=Paper=
{{Paper
|id=None
|storemode=property
|title=Curating Semantic Linked Open Datasets for Software Engineering
|pdfUrl=https://ceur-ws.org/Vol-1035/iswc2013_demo_16.pdf
|volume=Vol-1035
|dblpUrl=https://dblp.org/rec/conf/semweb/MaheshNBP13
}}
==Curating Semantic Linked Open Datasets for Software Engineering==
https://ceur-ws.org/Vol-1035/iswc2013_demo_16.pdf
Curating Semantic Linked Open Datasets for
Software Engineering
Kavi Mahesh, Aparna Nagarajan,
Apoorva Rao Balevalachilu, and Karthik Prasad
Centre for Knowledge Analytics and Ontological Engineering - KAnOE,
PES Institute of Technology, Bangalore 560085 India
{drkavimahesh,aparna.nagarajan26,
apoorva.rao.b,karthikprasad008}@gmail.com
http://kanoe.org/
Abstract. A typical software engineer spends a significant amount of
time and effort reading technical manuals to find answers to questions es-
pecially those related to features, versions, compatibilities and dependen-
cies of software and hardware components, languages, standards, mod-
ules, libraries and products. It is currently not possible to provide a
semantic solution to their problem primarily due to the non-availability
of comprehensive semantic datasets in the domains of information tech-
nology. In this work, we have extracted, integrated and curated a linked
open dataset (LOD) called LOaD-IT exclusively on this domain from
a variety of sources including other LODs such as Freebase and DBPe-
dia, technical documentation such as JavaDocs and others. Further, we
have built a technical helpdesk system using a semantic query engine
that derives answers from LOaD-IT. Our system demonstrates how pro-
ductivity of the software engineer can be improved by eliminating the
need to read through lengthy technical manuals. We expect LOaD-IT
to become more comprehensive in the future and to find other related
practical applications.
Keywords: Linked Open Data, software engineering, technical helpdesk,
information retrieval, semantic search.
1 Introduction
Software developers, test engineers, researchers and students face technical queries
in the course of their daily work. The solution to their problems, more often than
not, is to go through some technical documentation, search for information, look
for a similar question on on-line forums, or to ask somebody who is considered an
expert in that domain. Unfortunately, information from these sources is rather
unstructured, distributed and difficult to obtain. The process of going through
the technical documentation is tedious and time consuming. What the person
ideally wants is a quick and precise answer to the question perhaps augmented
by a suitable visualization to aid comprehension. The objective of our work is to
enhance workplace productivity of information technology workers by reducing
the effort and time spent in searching for technical information.
There is currently no readily available LOD that covers the factual knowl-
edge sought by a typical information technology (IT) worker. Thus, our solution
begins by semanticizing the relevant documentation by creating a linked open
data store, which we call LOaD-IT (available as an RDF dump in N-Quads for-
mat at http://datahub.io/dataset/load-it). The potential of this highly special-
ized linked data is tapped by building Kappa, a technical assistant application
prototype (Kappa is hosted at http://kanoe.org/lod/loadit-kappa.html). Kappa
takes keyword queries as input and provides specific factual answer(s) instead
of large number of HTML pages that may or may not contain the correct and
complete answer. It also saves the user from scrolling through long manuals and
documents to manually extract the answer. In order to provide results in the
form of facts, the application first performs query interpretation as described in
[1]. The user of the application need not have any knowledge of the structure of
LOaD-IT or knowledge of RDF formats. Simple keywords entered in a search bar
suffice. Search results are presented as a list of facts along with hyperlinks to rel-
evant pages. Further, a graphical representation of the relevant RDF sub-graph
is shown to help the user visualize the facts. Use of anontology to implement se-
mantic search together with Web search has been demonstrated before (see, e.g.,
[2]). However, such methods typically require semantic meta-data or annotations
to be manually added to existing documents or web pages.
2 Creating the Semantic LOD on Information
Technology: LOaD-IT
Existing semantic datasets [3] like DBPedia, Freebase and YAGO are quite large
and cover multiple domains while focusing heavily on common areas of interest
rather than technical domains such as IT. It is unlikely that a retrieval engine on
such a general-interest LOD will provide the necessary recall or precision for IT
users. Our aim is to curate an LOD specific to the IT and software engineering
domains which will also be of smaller size than the above LODs.
Another concurrent objective is to generate RDF quads from technical man-
uals, user guides and other documentation that are widely available but demand
considerable amount of time and effort to sift through. Semanticization of this
kind of documentation will contribute to raising the quality of answers (i.e., re-
call and precision) that the dataset is capable of providing at a fine grain of de-
tail. Extraction of RDF facts from free flowing paragraphs of explanatory texts
amounts to solving core problems of natural language processing and seman-
tic information retrieval. Therefore, we take the practical approach of selecting
‘low-hanging fruit? by focusing on relatively well-structured technical documen-
tation. Java Platform Standard Edition 7 Documentation API was selected for
this purpose because of its definite structure in HTML tables and consistent lay-
out, thereby facilitating extraction of clean and accurate facts. The widespread
use of Java in industry and academia is another factor why this documentation
API is a good choice for our purposes.
The methodology adopted for constructing the LOD has two parts:
2.1 Filtering Existing Datasets
A set of carefully chosen seed words from the domain of IT and software en-
gineering is used to select matching resources from Freebase and DBPedia. For
each of these resources, all triples matching the resource in their subject fields
are selected. All other triples are filtered out as irrelevant to our domain of inter-
est. These operations were carried out using Unix Shell scripts. The numbers of
triples obtained by this method from Freebase and DBPedia are shown in Table
1 below.
2.2 Extracting triples from technical documentation
1. A crawler is built to cover the entire documentation collection.
2. A screen scraping package is designed and built to parse the pages and
extract facts after a thorough study of the layout of each type of page.
3. Rules are written for logical mapping into quads:
(a) Check if any reuse of predicates is possible from existing vocabularies
like DC, FOAF, SKOS, and SIOC. Reuse wherever possible.
(b) If compatible predicates are not found, create new predicates that can
appropriately describe the relationships. For instance, to link a Java
method to its return type or modifier, a new predicate is required.
(c) Create and publish the required explanatory pages for predicates in RDF
and HTML formats. URLs are used to generate quads: (Subject, Pred-
icate, Object, Context) according to the conventions of Linked Open
Data.
4. A suitable software library is used to load the quads into a database or data
store.
The documentation of Java 7 SE is obtained from the official web site and parsed
using a Python Library called BeautifulSoup4 [4]. The LOaD-IT dataset thus
obtained has 1.4 million quads. Statistics of the dataset are given in Table 1.
Table 1. Numbers and Data Sizes in LOaD-IT
Source OriginalSize F ilteredSize RetainedQuads
Freebase 49 GB 128 MB 740149
DBPedia 44 GB 184 MB 433834
Javadocs < 1 GB 65 MB 268275
TOTAL LOaD-IT 377 MB 1442258
3 Kappa: The LOaD-IT Retrieval Engine
Kappa first constructs the top k query candidates for the given keywords us-
ing the methodology outlined in [1]. Once the SPARQL query candidates are
generated and the best query is selected, the results are obtained by sending an
HTTP request to a REST API provided by our wrapper for LOaD-IT. This API
access to LOaD-IT can be used also by other applications in the future. Results
obtained as JSON objects are rendered both as triples and graphically to help
the user in visualizing the relevant RDF sub-graph of LOaD-IT. An easy-to-use
browser-based interface is provided for this purpose. In addition, the query is
also sent to external Web search engines (e.g., Bing) as well as on-line technical
discussion forums (e.g., StackOverflow). Results from external engines are also
displayed to the user to help them look for external content related to the query.
Fig. 1(a) shows the overall architecture of LOaD-IT and Kappa.
4 Example Scenarios
Two example scenarios and results are presented below to illustrate the usage
of Kappa and LOaD-IT.
Scenario 1 The user wants to know what to write in the import statement in
a Java program if he has a need to use a method from the Driver Class in Java.
In fact, the user may not even know whether Driver is the name of a Class or a
Package.
Keywords entered: Package Driver
Facts retrieved:
java.sql type package
java.sql member Driver
Driver type class
A graphical rendering of these three triples in shown in Fig. 1(b)
Scenario 2 The user wants to know the details of the Integer Class in Java.
Keywords entered: Integer
Facts retrieved:
Integer comment Integer class wraps a value of the primitive type int in an object
Integer member highestOneBit,hashCode,valueOf,getInteger..
Integer type class
(a) Schematic diagram of the architecture of (b) Visualization of
Kappa Facts Returned by
Kappa
Fig. 1.
5 Conclusion and Future Work
Our primary aim was to create a linked open dataset for the IT domain. After
filtering existing datasets, we created a crawler and used screen scraping to ex-
tract RDF quads from semi-structured technical documentation. We integrated
and normalized the dataset to ensure its compatibility with other LODs and
our application Kappa. A reasonably large technical dataset was obtained and
its potential was illustrated by a simple yet meaningful and useful application,
a fact-finding engine that serves as a technical helpdesk for practicing software
engineers.
The dataset can be further enlarged and enriched to include RDF quads
extracted from other technical documentation and on-line discussion forums.
Further, it can be refined and grown over time by adopting a crowd-sourcing
approach. Additionally, a vocabulary can be created specifically for predicates
that are useful in a software engineering context and can be standardized to bring
in global acceptance. The facts that are rendered can be enriched by natural
language processing techniques to bring them closer to proper English.
We expect LOaD-IT to find several other useful applications in the near
future. An immediate possibility is to use it as a controlled vocabulary plus on-
tology for corporate knowledge management. It can also be used to further auto-
mate the generation and management of program comments and documentation
in software engineering by integrating it with IDE for various programming lan-
guages and platforms. None of these is readily possible without LOaD-IT wherein
a federated retrieval engine would have to query various general-purpose LODs
as well as document repositories and the Web. LOaD-IT together with Kappa
promises to deliver the levels of precision and recall to make them useful in
real-world software engineering situations.
A demonstration of our application Kappa is available at http://kanoe.
org/lod/loadit-kappa.html.
References
1. Tran, T., Wang, H., Rudolph, S. et al.: Top-k Exploration of Query Candidates for
Efficient Keyword Search on Graph-Shaped (Rdf) Data. (2009) 405-416
2. Karanth, P., Mahesh, K.: Integrating Knowledge Base Retrieval with Web Search
using Semantic Roles In: Lecture Notes in Engineering and Computer Science: Pro-
ceedings of the International MultiConference of Engineers and Computer Scientists
2012 (IMECS 2012, ICCS 2012), Vol. 1, ISBN 978-988-19251-1-4 pp. 344-349 IAENG
Hong Kong (2012)
3. Bizer, C., Jentzsch, A., Cyganiak, R.: State of the LOD Cloud. Version 0.3 (Septem-
ber 2011), (2011)
4. Richardson, L.: Beautiful Soup-HTML. XML parser for Python, (2008)