| With the rapid development in workflow management technology, enterprises can construct and update their process information management more efficiently. Therefore it would be no strange that an increasing number of workflow models in different area and in different format have been built. While how to manage those models and how to achieve useful information from those models become an important problem that we’d like to give a discuss in this article- process retrieval. While although much work has been down around this question, there doesn’t exist a well-accepted benchmark to evaluate those measures’effect.As for this paper, we first give a brief introduction of the existing workflow retrieval measures, including structure-based, behavior-based and semantic-based measures. We compare and analysis those method and discover some indicated limitations involved.Then after having a profound analysis of the structure and characteristic of workflow model, we construct a benchmark process dataset named JTang Workflow Benchmark according to a series of statement to judge the workflow similarity achieved by questionnaire answered by expert working in workflow areas with different background based on the public process repository provided by IBM. The benchmark consists of retrieval processes and the defined results by the experienced judgments with their rank. Various process retrieval algorithms can be implemented upon the workflow repository and we can calculate and compare their precision.At last, we implement some of those process similarity measure based process retrieval algorithm and analyze the results from different aspect respectively. Besides that, different manners have been involved to compute the precision of those results. In the meantime, the time cost by those algorithms is also mentioned. As the result shows, the cosine-similarity based method achieve the best precision while the greedy algorithm give a more delicate answer when considering the rank of the results. |
PDF Full Text Request