SOCNET - A Study of Student Societies for Mobile Social Networks
WLAN deployment across university campuses has risen rapidly in recent years. There has been a marked increase in mobile users and traffic as a result. Analyzing usage of WLANs is currently a major research issue. The SOCNET study is based on analyzing WLAN traces in campus-wide WLANs and discovering significant usage patterns according to different group classifications. Recent work in this direction has been based on classification by gender of wireless users and sub classification by area of interest based on AP locations. Further groupings are envisioned for investigation in the near future. There is a strong dependency between AP locations and our study, making it necessary to know this information. The procedure used in the SOCNET study is summarized in the diagram below.
Subsections below explain the studies carried out as part of SOCNET (further studies are ongoing and will be put online as and when they become available).
(1) Gender Based Feature Extraction in a Campus-wide WLAN* - Nikhil Yadav, Udayan Kumar. Research Supervisor: Prof. Ahmed Helmy
In order to carry out a meaningful analysis of the various groups of WLAN users, specifically targeting and differentiating between genders, the traces need to provide the following information:
- comprehensive syslog (or SNMP) logs for various buildings, dorms and sororities/fraternities for the general population of students (without population bias) and for extended periods of time (30 days or more)
- mapping between the APs (or point of collection) to specific building’s designation.
- differentiation between individual users and ability to track users (without necessarily knowing their identity).
We have investigated WLAN traces collected at USC, UNC and Dartmouth. Dartmouth traces do not provide AP-to-building mapping, which makes it difficult to do this kind of study. UNC traces on the other hand have limited number of APs in sororities and fraternities. We chose a one month long USC (http://nile.cise.ufl.edu/mobilib) trace for our study. It provides 12 fraternities and 7 sororities included in the WLAN traces and the AP-to-building mapping is also available. The stages and results in our study are summarized below
a) Parsing: Scripts were written to process the data from the USC WLAN traces and make them conform to a standard format. The parsers with descriptions can be downloaded here.
b) Thresholding: As fraternities and sororities have visitors, our classification needed further refinement. Since visitors are infrequent users, their associated number of sessions is, in general, less than that of regular users. These visitors are excluded for improved accuracy, using the following steps: i. Extract the number of sessions per MAC for each fraternity or sorority AP. ii. Vary the min. session duration (as a threshold for regular users) and observe its effect on the number of sessions and distinct users. iii. Obtain a suitable threshold for the session duration and session count to classify users above these limits as being either males or females2. After performing the above procedure for the studied trace, we plot the session count curves for sorority and fraternity users in Figure 1 in descending order with respect to the number of sessions. An interesting feature of the figure is the presence of a sharp bend (knee) as the number of sessions per MAC decreases. MACs below the knee have an order of magnitude less number of sessions (intuitively accounting for the difference between a regular user and a visitor). All users below the knee were classified as visitors and removed from the study. Changing the session duration had no effect on the shapes of the curves. The session duration threshold here was the average session duration of sororities and fraternities for all the users. The analysis was carried for different time durations to check for emergence of the pattern. The results from this study can be found here
c) DB queries: DB queries were written in SQL on the parsed, thresholded data to come up with the graphs shown in the results (below) for further analysis.
d) Location to AP : The USC campus's complete map can be found on http://www.usc.edu/assets/maps/upc_map.pdf . AP to location mapping files are also available for this (refer to http://nile.cise.ufl.edu/mobilib ).
RESULTS:
The graphs below show the results obtained on doing the analysis:
The use of SQL queries on the WLAN traces are generic. The results from this research are based on a sample of the user population, since gender may be identified based on sorority and fraternity AP associations. The concept of an ‘area’ is used, which includes majors as well as different buildings on campus to improve the richness of data set. To conclude, there is a distinct difference in WLAN usage patterns for different genders even with similar population sizes. Females seem to dominate in WLAN usage in areas of Social Science and Law and prefer Apple over Intel. Males have longer session durations than females in most cases. We hope for this study to open the door for other mobile social networking studies and profile-based service designs.
* Poster selected at Mobicom 2007, entered in 1st round SRC, to be published in MC^2R. Related poster selected in CRAWDAD poster session 2007.