Index de la documentation

Récupérez l’index complet de la documentation à l’adresse suivante : https://help.tritondigital.com/llms.txt

Utilisez ce fichier pour découvrir toutes les pages disponibles avant d’aller plus loin.

Méthodologie de classification contextuelle

Précédent Suivant

Sounder’s contextual category classification model identifies topics of conversation at the segment level of podcast episodes. It follows the IAB Content Taxonomy (3.0).

Classification contextuelle

Sounder determines the IAB categories of your content. First, it uses its proprietary automatic speech recognition (ASR) engine to transcribe podcast episodes into text. It then uses natural language processing (NLP) to break the transcription into segments that relate to specific IAB categories. To do this, it looks at the named entities, topics, tone, and other elements in each segment.

  • Segments: Sounder assigns one or more IAB categories to each segment.

  • Episodes: Sounder assigns an IAB category to an episode when it has at least three segments in the category.

  • Shows: Sounder assigns the IAB categories from all the show’s episodes from the past 90 days.

Example: A show's most recent episode was published on April 15. All episodes published and processed between January 15 and April 14 are included in the show-level score calculation.

Category Tiers

When it assigns a tier 2 or tier 3 category to a segment, episode, or show, it also assigns the category's parent tiers. Example: Sounder assigns "Auto Insurance" to a segment, which is a tier 2 category. Sounder also assigns the "Automotive" category, which is tier 1.

Contextual Category Confidence

Sounder uses the concentration and depth of conversations of identified categories to determine confidence scores. Par exemple, une discussion d’environ 6 à 8 minutes portant sur un seul sujet donne probablement lieu à un score de confiance de 100 % pour la catégorie contextuelle associée. En revanche, une conversation sur le même sujet qui ne dure que quelques minutes ou moins n’aboutira probablement qu’à un score de confiance de 60 % ou moins.

Sounder uses more sophisticated techniques than simple keyword-based classification. Instead it uses contextual signals such as sentiment and tone, so the precise number of keywords does not guarantee confidence score results.

Category Sentiment

Sounder labels each segment and episode’s IAB category with a sentiment: neutral, positive, negative, or mixed.

Sounder does not apply sentiment labels to episodes that were published before May 15, 2026.

Sentiment

Description

— Neutral

Doesn’t align with either positive or negative sentiment.

↑ Positive

Expresses a mostly favorable or supportive opinion.

↓ Negative

Expresses a mostly opposing or adverse opinion.

↑↓ Mixed

Has an equally positive and negative sentiment.

/ No sentiment detected

Sounder could not determine a sentiment, or the episode was published before May 15, 2026.


© 2026 Triton Digital, Inc. Tous droits réservés.