Digital-forms are commonly used for collecting structured information from users. However, filling digital-forms that include a large number of fields is tedious and error-prone. Auto-filling form fields for the user is highly beneficial for improving user experience and potentially collecting more valuable information (in cases where not all fields are mandatory). Online E-commerce marketplaces quite often utilize such forms to collect listing attributes from sellers. In this work, we describe Form-BERT -- a Transformer-based model which is optimized for auto-filling listing attributes given the following inputs: free-text, list of known attribute names, and zero or more attribute values. Form-BERT can be further used iteratively to leverage filled out attributes as the form filling progresses.

E-commerce platforms often use a predefined structured hierarchy of product categories. Apart from helping buyers sort between different product types, listing categorization is also critical for multiple downstream tasks, including the platform's main listing search. Traditionally, when creating a new listing, sellers need to assign the product they sell to a single category. However, the high diversity of product types in the platform, along with the hierarchy's low level of granularity result in tens of thousands of different possible categories that sellers need to pick from. This, in turn, creates a unique classification challenge, especially for sellers with a large number of listings. Moreover, the expected cost of making a category classification error is high, as it can impact the likelihood that their listing will get discovered by relevant buyers, and eventually sold.

Discovering similarities between online listings is a common backend task being used across different downstream experiences in eBay. Our baseline unstructured listing similarity method relies on measuring the semantic textual similarity between the embedding vectors of listing titles. However, we discovered that even with the latest contextualized embedding methods, our similarity fails to give the proper weight to the key tokens in the title that matter. This often results in identifying listing similarities that are not sufficient, which later hurts the downstream experiences. In this paper we present a method we call "Listing2Query", or "L2Q", which uses a Sequence Labeling approach to learn token importance from our users? search queries and on-site behaviour. We used pairs of listing titles and their matching search queries, and leveraged a contextualized character language model, to train L2Q as a bidirectional recurrent neural network to produce token importance weights. We demonstrate that plugging these weights into relatively straightforward listing similarity methods is a simple way to significantly improve the similarity results, even to the extent that it consistently outperforms those created by popular representations such as BERT. Notably, this approach is not reserved to only large online marketplaces but can be generalized to other cases that include a search-driven experience and a recall set of short documents.