Process model reuse is a foundational aspect of the semantic web vision. The new PSearch framework can search repositories of process flow models for process reuse matches. The authors use multi-dimensional indexes derived from semantic metadata annotations and perform fast process flow analysis. These semantic annotations were the vision for the semantic web and realized through languages like the web ontology language for web services (OWL-S). Now with PSearch, the input/output (I/O) flow activities of processes can be used for search and comparison. This paper is very detailed in describing how indexes are created from process graphs and then used for searching process repositories.

The reusable process graphs are made up of various OWL-S block structures representing the various flow types, such as Sequential, AND, XOR, and LOOP. These graphs are then semantically annotated for use in the PSearch framework. A query is presented to the framework’s filtering module that searches for graphs sharing an activity (which uses an offline index setup). The candidates retrieved are then further filtered by estimating similarities based on user selectable accuracy. The PSearch framework can continue this loop, matching and refining searches across complex graphs.

Over half of the paper discusses the process model representation use for fast retrieval. The authors feel their approach is unique for representing graphs with semantic annotations. Augmented with numerous flow dependency definitions, the paper breaks down the steps necessary to create valuable searchable indexes.

Since the graphs represent things like business management processes, the activities in those flows are key to determining their similarity. Inputs and outputs and flow types are used for further indexing. The result is a repository of graphs that have been “compiled” into their respective process types: (1) processes, (2) activities, and (3) flow dependencies. Process matching and retrieval is also very detailed.

The paper also includes testing and experimental performance results and future ideas for scaling the ideas.