8. Conversation and Text Generation - [E3] 3. Danescu-Niculescu-Mizil, C., West, R., Jurafsky, D., Leskovec, J. and Potts, C., 2013.

[JunsolKim]

Post questions here for this week's exemplary readings: 3. Danescu-Niculescu-Mizil, C., West, R., Jurafsky, D., Leskovec, J. and Potts, C., 2013. “No country for old members: User lifecycle and linguistic change in online communities.” In Proceedings of the 22nd international conference on World Wide Web: 307-318.

[konratp]

I found it odd that the authors chose two beer review websites as the communities they want to study. One thing I'm confused by is -- and this is my own view -- the lack of relevance of beer review websites to society as a whole. I understand that the goal of the paper is methodological in nature, but I think that even then, other papers do a much better job at showcasing the methodological contribution on a more relevant example. The beer review websites are not particularly popular online, and I also think that by focussing on the two websites, the authors focus on a rather limited subset of society, as beer-drinkers are often thought of to be mostly men. The authors don't really provide a demographic breakdown of the data either, so it's hard to control for this. Instead, I think they could have focussed on a different bizarre internet phenomenon that is less fringe and perhaps more representative of our society.

[Jiayu-Kang]

The theory about the two-stage lifecycle of linguistic change is interesting! I understand that the authors targeted the two beer review websites because they are good sources of online community-based forums that has easily identifiable terms and norms. However, like Konrat's mentioned, it would produce more generalizable knowledge if focused on samples that are more representative of the society. Going off of that, using a more diverse sample might help answer some further questions: for example, I'd like to see how the interaction among users and among communities shape the patterns of linguistic change and user lifespans.

[Sirius2713]

Adding on Konratp and Jiayu's comments, this paper and the orienting reading expose the limit of data source when we are trying to use textual data for social science research. Therefore, I'm wondering if there're any technical methods to lower the bias or cross-validate the results concluded from some specific settings.

[ValAlvernUChic]

The paper is really cool! It seems like the paper is centering linguistic change and characteristics on word usage. While the authors demonstrated that this is absolutely sufficient for prediction, I was wondering how much more difficult it would be to investigate it in terms of sentence structure or syntactical adoption? I was also thinking that it would be interesting to try and map this phenomenon to offline cicerone circles to determine if it's a special function of the online space that this linguistic change happens.

[pranathiiyer]

Very interesting read! The authors touch upon the lexical nature of the discussions in these specific communities. I wonder how much of the prediction in such cases can be attributed to a given user's lack of interest in the subject matter itself over time. Of course the results still look pretty promising, I suppose based on anecdotal evidence, this could hold more strongly for communities where traction might not be affected by the topic itself which is so niche and could vary for a person over his/her life span.

[Jasmine97Huang]

Something interests me about the findings in the paper is the predictive power of linguistics changes on members’ lifetime in the community, It seems intuitive that members who aren’t interested in stay in the community are less confirming of the normative vocabularies. However, I am curious about the causal effect- does linguistic conservatism causes end of community membership or vice versa?

[hsinkengling]

I'm having trouble finding the support for the following sentence on p.307 "We show that this increasing gap is explained by the user ceasing to respond to changes in community norms: because the language of the community is constantly evolving a user that is unreceptive to this change will appear as fathering from the community." Is this explanation (desiring a perception as fathering from other members) supported elsewhere in the paper?

To add on to the discussion brought by Konrapt, Jiayu and others, I think that one of the advantages in working with this kind of tightly-knit online communities could be that we can actually try to answer more qualitative questions with online surveys and interviews even. This could enable us to articulate and falsify some hypotheses regarding the cause for linguistic shift.

[sizhenf]

Very interesting read! The model that the authors developed to predict user life span is very cool. My question is, sometimes we see, in the long run, the linguistic pattern/characteristics of a certain online community would change due to whatever reasons, and may cause many users to leave. I wonder if the authors have taken that into the consideration.

[Hongkai040]

The idea of this paper is novel. It told us 'what', but I'm more interested in 'why'. One more thing to add on is that the paper was published in 2013, which is 9 years ago now, so online communications changed a lot in many ways. In many online communities, we're in favor of user-defined emojis, and those emojis are more contagious than mere texts in many ways. I am wondering will the framework still hold true for these patterns?

[mikepackard415]

The findings of this paper seem pretty intuitive and generalizable to me. For example, I wonder if we could study whether some of the patterns we see in these online beer communities are also found in political discourse generally? Younger folks driving the evolution of topics, older folks tending to be reluctant to accept new vocabulary.

[chentian418]

I am curious about the mechanism behind the snapshot language model: how can cross-entropy according to the snapshot language model reflect the surprising component of the post language with respect to the language the community was using at the time point it was written. Also, can we improve this model to calculate some dynamic evolution of this surprising component across a longer time domain?

[Qiuyu-Li]

I'm the presenter and I think the paper is very interesting! My only concern is that the predictive power of the features seems to be not very high. Even after 5 additional features are added, the F-score only raises from 30.5 to 42.9 But it might due to my lack of knowledge about this area and this is actually a decent increase.

[melody1126]

An interesting finding of this paper is that linguistic rigidness has less to do with biological age of the user but more to do with the user lifespan on that platform – the authors also used this to build language predictor for user lifespan. I think this paper goes well with the structural v.s. cultural embeddedness article we read for the orienting reading this week – it would be interesting to map both the structural embeddedness in terms of network brokerage in online communities, and the cultural embeddedness (which seems to be what this article showed) together.