Mobile devices and networks produce abundant data that exhibit geo-spatial and temporal properties mainly driven by human behavior and activities. We refer to such data as geoMobile data. Mining such data to extract meaningful patterns that are reflective of collective user activities and behavior can benefit mobile network resource management as well as the design and operations of mobile applications and services. However, diverse feature distributions inherent in such data make such a task challenging. In this paper we advocate an approach based on advanced machine learning algorithms to transform original data matrices into a feature distributional similarity graph and extract “latent” patterns from complex structures of geoMobile data. Our analysis is further aided with a visualization technique. Using mobile access data records from an operational cellular carrier, we demonstrate the potentials of our proposed approach under multiple settings, and make some very interesting observations from the obtained results.
|Original language||English (US)|
|Title of host publication||MobiData 2016 - Proceedings of the 1st Workshop on Mobile Data, co-located with MobiSys 2016|
|Publisher||Association for Computing Machinery, Inc|
|Number of pages||6|
|State||Published - Jun 30 2016|
|Event||1st Workshop on Mobile Data, MobiData 2016, co-located with MobiSys 2016 - Singapore, Singapore|
Duration: Jun 30 2016 → …
|Name||MobiData 2016 - Proceedings of the 1st Workshop on Mobile Data, co-located with MobiSys 2016|
|Other||1st Workshop on Mobile Data, MobiData 2016, co-located with MobiSys 2016|
|Period||6/30/16 → …|
Bibliographical noteFunding Information:
This research was supported in part by NSF grants CNS-1117536, CRI-1305237, CNS-1411636 and DTRA grant HDTRA1-14-1-0040 and DoD ARO MURI Award W911NF-12-1-0385.
© 2016 ACM.
- Call detail records
- Feature distributions
- Latent pattern extraction
- Similarity graph