Adobe Express is expanding internationally, but the US has a disproportionately large content supply and interaction volume. Learning-to-rank (LTR) models trained primarily on behavioral feedback inherit this imbalance: templates popular in US are over-served in non-US locales. This cross-locale exposure bias suppresses local content discoverability and degrades ranking quality in growth locales. We show that click-only training suppresses semantically informative localization features. Adding vision-language model (VLM) graded relevance labels as auxiliary supervision alongside clicks improves semantic alignment but does not preserve local content visibility. We propose a multi-objective framework combining behavioral supervision, VLM-derived relevance signals, and locale-aware boosting. Across five locales, the resulting model improves relevance while restoring stable localization, demonstrating the importance of disentangling exposure from semantic supervision.

In this work, we investigate how small language models (SLMs) can be scaled to support multimodal search and recommendation use cases while remaining efficient enough for real-time, resource-constrained deployments. We present a framework that combines upside-down reinforcement learning with synthetic data distillation from a large language model (Llama-3 [1]) to train a $\mathbf{1 0 0 M}$-parameter GPT-2 model [2] for multitask prompt generation. Despite being up to 80 times smaller than state-of-the-art large language models (LLMs), our SLM achieves relevance and diversity scores within 6% of competitive baselines such as Llama-3 8B, Qwen3 8B, and Ministral 8B. These results demonstrate that SLMs can effectively handle multimodal search and recommendation tasks, while dramatically reducing inference latency and memory overhead. Our study highlights the potential of lightweight models as practical engines for scalable multimodal discovery, bridging the gap between cutting-edge research and real-world multimodal applications such as media recommendations and creative content generation.

In this work, we investigate how small language models (SLMs) can be scaled to support multimodal search and recommendation use cases while remaining efficient enough for real-time, resource-constrained deployments. We present a framework that combines upside-down reinforcement learning with synthetic data distillation from a large language model (Llama-3) to train a 100M-parameter GPT-2 model for multitask prompt generation. Despite being up to 80 times smaller than state-of-the-art large language models (LLMs), our SLM achieves relevance and diversity scores within 6

The eCommerce search and recommendations space is a unique and dynamic domain within information retrieval (IR), characterized by its multimodality and industry-driven challenges. While the basic task of fulfilling a user's information need aligns with web search, the methodologies employed are distinct. On eCommerce platforms (e.g. Alibaba, Amazon, eBay, Etsy, Flipkart, Walmart), the data available for retrieval and ranking differs significantly, as do the success signals (e.g. adding items to a cart, purchasing). Our focus for 2025 is on fostering deeper engagement through interactive discussions, exploring crucial topics such as navigating irreproducibility in research-to-product pipelines, and addressing emerging topics such as evaluation metrics for LLMs, multimodality, and the interplay between organic and sponsored search. With our discussion-heavy format and structured facilitation, we aim to spark conversation among all participants.

Knowledge graphs are useful tools to organize, recommend and sort data. Hierarchies in knowledge graphs provide significant benefit in improving understanding and compartmentalization of the data within a knowledge graph. This work leverages large language models to generate and augment hierarchies in an existing knowledge graph. For small (<100,000 node) domain-specific KGs, we find that a combination of few-shot prompting with one-shot generation works well, while larger KG may require cyclical generation. We present techniques for augmenting hierarchies, which led to coverage increase by 98 and 99

Accurate explicit and implicit product identification in search queries is critical for enhancing user experiences, especially at a company like Adobe which has over 50 products and covers queries across hundreds of tools. In this work, we present a novel approach to training a product classifier from user behavioral data. Our semantic model led to >25% relative improvement in CTR (click through rate) across the deployed surfaces; a >50% decrease in null rate; a 2x increase in the app cards surfaced, which helps drive product visibility.

As user content and queries become increasingly multi-modal, the need for effective multi-modal search systems has grown. Traditional search systems often rely on textual and metadata annotations for indexed images, while multi-modal embeddings like CLIP enable direct search using text and image embeddings. However, embedding-based approaches face challenges in integrating contextual features such as user locale and recency. Building a scalable multi-modal search system requires fine-tuning several components. This paper presents a multi-modal search architecture and a series of AB tests that optimize embeddings and multi-modal technologies in Adobe Express template search. We address considerations such as embedding model selection, the roles of embeddings in matching and ranking, and the balance between dense and sparse embeddings. Our iterative approach demonstrates how utilizing sparse, dense, and contextual features enhances short and long query search, significantly reduces null rates (over 70%), and increases click-through rates (CTR). Our findings provide insights into developing robust multi-modal search systems, thereby enhancing relevance for complex queries.

The IR community as a whole is considering whether search and recommendations can move entirely to embedding-based technologies. This SIRIP panel discusses the future of embedding-based technologies in industry search given its broad range of document types, its specific query types, its performance requirements, and the features that accompany search. The panel comprises long-time industry experts and academics with industry ties. The panelists vary as to whether they believe that the industry in practice will move entirely to embeddings or will remain a hybrid domain.

With the advent of multimodal LLMs and release of open-source multimodal models, the potential for multimodal search and recommendations has significantly increased. Multimodal systems offer a next-gen customer experience by creating a shared embedding space for text, images, audio, etc. These advancements enable more accurate, personalized recommendations, enhancing user satisfaction and engagement. This workshop on Multimodal Search and Recommendations explores the latest advancements, challenges, and applications of multimodal search and recommendations.

This work leverages neural transformers to generate hierarchies in an existing knowledge graph. For small ( < 10,000 node) domain-specific KGs, we find that a combination of few-shot prompting with one-shot generation works well, while larger KG may require cyclical generation. Hierarchy coverage increased by 98

ECOM24 brings together researchers and practitioners from academia and industry to identify and discuss core research problems in eCommerce search and recommendation. The workshop aims to foster collaboration, to attract research funding, and to introduce IR researchers and postgraduate students to eCommerce product discovery. The workshop features a special theme of eCommerce search in the age of Generative AI and LLMs and a data challenge in collaboration with TREC on how end-to-end retrieval systems can be built and evaluated given a large set of products.

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%.

Search has become an integral part of Adobe products and users rely on it to learn about tool usage, shortcuts, quick links, and ways to add creative effects and to find assets such as backgrounds, templates, and fonts. Within applications such as Photoshop and Illustrator, users express domain-specific search intents via short text queries. In this work, we leverage sentence-BERT models fine-tuned on Adobe's HelpX data to perform multi-lingual semantic search on help and tutorial documents. We used behavioral data (queries, clicks, and impressions) and additional annotated data to train several BERT-based models for scoring query-document pairs for semantic similarity. We benchmarked the keyword-based production system against semantic search. Subsequent AB tests demonstrate that this approach improves engagement for longer queries while reducing null results significantly.

eCommerce Information Retrieval (IR) is receiving increasing attention in the academic literature and is an essential component of some of the largest web sites (e.g. Airbnb, Alibaba, Amazon, eBay, Facebook, Flipkart, Lowes's, Taobao, Target). SIGIR has for several years seen sponsorship from eCommerce organizations, reflecting the importance of IR research to them. The purpose of this workshop is (1) to bring together researchers and practitioners of eCommerce IR to discuss topics unique to it, (2) to determine how to use eCommerce's unique combination of free text, structured data, and customer behavior data to improve search relevance, and (3) to examine how to build datasets and evaluate algorithms in this domain. The theme of this year's eCommerce IR workshop is Foundation Models and Unified Information Access in eCommerce. The workshop solicits papers on this topic and includes a panel focused on this area. In addition, Lowe's is sponsoring an eCommerce data challenge on Cross-modal and Multi-modal Visual Search for eCommerce. The data challenge reflects themes from the successful SIGIR workshops in 2017, 2018, 2019, 2020, 2021, and 2022. ECOM23 will be held as a full day hybrid workshop to accommodate for diverse participation.

Spellchecking is one of the most fundamental and widely used search features. Correcting incorrectly spelled user queries not only enhances the user experience but is expected by the user. However, most widely available spellchecking solutions are either lower accuracy than state-of-the-art solutions or too slow to be used for search use cases where latency is a key requirement. Furthermore, most innovative recent architectures focus on English and are not trained in a multilingual fashion and are trained for spell correction in longer text, which is a different paradigm from spell correction for user queries, where context is sparse (most queries are 1-2 words long). Finally, since most enterprises have unique vocabularies such as product names, off-the-shelf spelling solutions fall short of users' needs. In this work, we build a multilingual spellchecker that is extremely fast and scalable and that adapts its vocabulary and hence speller output based on a specific product's needs. Furthermore, our speller out-performs general purpose spellers by a wide margin on in-domain datasets. Our multilingual speller is used in search in Adobe products, powering autocomplete in various applications.

Search has become a key component of many on-line user experiences. Search queries are usually textual and hence should benefit from improvements in natural language processing. However, many of the NLP algorithms used in production systems fail for queries that require structured understanding of the query and document or that require reasoning. These issues arise because of the way information is stored in the search index and the need to return results quickly. The issues are exacerbated when searching over non-textual documents, including images and structured data. The use of embedding-based techniques has helped with some types of searches, especially when the query vocabulary does not match that of the documents and when searching over images. However, these techniques still fail for many searches, especially ones requiring reasoning. Simply combining classic word-level search and embedding-based search does not solve these issues. Instead, in this position paper, I argue that we need to create hybrid systems from traditional search techniques, embedding-based search, and the addition of structured data and reasoning. Enabling such hybrid systems will require a deep understanding of linguistic representations of meaning, of information retrieval optimization, and of the types of information encoded in the queries and documents. It is my hope that this paper inspires further collaboration across disciplines to improve these complex search problems.

eCommerce Information Retrieval (IR) is receiving increasing attention in the academic literature and is an essential component of some of the world's largest web sites (e.g. Airbnb, Alibaba, Amazon, eBay, Facebook, Flipkart, Lowe's, Taobao, and Target). SIGIR has for several years seen sponsorship from eCommerce organisations, reflecting the importance of IR research to them. The purpose of this workshop is (1) to bring together researchers and practitioners of eCommerce IR to discuss topics unique to it, (2) to determine how to use eCommerce's unique combination of free text, structured data, and customer behavioral data to improve search relevance, and (3) to examine how to build datasets and evaluate algorithms in this domain. Since eCommerce customers often do not know exactly what they want to buy (i.e. navigational and spearfishing queries are rare), recommendations are valuable for inspiration and serendipitous discovery as well as basket building. The theme of this year's eCommerce IR workshop is Bridging IR Metrics and Business Metrics and Multi-objective Optimization. The workshop includes papers on this topic as well as a panel focused on this area (see Section 3). In addition, Farfetch is sponsoring a recommendation challenge focused on outfit completion: as part of the event, Farfetch will release to the research community a novel, large dataset containing multi-modal information and extensive labels curated by fashion experts. The data challenge reflects themes from prior SIGIR workshops in 2017, 2018, 2019, 2020, 2021.