Our Latest Release Brings IAB Tech Lab Content Taxonomy 2.2 Topic Classification – Adds Sensitive Topics analysis
Right on the heels of the IAB Taxonomy 2.1 classification release last month, Valossa AI has evolved again to detect an even broader set of video topics based on IAB Tech Lab Content Taxonomy v2.2, including 11 sensitivity categories to support brand safety and content appropriateness for advertising.
IAB content topics are being analyzed contextually inside the video modalities of speech, audio, visuals, and OCR. At the document level, Valossa AI summarises the most prominent categories that describe the entire video. Over 500 categories have been trained into the system.
As an example, we have broadened the NLP-based understanding of sensitive, inappropriate, or unwanted language use across speech and visual text to support a broad range of content styles. The improvement has also facilitated adding new topic categories such as “crime”, “death” and “hateful language”. The topics are complex even for modern content analysis, and we will keep working on the new categories in future updates.
Let’s look into an example of sensitive topics extraction from a collage of movie clips depicting death: Valossa AI identifies narrative topics about death from speech, visuals, and OCR. This is useful in contextual advertising scenarios where videos focusing on graphic depictions of death should be avoided.
Time coded content topics in a “movie deaths” video collage.
Rows of balloons visualize the location of time-coded topics in the video timeline
Also, with the powerful IAB 2.2 Content Taxonomy classification feature, Valossa AI provides proprietary categories that have been expanded to support over 60 different video topic types.
Intrigued? Our API documentation provides the full extent of categories AI detects from the video content.
Wikipedia References Added to Valossa Celebrity Recognition.
We have added Wikipedia references to the nearly 7000 globally notable celebrity identities trained into our celebrity face recognition engine.
Note! Future Changes in Valossa Metadata Format.
The following “explicit_content” detection types will be deprecated in Valossa Core Metadata and removed completely in the next update planned for October 2021:
- explicit_content.audio.offensive
- explicit_content.transcript.offensive
- explicit_content.ocr.offensive
The same detections can now be found in the following detection types tagged with Valossa tag category “bad_language”:
- audio.keyword.compliance
- transcript.keyword.compliance
- visual.text_region.keyword.compliance (OCR)
As always, these new features keep broadening our audio-visual intelligence abilities and are immediately available to our customers through Valossa AI subscriptions.
If you are interested in trialing our newest features and tools, fill up the form by clicking below.
