Text labelling app · Lab · Çoruh Şimşek

A Flutter phone app our team used to label text for training our psychology models; Barış Ülgen and I built it together around the turn of 2023.

What it is

FIG. 1One text, one rater. Several raters label the same texts, so each text ends up with a count per label rather than one answer. The app saves where each rater stopped.
Source: Labeler lib/screens/label_page.dart and home_page.dart, HEAD abc7cff

At dert., the company that later became CogniScope, our psychology models needed labelled text, so we built our own labelling app for phones. Each dataset sits in Firestore with its own set of labels. A rater signs in with a code sent to their phone, fills in a short profile, picks a dataset and gets one text at a time. They can choose several labels for a text, or a skip label when it is neutral, not applicable or ambiguous, and then move on. Every text keeps a running count of votes per label, and the app remembers where each rater stopped.

What happened

I wrote the first version in the last week of 2022: the main pages, choosing several labels at once, reading the datasets from Firebase, saving the labels and the phone sign-in screen. Barış Ülgen set up Firebase and the sign-in backend, fixed the Android build, added the skip label and a page that explains each dataset, and carried the app into 2023. The commits split evenly, 18 each.

It was our in-house tool for new training data. I have no record of how many texts were labelled with it.