Contextual & Semantic Data Providers — Terminology
PDF download form at the bottom of this page.
ASR/OCR (Audio & Image Text Extraction)
Summary: Automatic speech recognition and optical character recognition to turn audio/video/imagery into text for classification.
Use Cases:
-
Extract podcast/video transcripts for topic analysis
-
Use OCR on screenshots to detect brand or sensitive text
Attention Context Signals
Summary: Layout and format cues (position, size, media type) that correlate with attention and engagement.
Use Cases:
-
Prefer placements adjacent to long-form or interactive elements
-
Use attention cues to weight contextual scores
Brand Adjacency
Summary: Proximity to specific content entities or keywords used to align or avoid context around the ad.
Use Cases:
-
Ensure ads appear next to favorable topics or partners
-
Exclude adjacency to competitor or sensitive entities
Brand Safety Frameworks (GARM)
Summary: Industry frameworks defining safety/suitability categories and risk tiers.
Use Cases:
-
Align client policies to GARM tiers
-
Audit partners for consistent safety labeling
Context Drift & Re-Training Cadence
Summary: Monitoring shifts in language/content and retraining models to maintain precision and recall.
Use Cases:
-
Refresh classifiers quarterly for new slang/events
-
Backtest drift windows against ground truth
Cookieless Targeting
Summary: Using contextual and environmental signals as primary targeting inputs when user IDs are limited or unavailable.
Use Cases:
-
Reach scale on browsers without third‑party cookies
-
Activate campaigns purely on content signals in CTV/OTT
Creative-to-Context Matching
Summary: Optimizing creative message and format to the detected context and sentiment.
Use Cases:
-
Swap copy when sentiment is negative vs. positive
-
Use lightweight formats on fast, mobile contexts
CTV/OTT Contextual Metadata (EIDR/Genre/Rating)
Summary: Program-level metadata (genre, rating, EIDR/IMDB IDs) enabling contextual targeting in streaming.
Use Cases:
-
Target family-friendly genres and exclude mature ratings
-
Attribute outcomes by show/genre rather than user IDs
Entity Disambiguation
Summary: Resolving ambiguous names/terms to the correct real‑world entity or concept using context.
Use Cases:
-
Differentiate ‘Apple’ the brand from the fruit in classification
-
Map ‘Jaguar’ to the car brand vs. the animal as needed
Explainability & QA (Model Audits)
Summary: Techniques and processes to explain model decisions and validate outputs for bias and drift.
Use Cases:
-
Use SHAP-style feature attributions in reviews
-
Run human-in-the-loop spot checks on high-risk topics
Geo/Language Detection (Multilingual NLP)
Summary: Detecting page language/locale and applying region-specific models and taxonomies.
Use Cases:
-
Route Spanish pages to ES taxonomy models
-
Block regionally sensitive topics by locale
Knowledge Graphs & Ontologies
Summary: Structured graphs that connect entities, topics, and relationships to improve classification and disambiguation.
Use Cases:
-
Expand topics via graph neighbors for recall
-
Resolve aliases and brand families consistently
Named Entity Recognition (NER)
Summary: NLP task that identifies entities (people, orgs, products, places) mentioned in content.
Use Cases:
-
Trigger brand adjacency rules when entities appear
-
Enrich contextual segments with entity facets
On-Device NLP
Summary: Classification and enrichment performed locally on the device or browser to minimize data transfer.
Use Cases:
-
Score page topics in-browser for privacy
-
Enable mobile app contextual targeting without server calls
Page Quality & Readability Signals
Summary: Signals such as ad density, readability, CLS/LCP proxies, and spam heuristics to filter low-quality pages.
Use Cases:
-
Exclude MFA (made-for-advertising) pages
-
Favor pages with higher readability for brand campaigns
Page- vs. Article-Level
Summary: Whole-page classification (all elements) versus the specific content unit (article/video/post).
Use Cases:
-
Use article-level classification for precise targeting
-
Fall back to page-level when unit granularity is unavailable
Privacy Sandbox Topics API
Summary: Browser-provided interest topics designed to enable interest-based ads without third‑party cookies.
Use Cases:
-
Augment contextual signals with browser topics (when available)
-
Respect per‑user browser privacy controls
Real-Time Classification
Summary: On‑the‑fly page/app categorization to support programmatic bidding and brand safety decisions.
Use Cases:
-
Return topics within bid timeouts for OpenRTB
-
Live-score UGC feeds before monetization
Safety vs. Suitability
Summary: Avoiding unsafe content (safety) versus aligning with nuanced contexts that fit brand values (suitability).
Use Cases:
-
Block violence/misinformation universally
-
Allow news but exclude specific sub-topics per brand
Semantic Similarity
Summary: Vector representations (embeddings) that capture meaning beyond exact keyword matches.
Use Cases:
-
Expand reach to semantically related content
-
Reduce false negatives from keyword-only rules
Sentiment & Emotion Analysis
Summary: Classification of tone and affect (positive/negative/neutral; emotions) to refine suitability and messaging.
Use Cases:
-
Avoid negative sentiment around brand placement
-
Amplify positive product reviews contexts
Taxonomy (IAB/Custom)
Summary: Structured content categories used for targeting, reporting, and suitability controls.
Use Cases:
-
Map to IAB Tech Lab Content Taxonomy
-
Maintain custom tiers for stricter client policies
Topic Modeling & Keyphrase Extraction
Summary: Unsupervised (e.g., LDA) and supervised methods to derive topics and keyphrases from text.
Use Cases:
-
Discover emergent themes to add to taxonomy
-
Generate keyphrases for SEO and contextual bidding
Visual Context (Computer Vision)
Summary: Use of image/video understanding (objects, scenes, logos) to add non-textual context.
Use Cases:
-
Detect alcohol/firearms imagery for safety filters
-
Match creatives to detected scenes (e.g., beach, sports)
