| Introduction | In a random sample of recipe reviews from allrecipes.c0m, we found that 57.8% contain refinements of the original recipe. |
| Introduction | We created a new recipe data set, and manually labeled a random sample to evaluate our model and several baselines. |
| Models | In a manually labeled random sample of recipe reviews, we find that refinement segments tend to be clustered together in certain reviews (“bursty”), rather than uniformly distributed across all reviews. |