Exploring Hierarchical Interest Representation For Meta Ads Deep Funnel Optimization

Summary

Hierarchical Interest Representation is an upstream representation layer for Meta Ads that addresses sparse deep-funnel feedback by connecting users’ inferred interests with advertisers, products, services, and other ads entities. It models a typed, heterogeneous engagement graph with a transformer architecture, sparse attention for long-range relationships, bias-aware attention, self-supervised cross-view distillation, and multimodal content features processed through vision or language models. The system projects the raw graph into learned interest primitives and cascading super-graphs, producing universal embeddings and Bag-of-Meaning tokens at multiple granularities. Temporal cutoffs and chronological splits prevent future-information leakage, while a pipelined online graph engine reportedly achieved a 30x wall-clock speedup over a synchronous baseline during testing. Trained end-to-end on billions of interactions, the representations are designed for personalization, retrieval, supervision, and ranking integrations including GEM, Andromeda, and the Adaptive Ranking Model.

Takeaways

1. 1

   The raw engagement graph is projected into a configurable super-graph whose learned interest primitives make sparse user-ad connections denser and provide a more stable vocabulary.

2. 2

   Bag-of-Meaning tokens discretize continuous embeddings into a compact, unordered interest vocabulary suitable for inverted-index retrieval, set aggregation, personalization, supervision, and interpretation.

3. 3

   Temporal cutoffs, chronological data windows, pipelined graph fetching, and bit-exact reproducibility support leakage prevention and reliable infrastructure operation; testing showed a 30x wall-clock speedup over a synchronous baseline.