We present an application of encoder-only Transformers to the task of recommending filters to eCommerce search queries. In particular, we operate in a dynamic setup where new recommendations are computed online whenever the user selects one or more filters, conditioned on the search query and the filters selected so far. Our method leverages the world knowledge imparted into a pretrained model, setting it apart from purely memory-based or statistical models, which we use as baselines for evaluation. We review experimental results on offline benchmarks using data generated from eBay search logs, comparing the performance of the proposed model to the baselines. The results show a significant increase in filter recommendation accuracy, as measured by NDCG.

Aspect extraction - the task of identifying attributes such as model, color, or size from textual entities like question-answer pairs - is one of the key tasks in eCommerce. Given the large number of possible aspects per entity, retrieving the most relevant ones is challenging. Traditional methods rely on high-quality labeled data for training, which is costly to obtain at scale. In this work, we propose a training-free aspect extraction approach using LLMs. Leveraging in-context learning and a novel Forward-Backward method that combines retrieval-augmented generation (RAG) with embedding-based matching, our method effectively extracts relevant aspects from text without requiring training data.

In search, autocomplete (AC) is an essential tool that provides suggestions for each keystroke, functioning well with token-based queries. However, it is challenging to handle at scale when input queries are conversational and semantically rich. Identifying relevant queries for sub-tokens requires efficient lookup strategies, real-time ranking, and relevance in the results. This work integrates Retrieval-Augmented Generation (RAG), AI safety, and relevance ranking to produce autocomplete suggestions for conversational queries in a production system. RAG-based responses ensure a high hit ratio for popular AC inputs and maintain a very low risk category by not triggering any critical AI safety concerns.

Compound words are a grammatical structure that allows forming new words by composing existing words. For e-commerce search in German, it is essential to split these compounds into meaningful parts because item titles often use the joint form while search queries are often split. We propose a method for German compound splitting leveraging a large language model (LLM) with a voting mechanism and a hyperparameter search for automatically optimizing prompt and parameter combinations. Our evaluation of the proposed method on human-created gold standard data for e-commerce shows that it outperforms existing methods for compound splitting in this domain.