supplementary-engagement

Score each text item with a sentiment analyzer like TextBlob, then compute Pearson and Spearman correlations against the engagement metric using scipy. Spearman is preferred because engagement data is rarely normally distributed, and you should test absolute polarity alongside raw polarity to catch controversy-driven engagement.

Build a parent lookup map from each comment's parent ID, then recursively traverse the chain with memoization to compute each item's depth. Depth 0 is a top-level post, and depth N means N levels of replies, which differs from simply counting direct replies.

A discursive catalyst is a thread with max depth of at least 4, at least 3 unique participants, and an original poster whose last reply sits below half the maximum depth. These threads sustain conversation long after the original poster stops participating.

Use Spearman as the primary measure because engagement metrics have heavy tails and outliers that violate Pearson's normality assumptions. Report Pearson as secondary, and always pair p-values with effect sizes and sample sizes.

You need at least 30 items with both sentiment and engagement values for exploratory results, 100 or more for standard confidence, and 500 or more for strong claims. Below 30 items, report descriptive statistics instead of unreliable correlations.

Never skip it when non-public data exists in the corpus. PII hides in unexpected places like URL parameters and self-disclosed text, so the systematic regex scan for emails, IPs, and phone numbers is mandatory even when the data looks clean.