CS 6474 Social Computing: Analyzing Language II

Transcription

1 CS 6474 Social Computing: Analyzing Language II Munmun De Choudhury Week 9 October 15, 2018

2 Midterm Presentations On Oct 29 and Oct teams in all Each team gets 15 minutes 10 minutes for actual presentation 5 minutes for Q&A All team members need to attend both days Any combination can present Schedule at the following link (also on class website): THi-ciBDoW5IyiVmVh547h2CkE8zY/edit#gid=0

3 Midterm Presentations Introduction and motivation Background Project Goals Expected outcomes Prior Work Work accomplished so far Projected plan

4 Gender and Power: How Gender and Gender Environment Affect Manifestations of Power

5 Summary

6 Summary

7 0.16% 0.135% 0.12% 0.114% 0.113% 0.08% 0.091% 0.086% 0.096% 0.072% 0.086% 0.04% 0% Subordinates% Superiors% Female% Male% Female% Subordinates% Female% Superiors% Male% Subordinates% Male% Superiors% 1.00$ 0.79$ 0.75$ 0.50$ 0.60$ 0.52$ 0.61$ 0.54$ 0.56$ 0.48$ 0.57$ 0.51$ 0.56$ 0.55$ 0.25$ 0.00$ Subordinates$ Superiors$ Female$Env$ Mixed$Env$ Male$Env$ Subordinates$ Superiors$in$ in$female$env$ Female$Env$ Subordinates$ in$mixed$env$ Superiors$in$ Mixed$Env$ Subordinates$ Superiors$in$ in$male$env$ Male$Env$

8 Would these results hold in modern enterprise social media contexts?

9 As people in corporations increasing adopt platforms like Facebook and Twitter, how do you expect these findings to generalize/change?

10 How do the preexisting social/corporate structures and biases at a large corporation impact the data, especially given that less than 35% of the unique discourse participants were women? The style of communication depends on the context or environment

11 How can the sociolinguistic findings on gender, gender environment, and power be useful for social computing research?

12 Ex: The power framework provided an interpretable and actionable set of hypotheses that could apply productively to other social situations, such as the difference in moderator vs. user behavior in an online forum.

13 Ex: Danescu-Niculescu-Mizil et al. identified connections between linguistic coordination and social power relations using discussions among Wikipedians and arguments before the U. S. Supreme Court

14 What kind of design considerations could incorporate the sociolinguistic findings on gender, gender environment, and power? What would they enable/what are the advantages?

15 No Country for Old Members: User lifecycle and linguistic change in online communities

16 Summary

17 Cross entropy Cross entropy Jan an All Newcomers Jan an Jan All Newcomers Jan Jan (a) BeerAdvocate an an an (a) BeerAdvocate Jan an Jan an Jan an (a) BeerAdvocate at their community-life in a way that allows comparison across users of with different activity levels and lifespans. ing To this end we define the life-stage of a user as the percentage tic of posts the user has already written, out of the total number of posts mi the user will ultimately write before abandoning the community. of 5 Thus, a life-stage of 0% corresponds to birth the moment the user app joined the community and a life-stage of 100% corresponds to ent death the moment the user leaves the community. pan User s distance com from the language of the community. Another tha key element of the proposed framework is the ability to measure a user s All reaction to linguistic change at a given stage in her Newcomers it i community-life. In the following we use several measures for linguistic change, each of them providing different perspectives on the the is t phenomenon. We start by quantifying the extent to which a userres is in tune with the community s norms by employing the snapshot aw language models defined in Section 2. In Figure 6 we (b) plot RateBeer end the average cross-entropy of a user s posts hy at different life-stages according to the snapshot language model Figure 5: ofexample the months ofincommunity-level which the respective change: posts were Predictability of eaching month s the evolution language, of cross-entropy calculated as over the the average users cross- lifespan we no- written. Observentropy of tice that that, month s in bothposts communities according under to snapshot study, users language follow a determined respective lifecycle: month When (blue, usersround join, markers; their language errorisbars far from that model of the are very tight, of thesometimes communitynot 6 (high visible). cross-entropy) Lower values and mean then users eas-graduallier to predict. approach Compare the current withlanguage the predictability of the community of the language (decreasing cross- used by users entropy); joining interestingly, that month after(green, about adiamond third of users markers). (ultimate) lifespan, their(b) language RateBeer. starts to again distance itself from that of the a u In (a) BeerAdvocate; community. It appears as if a user s language falls out of tune with at that of the community before she abandons the community. cal 3.2 UserSince lifecycle communities as well as individuals simultaneously evolve, by it is not clear whether the change in cross-entropy we just described We now is transition the result of from the user exploring actively user-level changing and her language communitylevel linguistic then away) changefrom in isolation that of to the the community main goalor, of the on the present lan Fig towards (and contrary, the work: Building result of a framework (b) RateBeer the evolvingfor community understanding normsthe getting interaction closer (respectively between these away) levels from of change. a static user. In particular, Thus, thewe increase focus on in cross-entropy analyzing in the 5 In sur Cross entropy Jan Jan Jan Jan Jan Jan Jan Jan lan

18 Class Exercise II Danescu-Niculescu-Mizil et al. say that [their] framework can be used to detect, early in a user s career, how long she will stay active in the community Describe two scenarios where this knowledge will be beneficial. Who are these stakeholders who can derive benefit?

19 How do evolving linguistic norms impact participation in anonymous communities?

20 Will the two-phase lifecycle (linguistic innovation learning and conservative phases) hold for/generalize to other online communities?

21 Chancellor, S., Pater, J. A., Clear, T., Gilbert, E., & De Choudhury, M. (2016, February). # thyghgapp: Instagram content moderation and lexical variation in pro-eating disorder communities. In Proc. CSCW.