An algorithm to mine www.meetup.com to identify best activity groups to attend in a city.
Meetup.com is fantastic way to meet new people and do fun activities. However, identifying the correct meetup groups can be a challenge. Some meetup groups are older and inactive whereas others just do not have enough people attending them.
This R script fetches upcoming meetup events from trending groups in a user-specified location and exports the list as weekly files. Since many meetup groups have low attendees, the script also exports an additional filtered version of the data showing only events with at least X number (user-specified threshhold) attendees. After running this script for a few months, I analysed the data to identify the best meetup groups in a location, ie the meetups with frequent events and decent amount of attendees.
Meetup reserves the rights to the data.
-
This script requires access to meetup's API. You can get your key by registering at https://secure.meetup.com/meetup_api/key/. Store your key in the apikey variable.
-
Location information. The script indirectly passes location information as GET request. For example, the bit corresponding to Boulder, Colorado, USA is stored in the location variable (shown below).
location <- "country=us&city=boulder&state=co"Setting the location information may require trial and error. You can also supply latitude and longitude as 'lat' and 'lon' in the above format. See https://www.meetup.com/meetup_api/docs/2/open_events/ for more information.
If everything is correct, the script should print out a number corresponding to the number of upcoming events.
- Filter and export data. The rest of script is straightforward. It will export two files for each week in /upcoming_events folder: all events and filtered events. I uploaded the exported files for Boulder as an example.
The script was run almost weekly to collect data from November 2017 to August 2018; however, there are some weeks where the data was not collected.
I empirically noticed that Sundays have lowest number of meetups. Here, I plotted the number of attendees/RSVP versus each day of the week to see how it varies in a week.
The RSVP number for meetups on Wednesdays seem to have slightly higher number than the rest.
In addition, I used one-way ANOVA to see if any changes observed are statistically significant, ie P-value < 0.05. Code snippet below.
print("Using One way ANOVA to see if a specific day in a week has more attendees than rest")
fit <- aov(RSVP ~ WeekDay, df)
summary(fit)
stat <- as.data.frame(TukeyHSD(fit)[[1]])
print(stat[stat$`p adj` < 0.05, ]) diff lwr upr p adj
Wednesday-Sunday 4.401224 2.279118 6.523329 4.409442e-08
Wednesday-Saturday 3.759404 1.920911 5.597897 5.936262e-08
Wednesday-Friday 4.357021 2.209332 6.504711 7.185972e-08
Wednesday-Thursday 3.973141 2.061353 5.884928 4.246177e-08
Tuesday-Wednesday -3.500400 -5.327718 -1.673081 3.800073e-07
Monday-Wednesday -3.433965 -5.215838 -1.652093 3.159554e-07Each row represents comparision of the RSVP number versus days. The columns 'diff' and 'p adj' represent the difference between two days in number of RSVPs and P-values corrected for testing multiple hypotheses, respectively.
The test shows that Wednesdays indeed have higher number of attendees than events falling on Thursdays to fridays. However, the magnitude of effect, ie 'diff', ranges from 3-4 more RSVP numbers which is practically insignificant. In addition, the 'diff' is negative for Wednesday versus Monday and Tuesday, but, again, it is negligible.
In summary, all weekdays have equal number of attendees in Boulder for practical purposes.
Meetup groups with highest number of events and attendees (top 25% for both criteria) were deemed as the 'best' in this example, although the definition is rather subjective (Boulder_meetup_analysis.R).
In the figure below, the height of the bar represents median RSVPs for a group, and the number of events from the group is displayed next to the bar. The colour of the graph also responds to number of events.
'Boulder Python' meetup has the highest number of RSVPs but has fewer events whereas 'Boulder Collective Soccer' has around 10 RSVPs for events but a lot of events. The choice of what meetups to attend is personal. In the case of Boulder Collective Soccer, having 10 people may be good enough to form teams for games and so on. However, the script rapidly identifies suitable meetup groups that interests an individual. I attended several events from the 'Boulder 20s and 30s social' meetup (median RSVP: ~25 and 20 events in the time period) and it was fantastic.