamic and diversified strategies companies use to hide their fraudulent operations. To attack the problem of data volume, we have used two data mining model functions, clustering and association rules, and a multi-agent approach to address the dynamic strategies of companies that are involved in cartel formation. To integrate both solutions, we have developed AGMI, an agent-mining tool that was validated using real data from the Brazilian Office of the Comptroller General, an institution of government auditing, where several measures are currently used to prevent and fight corruption. Our approach resulted in explicit knowledge discovery because AGMI presented many association rules that provided a 90% correct identification of cartel formation, according to expert assessment. According to auditing specialists, the extracted knowledge could help in the detection, prevention and monitoring of cartels that act in public procurement processes.

Improving Business Process Models with Agent-Based Simulation and Process Mining

Business processes are usually modeled at a high level of abstraction, while the analysis of their run-time behavior through process mining techniques is based on low-level events recorded in an event log. In this scenario, it is difficult to discover the relationship between the process model and the run-time behavior, and to check whether the model is actually a good representation for that behavior. In this work, we introduce an approach that is able to capture such relationship in a hierarchical model. In addition, through a combination of process mining and agent-based simulation, the approach supports the improvement of the process model so that it becomes a better representation for the behavior of agents in the process. For this purpose, the model is evaluated based on a set of metrics. We illustrate the approach in an application scenario involving a purchase process.

Mining the low-level behaviour of agents in high-level business processes

Currently there is a gap between the high level of abstraction at which business processes are modelled and the low level nature of the events that are recorded during process execution. When applying process mining techniques, it is possible to discover the logic behind low-level events but it is difficult to determine the relationship between those low-level events and the high-level activities in a given process model. In this work, we introduce a hierarchical Markov model to capture both the high-level behaviour of activities and the low-level behaviour of events. We also develop an expectation-maximisation technique to discover that kind of hierarchical model from a given event log and a high-level description of the business process. We use this technique to understand the behaviour of agents in business processes, from the control-flow perspective and from the organisational perspective as well. Using an agent-based simulation platform (AOR), we implemented a purchasing process and generated an event log in order to illustrate the benefits of the proposed approach and to compare the results with existing process mining techniques, namely the ones that are available in the ProM framework.

A Cooperative Multi-agent Approach to Computer Forensics

This article proposes the use of a collaborative multi-agent approach to develop a toolkit to assist the experts during the forensic examination process: MADIK - a Multi-Agent Digital Investigation ToolKit. The use of a multi-agent approach has been proved adequate, specially regarding the cooperative action of the autonomous specialized agents: HashSetAgent, FilePathAgent, TimelineAgent, FileSignatureAgent. Also the distributed nature of the multi-agent approach allows for better usage of computational resources, since agents can operate autonomously in different machines and environments. As part of our work, we have defined a four layer multi-agent architecture, as a metaphor to the organizational hierarchy levels, which is divided in strategic, tactical, perational and specialist levels. The proposed architecture was the base to the development of the toolkit, which was developed with a blackboard approach, implemented over the Java Agent DEvelopment Framework - JADE, using Java Expert System Shell - JESS. We have done some experiments with MADIK using real data and the results are encouraging. This paper focuses on the benefits of using the multi-agent approach to aid in the forensic examination process, specially regarding the cooperative action of the autonomous specialized agents, which we deem as a flexible and promising possibility that should be further explored in the computer forensics scenario.

MASE-BDI: agent-based simulator for environmental land change with efficient and parallel auto-tuning

This paper presents an agent-based simulator for environmental land change that includes efficient and parallel auto-tuning. This simulator extends the Multi-Agent System for Environmental simulation (MASE) by introducing rationality to agents using a mentalistic approach—the Belief-Desire-Intention (BDI) model—and is thus named MASE-BDI. Because the manual tuning of simulation parameters is an error-prone, labour and computing intensive task, an auto-tuning approach with efficient multi-objective optimization algorithms is also introduced. Further, parallelization techniques are employed to speed up the auto-tuning process by deploying it in parallel systems. The MASE-BDI is compared to the MASE using the Brazilian Cerrado biome case. The MASE-BDI reduces the simulation execution times by at least 82 × and slightly improves the simulation quality. The auto-tuning algorithms, by evaluating less than 0.00115 % of a search space with 6 million parameter combinations, are able to quickly tune the simulation model, regardless of the objective used. Moreover, the experimental results show that executing the tuning in parallel leads to speedups of approximately 11 × compared to sequential execution in a hardware setting with 16-CPU cores.

Towards a Cognitive Meta-Model for Adaptive Trust and Reputation in Open Multi-Agent Systems

Computational trust and reputation models are key elements in the design of open multi-agent systems. They offer means of evaluating and reducing risks of cooperation in the presence of uncertainty. However, the models proposed in the literature do not consider the costs they introduce and how they are affected by environmental aspects. In this paper, a cognitive meta-model for adaptive trust and reputation in open multi-agent systems is presented. It acts as a complement to a non-adaptive model by allowing the agent to reason about it and react to changes in the environment. We demonstrate how the meta-model can be applied to existent models proposed in the literature, by adjusting the model’s parameters. Finally, we propose evaluation criteria to drive meta-level reasoning considering the costs involved when employing trust and reputation models in dynamic environments.

Improving process models by mining mappings of low-level events to high-level activities

While it is possible to analyze the run-time behavior of a business process through process mining techniques, in practice there is often a gap between the low-level nature of the events recorded in an event log and the high-level of abstraction at which the process is modeled. This makes it difficult to understand the recorded behavior in terms of the high-level activities in the process model. Also, it makes it difficult to improve the model based on run-time data about the process. In this work we present an approach to mine mappings between the events in the log and the activities in the model. These mappings can be used to generate suggestions of how the process model can be extended in order to capture the behavior recorded in the event log. Using a real-world and publicly available event log, we show how the approach can improve the model in a stepwise manner, until it covers all the behavior recorded in the event log.

An agent-based solution for dynamic multi-node wavefront balancing in biological sequence comparison

Abstract Many parallel and distributed strategies were created to reduce the execution time of bioinformatics algorithms. One well-known bioinformatics algorithm is the Smith–Waterman, that may be parallelized using the wavefront method. When the wavefront is distributed across many heterogeneous nodes, it must be balanced to create a synchronous data flow. This is a very challenging problem if the nodes have variable computational power. This paper presents an agent-based solution for parallel biological sequence comparison applications that use the multi-node wavefront method. In our approach, autonomous agents are able to identify unbalanced computations and dynamically rebalance the load among the nodes. Two strategies were developed to the balancer agent in order to identify if the computations are balanced, one using global information and other using only local information. The global strategy demands a huge amount of data transfers, incurring in more communication, whereas the local strategy can decide about the balancing status using only local information. The results show that the balancing gains of strategies are very close. Thus, the local strategy is preferred, since it can be implemented in real wavefront balancers with almost the same benefits as the global strategy.

Towards the Integration of Multiagent Applications and Data Mining

This chapter has the objective to present research on combining two originally separated areas, agents including distributed multiagent systems and data mining, which are increasingly interrelated. Recent research has present that such interaction features are bilateral and complementary, since new approaches and techniques are developed to benefit from the synergetic enhancement of intelligence and infrastructure for information processing and systems. This chapter draws attention to illustrate agent-mining interaction with two different domain multiagent applications: BioAgents at the bioinformatics area and MADIK at the computer forensics area. The presented case studies are driving forces towards the integration of the agent-mining challenging area. As ongoing research works we discuss the prospects of both agent-mining projects.

A Hierarchical Markov Model to Understand the Behaviour of Agents in Business Processes

Process mining techniques are able to discover process models from event logs but there is a gap between the low-level nature of events and the high-level abstraction of business activities. In this work we present a hierarchical Markov model together with mining techniques to discover the relationship between low-level events and a high-level description of the business process. This can be used to understand how agents perform activities at run-time. In a case study experiment using an agent-based simulation platform (AOR), we show how the proposed approach is able to discover the behaviour of agents in each activity of a business process for which a high-level model is known.