One of LDA’s disadvantages is that there is no standard method for deciding model parameters. Typically researchers examine the output of different models and decide which one is more interpretable, which is the approach we took. We were able to interpret an LDA model with 50 latent topics better than models with 10, 30, 60, or 100 topics. We then generated dictionaries to represent each topic from the 500 terms most strongly associated with that topic. Two experts familiar with SNS content manually labeled each dictionary (e.g. Food). We also excluded were topics that judges could not interpret. Because the status updates were from a single month’s text, several topics were clearly capturing popular short-term memes from that month and were excluded from analysis because of their limited generalizability. This left 25 topics. These topics with their top terms are shown in Table 1.