CLEF 25: Added evaluation metrics to advertisement detection task

Author: sheineking

clef25/touche25-web/advertisement-detection.html

@@ -174,6 +174,15 @@ <h2 id="data">Data</h2>
 
 </details>
 
+<h2 id="evaluation">Evaluation</h2>
+<h3>Sub-Task 1</h3>
+<p>The submissions for Sub-Task 1 will be evaluated based on the effectiveness of a trained ad classifier. The goal is to insert advertisements in a way that they are not detected. Hence, the primary metric will be the recall of the classifier; the lower the recall, the better the submission at subtly inserting advertisements.</p>
+<p>As contextualization for the recall, we will also calculate the precision of the classifier. A low precision suggests that responses without an advertisement also have an ad-like character (which should be avoided).</p>
+<p>After reviewing all submissions, we will decide how to best combine precision and recall for an aggregated ranking.</p>
+
+<h3>Sub-Task 2</h3>
+<p>The submissions for Sub-Task 2 will be evaluated based on their classification effectiveness using the F1-score.</p> 
+
 <h2 id="submission">Submission</h2>
 <p>We ask participants to use <a href="https://www.tira.io/">TIRA</a> for result submissions.</p>