Dynamic resource allocation is considered a key aspect within business process management. Selecting the most suitable resources is a challenge for those in charge of making the allocation, because the e ciency with which this task is executed, can contribute to the quality of the results, and improve the process performance. Di erent mechanisms have been proposed to improve resource allocation. However, there is a need for more exible allocation methods that integrate a set of conditions and requirements de ned at run-time, and also, allow the combination of di erent criteria to evaluate resources. In this paper, we present ResRec, a novel Multi-factor Criteria tool that can be used to recommend and allocate resources dynamically. The tool provides the feature of solving individual requests (On-demand), or requests made in blocks (Batch) through a recommender system developed in ProM.
whereas, a Nave Bayes Model approach has been proposed to select the best
performer to execute an activity based on the total completion time [
The proposed framework for allocating/recommending the most suitable resources for executing activities in a business process is shown in Figure 1. For allocation we mean selecting the most appropriate resource. For recommendation we mean retrieving a ranking of the most suitable resources that could be considered by the person in charge of the allocation. The user responsible for undertaking the allocation speci es the characterization of a resource allocation/recommendation. This characterization is integrated by di erent factors that represent the desired request properties, e.g., what part of the process is the resource request for or what speci c characteristic the process instance has. For example, in a Help-Desk process, Contact level 1 English are two factors that can characterize a request. Resources are evaluated considering several criteria: frequency, performance, quality, cost, expertise and workload. Furthermore, speci c weights are de ned according to the importance level that the responsible wants to give to each criterion.
We use an information repository to store the resource contextual data (e.g.,
expertise), and the historical information about past process execution (e.g.,
frequency, performance, quality, and cost). We created a knowledge base, where
we use a Resource Process Cube to abstract at a conceptual level (not at an
implementation level) the historical information, and expertise matrices to
represent the expertise of a resource and the desired level of expertise required
to execute a given characterization. Each available resource to execute a given
characterization is evaluated considering di erent criteria, which are computed
across several metrics de ned in [
WEIGHTS
Resource Resource
Allocation Recommendation
four use cases: 1) On-demand Resource Allocation: a resource is allocated to a
single request; 2) On-demand Resource Recommendation: a ranked list of
resources is recommended for a single request; 3) Batch Resource Allocation: a
batch of requests is evaluated, and a resource is allocated to each of them; and
4) Batch Resource Recommendation: a batch of requests is considered, and a
ranked list of resource tuples is recommended to perform all requests. We use
a recommender system that uses two methods to allocate/recommend the most
suitable resources: a method based on Integer Linear Programming (to support
On-demand or Batch Resource Allocation), and a method based on the Best
Position Algoritm (BPA2) [
To perform a dynamic resource allocation/recommendation, the ResRec tool provides two plugins implemented in ProM: 1) GenerateResourceKnowledge, and 2) Recommend Resources. Both plugins are available in the Resource Recommendation package in the open-source framework ProM (available in ProM nightly-builds 2). 3.1
To use the ResRec tool, we have rst to import the information (contextual and historical) needed for the decision making within the framework (Figure 2a). For this purpose, we created a standardized format to store the information that is 2 www.promtools.org/prom6/nightly considered. ResRec includes an extension to the Java OpenXES library 3 that allows to standardize and manipulate the information that is used to generate the required knowledge. Using the GenerateResourceKnowledge plugin (Figure 2b), the di erent criteria are evaluated through di erent metrics, generating the resource knowledge. 3.2
Based on the resource knowledge, we apply the RecommendResources plugin (Figure 2c). This plugin allows the con guration of the required parameters to obtain the nal allocation/recommendation. Figure 2d shows the needed con guration. First, the person in charge of making the allocation must decide which method wants to perform: resource allocation (single resource), or resource recommendation (a ranking of resources). Then, the person in charge speci es the amount of requests for each type of available characterizations (On-demand for 1 request, Batch for 2 or more requests). After that, weights are de ned to describe the importance of each criterion considered in the framework. Finally, the recommender system computes the allocation/recommendation (Figure 2e) according with the chosen con guration, showing as an output the resource(s) allocated/recommended for each request.
Note that the ResRec tool is designed to be generic and extensible, being able to incorporate new criteria to evaluate the resources, so as to make the resource allocation a more e ective task. Moreover, the framework can be adapted to be used in any domain or scenario.
A screencast that demonstrates the usage of the ResRec tool is available on the web (http://is.ing.uc.cl/dcc/index.php/resrec/). The screencast shows the steps followed to generate the resource allocation/recommendation considering the use cases introduced in Section 2. We used a Help-Desk process as a running example. First, we add the contextual and the historical information required to evaluate the resources. Then, we create the knowledge base needed to perform the resource recommendation. After that, we con gure the corresponding parameters in order to generate a Batch Resource Recommendation, de ning a batch of requests and the weights describing the importance of each criterion. We show the obtained results by applying the method based on BPA2, performing a Batch Resource Recommendation. Additional to the Batch Resource Recommendation, the other three use cases outlined are also presented. Furthermore, we used the ResRec tool to recommend resources in a real life scenario. The case study included the resource allocation/recommendation for a consulting rm specializing in the sale of business software solutions in Costa Rica, based on its help-desk process. 5
We presented ResRec, a multi-criteria tool that consider historical and contextual information to allocate/recommend the most suitable resources to execute either a single request or a batch of requests, de ned at run-time. To obtain the nal recommendation, we propose a recommender system that use two methods: 1) Resource allocation based on Integer Linear Programming, and 2) Resource recommendation based on the Best Position Algorithm (BPA2). This tool also has been tested through a real-life help desk scenario of a software consulting company that handles a management solution developed for the automotive industry named DMS-One SAP system. We considered a event log with 1.778 cases registered between August and November 2015. As a future work, we may consider integrating a greedy-based approach as an alternative method to produce a Batch Resource Recommendation with several requests that need to be resolved simultaneously. We also plan to evaluate the incorporation of new criteria to enrich the knowledge base available to determine the most suitable resources. Finally, we aim to apply the framework to allocate/recommend resources with more case studies.
Acknowledgments. This project was partially supported by the Ph.D. Scholarship Program of CONICYT Chile (CONICYT-Doctorado Nacional 2014 63140181), Universidad de Costa Rica Professor Fellowships, and by Fondecyt (Chile) Project No.1150365.
Arias et al.