Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Contextual Font Recommendations based on User Intent

Sanat Sharma, Jayant Kumar, Jing Zheng, Tracy Holloway King

TL;DR

The paper addresses the challenge of selecting appropriate fonts from a vast Adobe Fonts library by introducing an intent-driven, multimodal recommendation system. It combines a large, curated dataset of text–intent–font triplets with a DistilBert-based creative-intent taxonomy and a multilingual text–intent classifier, trained alongside a font embedding model using Triplet Margin Loss to align fonts with user intents in a shared space. The system is deployed in Adobe Express with a scalable, low-latency service architecture and demonstrates strong production impact: CTR >25% and downstream export rates >50% when a recommended font is engaged, along with robust external and internal evaluations showing high relevance. This work enables diverse, contextually appropriate font recommendations across languages and platforms, highlighting practical gains in font discoverability and user engagement. It also outlines future work on harmonizing font choices with existing designs, considering global intents, and supporting RTL languages.

Abstract

Adobe Fonts has a rich library of over 20,000 unique fonts that Adobe users utilize for creating graphics, posters, composites etc. Due to the nature of the large library, knowing what font to select can be a daunting task that requires a lot of experience. For most users in Adobe products, especially casual users of Adobe Express, this often means choosing the default font instead of utilizing the rich and diverse fonts available. In this work, we create an intent-driven system to provide contextual font recommendations to users to aid in their creative journey. Our system takes in multilingual text input and recommends suitable fonts based on the user's intent. Based on user entitlements, the mix of free and paid fonts is adjusted. The feature is currently used by millions of Adobe Express users with a CTR of >25%.

Contextual Font Recommendations based on User Intent

TL;DR

The paper addresses the challenge of selecting appropriate fonts from a vast Adobe Fonts library by introducing an intent-driven, multimodal recommendation system. It combines a large, curated dataset of text–intent–font triplets with a DistilBert-based creative-intent taxonomy and a multilingual text–intent classifier, trained alongside a font embedding model using Triplet Margin Loss to align fonts with user intents in a shared space. The system is deployed in Adobe Express with a scalable, low-latency service architecture and demonstrates strong production impact: CTR >25% and downstream export rates >50% when a recommended font is engaged, along with robust external and internal evaluations showing high relevance. This work enables diverse, contextually appropriate font recommendations across languages and platforms, highlighting practical gains in font discoverability and user engagement. It also outlines future work on harmonizing font choices with existing designs, considering global intents, and supporting RTL languages.

Abstract

Adobe Fonts has a rich library of over 20,000 unique fonts that Adobe users utilize for creating graphics, posters, composites etc. Due to the nature of the large library, knowing what font to select can be a daunting task that requires a lot of experience. For most users in Adobe products, especially casual users of Adobe Express, this often means choosing the default font instead of utilizing the rich and diverse fonts available. In this work, we create an intent-driven system to provide contextual font recommendations to users to aid in their creative journey. Our system takes in multilingual text input and recommends suitable fonts based on the user's intent. Based on user entitlements, the mix of free and paid fonts is adjusted. The feature is currently used by millions of Adobe Express users with a CTR of >25%.
Paper Structure (13 sections, 1 equation, 6 figures, 1 table)

This paper contains 13 sections, 1 equation, 6 figures, 1 table.

Table of Contents

  1. Introduction: Application
  2. Data
  3. Model
  4. Intent Detection Model
  5. Creative Intent Taxonomy
  6. Multilingual Text-Intent Model
  7. Font Recommendation Model
  8. Service Architecture
  9. Evaluation
  10. External Manual Evaluation
  11. Internal Manual Evaluation
  12. In-Production Analysis
  13. Conclusion and Next Steps

Figures (6)

  • Figure 1: Example font recommendations in Adobe Express.
  • Figure 2: Training dataset of texts, intents and font used
  • Figure 3: Font characteristics based on intent: The font name is followed by the related intents (from most to least relevant). An example of the font is shown below the font name.
  • Figure 4: Triplet pairs used for training. We use intents and positive, negative fonts for training
  • Figure 5: Service Architecture for Font Recommendation
  • ...and 1 more figures