Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Concept Extractor - automatic extraction of product concepts from different natural language product artefacts

: Darting, Simon
: Rombach, Dieter; Maier, Andreas

Fulltext urn:nbn:de:0011-n-5250291 (1.0 MByte PDF)
MD5 Fingerprint: 415faa4eb83271312f215c3874a54ffc
Created on: 19.12.2018

Kaiserslautern, 2018, IX, 68 pp.
Kaiserslautern, TU, Master Thesis, 2018
Master Thesis, Electronic Publication
Fraunhofer IESE ()
text mining; systems engineering; product concept; systematic mapping study; case study; tool-chain

The daily work of a systems engineer involves the challenge of dealing with a large number of natural language documents. This data is the source of the relevant information for later systems engineering (SE) activities and therefore must be elicited. Since this manual work is not only time-consuming, frequently recurring, and very error-prone, it is worth considering a suitable software support. A research gap was identified by conducting a systematic mapping study with an outcome of 27 relevant papers. Based on these results and a manual target search, we describe the process of deriving and creating a prototype by following the CRoss-Industry Standard Process for Data Mining (CRISP-DM). The prototype consists of three different text mining approaches, namely TF-IDF, Word2Vec, Doc2Vec, represented in five tool-chains. These text mining approaches extract information in form of topic clusters, so-called product concepts, from a set of product-related documents such as requirements specifications, use cases or tender documents. In order to investigate the support of the results of the respective tool-chains for SE experts' work, a case study was conducted. The results of the case-study show potential for improvement and further development. In conclusion it must be stated, that although not all relevant information was revealed, the tool-chains identified information that has not been known by the experts.