Ebay, Incorporated

Mingkuan Liu

Hao Zhang

Xianjing Liu

Alan Qing Lu

Technology for the improved processing of search queries is provided. In one embodiment, methods may return semantically relevant search results for a search query. During a pre-computing offline processing, an inventory semantic index may be generated and may include inventory binary hashing signatures that are associated with inventory listings, such as goods or services for sell, and the index may be partitioned by categories and shards. When a search query is received, relevant categories are determined using a relevant category recognition service, and a search query binary hashing signature maybe generated for the search query. The relevant categories are searched to determine hamming distances between the inventory binary hashing signatures and the search query binary hashing signature, where the hamming distance indicates semantic relevance.

Search system for providing search results using query understanding and semantic binary signatures

In an example, one or more leaf category specific unsupervised statistical language model (SLM) models are trained using sample item listings corresponding to each of one or more leaf categories and structured data about the one or more leaf categories, the training including calculating an expected perplexity and a standard deviation for item listing titles. A perplexity for a title of a particular item listing is calculated and a perplexity deviation signal is generated based on a difference between the perplexity for the title of the particular item listing and the expected perplexity for item listing titles in a leaf category of the particular item listing and based on the standard deviation for item listing titles in the leaf category of the particular item listing. A gradient boosting machine (GBM) fuses the perplexity deviation signal with one or more other signals to generate a miscategorization classification score corresponding to the particular item listing.

Miscategorized outlier detection using unsupervised SLM-GBM approach and structured data

In accordance with an example embodiment, large scale category classification based on sequence semantic embedding and parallel learning is described. In one example, one or more closest matches are identified by comparison between (i) a publication semantic vector that corresponds to at least part of the publication, the publication semantic vector based on a first machine-learned model that projects the at least part of the publication into a semantic vector space, and (ii) a plurality of category vectors corresponding to respective categories from a plurality of categories.

Semantic category classification

Embodiments of the present disclosure relate generally to semantic indexing to improve search results of a large corpus. Some embodiments identify one or more closest matches between (i) a search semantic vector that corresponds to a search query, the search semantic vector based on a first machine-learned model that projects the search query into a semantic vector space, and (ii) a plurality of publication vectors corresponding to respective publications in the publication corpus, the plurality of publication vectors based on a second machine-learned model that projects the plurality of publication vectors into the semantic vector space.

Semantic forward search indexing of publication corpus

Embodiments of the present disclosure relate generally to index trimming to improve search results of a large corpus. Some embodiments, prior to receiving, from a user device, a search query of one or more keywords searching for a relevant set of publications in a publication corpus, trim candidate publications from a plurality of candidate publications to generate a trimmed plurality of candidate publications.

Search index trimming

Embodiments of the present disclosure relate generally to semantic indexing to improve search results of a large corpus. Some embodiments, with at least one of the keywords of the search query encoded by a semantic vector in a semantic vector space, identify a plurality of candidate publications in the publication corpus, the plurality of candidate publications encoded by a cluster of a plurality of semantic vectors in the semantic vector space, the identifying based on proximity in the semantic vector space between the at least one of the keywords of the search query and keywords in the plurality of candidate publications, the proximity based on a first machine-learned model that projects the at least one keyword in the search query and the keywords in the plurality of candidate publications into the semantic vector space.

Semantic reverse search indexing of publication corpus

In accordance with an example embodiment, an automated taxonomy mapping system that uses sequence semantic embedding techniques is described. Sequence sematic embedding models are used to generate the sequence vectors. The sequence semantic embedding models are trained offline and can be shared across different systems having different taxonomies and various versions of a category taxonomy.

Growing community of inventors

San Jose, CA, United States of America

Mingkuan Liu