[SystemSafety] HC2AS - Human Contribution to Aviation Safety
Derek M Jones
derek at knosof.co.uk
Wed Jun 6 19:27:59 CEST 2018
Brian,
> Textual analysis of narrative incident reports can produce word counts of phrases that are indicative of, say, flight-crew problem solving.
I would not hold out too much hope of getting any useful
from using text analytic. Tools are getting better,
https://spacy.io/ is very good, but still needs lots of
manual direction and the resulting special purpose tools
tend to be fragile.
Even doing something as simple as extracting numeric values from
text can be difficult:
http://shape-of-code.coding-guidelines.com/2017/01/22/full-fact-checking-of-number-words/
Sentiment analysis is a trendy text analysis topic.
A commonly used model was trained on film reviews.
Domain specific models can produce results that are a lot
more realistic:
https://www.researchgate.net/publication/308729996_Choosing_your_weapons_On_sentiment_analysis_tools_for_software_engineering_research
Various people have used text analytics of fault reports,
with somewhat disappointing results.
--
Derek M. Jones Software analysis
tel: +44 (0)1252 520667 blog:shape-of-code.coding-guidelines.com
More information about the systemsafety
mailing list