Searching Repositories of Web Application Models

Abstract

Abstract. Project repositories are a central asset in software development, as they preserve the technical knowledge gathered in past development activities. However, locating relevant information in a vast project repository is problematic, because it requires manually tagging projects with accurate metadata, an activity which is time consuming and prone to errors and omissions. This paper investigates the use of classical Information Retrieval techniques for easing the discovery of useful information from past projects. Differently from approaches based on textual search over the source code of applications or on querying structured metadata, we propose to index and search the models of applications, which are available in companies applying Model-Driven Engineering practices. We contrast alternative index structures and result presentations, and evaluate a prototype implementation on real-world experimental data.

Graph-Based Search over Web Application Model Repositories

“…Moogle is a model search engine that uses UML or Domain Specific Language (DSL) metamodels to create indexes for evaluation of complex queries [8]. Our previous work [1] performs model search using textual information retrieval methods and tools, including model segmentation, analysis and indexing; the query language adopted is purely keyword-based. Nowick et al [13] introduce a model search engine that applies a user-centric and dynamic classification scheme to cluster user search terms.…”

“…This is realized by performing similarity search using graph matching based on the calculation of graph edit distance. The contribution of this paper includes: 1) a graph based approach for content-based search in model repositories; 2) a flexible indexing strategy adaptive to the relationships among model elements found in the DSL metamodel; 3) a similarity measure that exploits semantic relationships between model element types; and 4) implementation and evaluation of the proposed framework by using projects and queries encoded in a Web Domain Specific Language called WebML (Web Modeling Language) 1 . Although the evaluation is performed on a repository of WebML projects, the approach is general and can be applied to any DSL described through a metamodel.…”

Textual and Content-Based Search in Repositories of Web Application Models

“…Some examples of existing on-line tools for sharing and retrieving source code are Google code, Snipplr, Koders, and Codase 9 . As explained in [Bozzon et al 2010],the most basic solution is the case where queries in form of keyword(s) are simply matched to the code and the results are the exact locations where the keyword(s) appear in the matched code snippets. However, online tools allow advanced search by using reg-ular expressions (Google Codesearch), wildcards (Codase); supporting search of specific syntactical categories, like class names, method invocations, variable declarations (Jexamples and Codase); making the search more specific by indicating fixed set of metadata (e.g., programming language, license type, file and package names).…”

“…However, online tools allow advanced search by using reg-ular expressions (Google Codesearch), wildcards (Codase); supporting search of specific syntactical categories, like class names, method invocations, variable declarations (Jexamples and Codase); making the search more specific by indicating fixed set of metadata (e.g., programming language, license type, file and package names). Source code online tools also have to consider a way to compute a relevance score between the query and the matched source code, and present the corresponding results to the user [Bozzon et al 2010]. Regarding this aspect, some approaches retrieve a list of matches without providing ranking, while others implement IR-style ranking using the standard TF/IDF measure, or ranking which besides the matches with the source code takes into account the project properties such as recency of the project, number of downloads, activity rates etc.…”

“…The results described in this paper are rooted in our previous work on model search. The early work [Bozzon et al 2010] defined the problem of searching over DSL repositories and proposed a technical architecture for the case of keyword-based model retrieval; no evaluation with a gold standard for the keyword-based retrieval system was reported yet. The subsequent works [Bislimovska et al 2011b;Bislimovska et al 2011a] introduced the approach of content-based search, supported with the A-star algorithm, and reported a preliminary evaluation of results with a limited ground truth dataset compiled by the authors.…”