Sentiment Analysis is a feature that helps organizations control which photos are sent to publishers based on the tone of the photo’s caption or metadata. It does not analyze the image itself—only the descriptive text that comes with the photo (for example, “Mahomes stiff arms Ballentine during the first half…”).
Credit System: Sentiment Analysis uses a credit system. Each organization has a set number of sentiment analysis credits.
Opt-In Sources: The feature only runs for content sources you’ve opted in (e.g., AP photos).
Caching: Results are cached, so the same photo/caption combination isn’t analyzed multiple times.
Current Usage: The feature is already in use by organizations like the NFL.
Article Contents:
How Does Sentiment Analysis Work?
Text-Only Analysis: The system reads the caption or metadata text attached to each photo.
Player Sentiment: For every player tagged in the photo, the system classifies the caption as portraying them positively, negatively, or neutrally.
Content Routing: These sentiment results feed into your content routing rules. For example, you can set up a rule to “only send photos to a player’s publisher if the sentiment is positive or neutral.” This helps avoid sending photos of a player in a negative context (like being tackled) to their publisher.
How Is Sentiment Determined?
AI-Powered: The feature uses OpenAI’s GPT-4o-mini model, guided by a custom prompt with examples. It’s not a custom-trained model, but we can refine the prompt and examples to improve accuracy.
No Fine-Tuning: We do not train the model from scratch or fine-tune it for specific organizations. If you need better accuracy for certain scenarios, we can adjust the prompt and examples.
Example Use Case
Suppose you want to ensure that only positive or neutral photos of a player are sent to their publisher. You can set up a routing rule based on sentiment analysis results, so negative portrayals are automatically filtered out.
For more details or to adjust your sentiment analysis settings, please contact your account manager or support team.