How Meta leverages generative AI to understand user intent

dead — the parent company of Facebook, Instagram, WhatsApp, Threads, and more — runs one of the world’s largest recommendation systems.

In two recent papers, its researchers reveal how generative models can be used to better understand and respond to user intent.

By viewing recommendations as a generative problem, you can approach them in new ways that are richer in content and more efficient than classical approaches. This approach can have important uses for any application that requires retrieving documents, products, or other types of objects.

Dense versus generative retrieval

Standard approach to creation Recommendation systems It is the computation, storage and retrieval of dense representations of documents. For example, to recommend items to users, the application must train a model that can calculate Implications For both users and items. Then it must create a large store of item inclusions.

At inference time, the recommender system attempts to understand the user’s intent by finding one or more items whose embeddings are similar to those of the user. This approach requires an increasing amount of storage and computation capacity as the number of items increases because each item embedding must be stored and each recommendation operation requires comparing the user’s embedding against the entire item store.

Generative retrieval is a newer technique that attempts to understand user intent and make recommendations by predicting the next item in a sequence rather than searching a database. Generative retrieval does not require storing item embeddings and the inference and storage costs remain constant as the list of items grows.

The key to doing generative retrieval work is to calculate “semantic identifiers” (SIDs) that contain contextual information about each item. Generative retrieval systems e.g tiger Work in two stages. First, the coding model is trained to generate a unique embedding value for each item based on its description and properties. These include values ​​become security identifiers (SIDs) and are stored with the item.

In the second stage A Transformers model It is trained to predict the next SID in the input sequence. The list of input SIDs represents the user’s interactions with previous items and the model prediction is the SID of the recommended item. Generative retrieval reduces the need to store and search across embeddings of individual items. It also enhances the ability to capture deeper semantic relationships within data and provides other benefits to generative models, such as adjusting temperature to adjust the diversity of recommendations.

Advanced generative retrieval

Despite its low storage and inference costs, generative retrieval suffers from some limitations. For example, it tends to over-fit items it saw during training, which means it has trouble handling items that were added to the catalog after the model was trained. In recommender systems, this is often referred to as “ Cold start problem“, which relates to new users and items and has no interaction history.

To address these shortcomings, Meta has developed a hybrid recommendation system called Ligerwhich combines the computational and storage efficiencies of generative retrieval with the strong embedding quality and classification capabilities of dense retrieval.

During training, LIGER uses both similarity scores and next symbol targets to improve the model’s recommendations. During inference, LIGER selects several candidates based on the generative mechanism and supplements them with some cold start elements, which are then ranked based on the embeddings of the generated candidates.

The researchers noted that “integrating dense and generative retrieval methods holds tremendous potential for the development of recommender systems” and as models develop, “they will become increasingly practical for real-world applications, enabling more personalized and responsive user experiences.”

In a separate paper, the researchers presented a new multimodal generative retrieval method called Multimedia preferences selector (Mender), a technology that can enable generative models to capture implicit preferences from a user’s interactions with different elements. Mender builds on generative retrieval methods based on security identifiers (SIDs) and adds some components that can enrich recommendations with user preferences.

Mender uses a large language model (LLM) to translate user interactions into specific preferences. For example, if a user praises or complains about a particular item in a review, the model will summarize that into a preference about that product category.

The main recommendation model is trained to be conditional on both the sequence of user interactions and user preferences when predicting the next semantic identifier in the input sequence. This gives the recommendation model the ability to generalize, perform learning in context, and adapt to user preferences without being explicitly trained in them.

“Our contributions pave the way for a new class of generative retrieval models that unlock the ability to use organic data to guide recommendations via user textual preferences,” the researchers write.

Implications for enterprise applications

The efficiency provided by generative retrieval systems can have important implications for enterprise applications. These advances translate into immediate practical benefits, including lower infrastructure costs and faster reasoning. The technology’s ability to keep storage and inference costs constant regardless of catalog size makes it especially valuable for growing companies.

The benefits span across industries, from e-commerce to enterprise search. Generative retrieval is still in its early stages and we can expect applications and frameworks to emerge as it matures.

Daily insights into business use cases with VB Daily

If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from organizational transformations to hands-on deployments, so you can share insights to maximize ROI.

Read our privacy policy

Thanks for subscribing. Check more VB newsletters here.

An error occurred.