HASHTRACKER

Social listening 2 min read

Social media data for research that survives peer review

Collecting social media data for research raises three questions every time: how the sample was gathered, what the query was and whether the data can be inspected.

Cover drawn as mentions landing over time

Using social media data for research has a requirement most commercial listening does not: someone is going to check it.

That changes which tool you can use and what you have to write down.

What a reviewer will ask

Three questions come up every time. How was the sample collected, what exactly was the query, and can the data be inspected. A dashboard screenshot answers none of them. A dataset with a stated methodology answers all three.

Tell us what you are studying and we will scope the collection around the methodology you need to defend.

Tell us what you are studying

Why the collection method decides publishability

Data scraped without permission creates problems at review and at publication, and increasingly at the ethics committee before either. Collection through the official API on its stated terms is a documentable method. You can write down what was collected, under what terms and with what limits, and that paragraph is what makes the rest of the paper defensible.

Everything we deliver is gathered Twitter compliant, which exists precisely for this situation.

Bounding a sample you can describe

A sample described as "posts about the election" is not reproducible. One described as a query, a period and an explicit set of exclusions is.

("candidate name" OR #ElectionHashtag) lang:es

Write the query into the methodology section verbatim. If it changed during collection, write both versions and when the change happened.

What to record beyond the posts

For each post: text, timestamp, language, engagement and impressions. For each author: account age, follower and following counts, and whether verified. The author fields are what allow the sample to be characterised. A study that cannot describe who was posting cannot claim much about what was said.

Deletion, and what it does to a dataset

Posts disappear. Authors delete them, accounts get suspended, and a dataset collected today will not match one collected next month. Record the collection date and treat it as part of the method. A sample is a snapshot of a moment, and saying so is more rigorous than implying completeness.

The platform's terms also govern what can be stored and for how long. That constraint is part of the research design, not an afterthought.

Working with what you get

A CSV with one row per post loads into R, Python or anything else. Code the text however your method requires, join it against your own variables, and keep the raw file so the coding can be redone.

The point is that the analysis belongs to you and can be repeated by someone else. Tell us the period and the terms and we will put the collection together.

Tell us the period and the terms

Choosing a subscription

Every subscription carries the same product. The only decision is how many tweets you need to analyse.

See the subscriptions