These are chat archives for FreeCodeCamp/DataScience
discussion on how we can use statistical methods to measure and improve the efficacy of http://freeCodeCamp.com
@BerkeleyTrue, @QuincyLarson, @dcsan, @benmcmahon100
@abhisekp, @biancamihai, @Lightwaves, @cdikibo, @AdventureBear, @SaintPeter, @mildused, @ArielLeslie, @qmikew1, @dting
Did you know that…? If we consider that a person becomes no-active in Gitte Help room (ie. doesn't show activity) if they don't send any message after 15 days of their last participation, we got that 721 of the total participants (1726) in the Gitter FCC/Help room wer considered active when observing activity during 2014-12-30 and 2015-08-16. Similarly, if we assume that a participant is a frequent one if he/she has at least 5 messages at different days, that number is drastically lower: only 308 people were considered as frequent visitors, visiting Gitter Help room to an average rate of 42% of the all days between their first and last message, being the longest stack 83 days long.
The data above should be cautiously manipulated: there are sure lots of invisibles: for example people who are passively visiting Gitter (eg. those who come just to read). There are other factors that affect the presence of users at Gitter rooms. That doesn't make the information useless. Data of this kind is useful if you can find that it is "stable" or a "signature" of a particular room, and you want to see if some changes "disturb" that stability.
@cdikibo hasn't been involved much in the project due to problems with her computer that it has taken her about 2 weeks long.
-- We found some small issues and we dedicated the time to them: we were discussing the best way to organise files and today I was working on normalising the date representations of the data, to allow for future comparisons. In the next few days I will still present a first demo through C9. Let you know.
-- In the few hours, between today and tomorrow I will load the results of a (very small) test of classification of randomly selected sentences from the Help room using nltk and the nps chat corpus.
-- Additionally I am working on a modification of the ubuntu chat corpus, in particular the bot-answerable questions (BAQ) corpus, in order to see if by making it more generic in its content, we can improve its precision. I will be also using the factoids list that comes with the corpus to modify the content of what we are considering as utterances (visit wikipedia). The modified file is ready but still not available.
-- The plan is to combine both corpora, the nps chat and the modified BAQ corpus, to check if that approach is good enough to capture tech questions in Gitter rooms.
-- There are some relevant challenges and "discoveries" about my last analysis that I could happily share with you if you want.
-- There is a proposal for carrying out a simple survey in the Gitter channels to collect demographic data. The proposal was mentioned to @BerkeleyTrue.
-- Load first demo in C9, with stacked views for at least two rooms.
-- Evaluate an implementation of redis.
-- Publish the modified files and scripts, likely presenting some python notebook.
-- Combine corpora, run the modifications and do simple tests.
-- Waiting for revision from FCC.
-- We will for the first time promote the room between other participants.
-- Load scripts on repos.