Improving the OpenStreetMap Data Set using Deep Learning
(2018) In LU-CS-EX 2018-05 EDAM05 20181Department of Computer Science
- Abstract
- OpenStreetMap is an open source of geographical data where contributors can change, add, or remove data. Since anyone can contribute, the data set is prone to contain data of varying quality. In this work, we focus on three approaches for correcting Way component name tags in the data set: Correcting misspellings, flagging anomalies, and generating suggestions for missing names. Today, spell correction systems have achieved a high correction accuracy. However, the use of a language context is an important factor to the success of these systems. We present a way for performing spell correction without context through the use of a deep neural network. The structure of the network also makes it possible to adapt it to a different language by... (More)
- OpenStreetMap is an open source of geographical data where contributors can change, add, or remove data. Since anyone can contribute, the data set is prone to contain data of varying quality. In this work, we focus on three approaches for correcting Way component name tags in the data set: Correcting misspellings, flagging anomalies, and generating suggestions for missing names. Today, spell correction systems have achieved a high correction accuracy. However, the use of a language context is an important factor to the success of these systems. We present a way for performing spell correction without context through the use of a deep neural network. The structure of the network also makes it possible to adapt it to a different language by changing the training resources. The implementation achieves an F1 score of 0.86 (ACR 0.69) for Way names in Denmark. (Less)
- Popular Abstract (Swedish)
- Vi demonstrerar i denna artikel en generell tre-stegs-lösning för att förbättra vägnamnsdata i olika länder genom att fylla i saknade namn, rättstava felstavade namn och flagga anomalier.
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8951995
- author
- Londögård, Hampus LU and Lindblad, Hannah
- supervisor
- organization
- alternative title
- Förbättra OpenStreetMaps Vägnamnsdata genom Neurala Nätverk
- course
- EDAM05 20181
- year
- 2018
- type
- H3 - Professional qualifications (4 Years - )
- subject
- keywords
- MSc, Machine Learning, OpenStreetMap, Random Forest, Neural Network, Sequence-2-Sequence
- publication/series
- LU-CS-EX 2018-05
- report number
- LU-CS-EX 2018-05
- ISSN
- 1650-2884
- language
- English
- id
- 8951995
- date added to LUP
- 2018-11-09 13:43:29
- date last changed
- 2018-11-09 13:43:29
@misc{8951995, abstract = {{OpenStreetMap is an open source of geographical data where contributors can change, add, or remove data. Since anyone can contribute, the data set is prone to contain data of varying quality. In this work, we focus on three approaches for correcting Way component name tags in the data set: Correcting misspellings, flagging anomalies, and generating suggestions for missing names. Today, spell correction systems have achieved a high correction accuracy. However, the use of a language context is an important factor to the success of these systems. We present a way for performing spell correction without context through the use of a deep neural network. The structure of the network also makes it possible to adapt it to a different language by changing the training resources. The implementation achieves an F1 score of 0.86 (ACR 0.69) for Way names in Denmark.}}, author = {{Londögård, Hampus and Lindblad, Hannah}}, issn = {{1650-2884}}, language = {{eng}}, note = {{Student Paper}}, series = {{LU-CS-EX 2018-05}}, title = {{Improving the OpenStreetMap Data Set using Deep Learning}}, year = {{2018}}, }