Abstract: A search results user interface enables a user conducting a search of an electronic catalog to identify and access browse nodes (item categories) associated with the search query when the search query is determined to be broad. In one embodiment, if the catalog items that are responsive to the search query fall within more than a threshold number of browse nodes, a search results page is presented that provides functionality for the user to view the browse nodes associated with the search query and to navigate to associated browse node pages.

Abstract: Devices and techniques are generally described for generating personalized search query suggestions. In some examples, a first search query entered by a user into a search field may be determined. A first keyword associated with the first search query may be determined based at least in part on past interactions of the user with content. In various examples, first string data comprising at least a portion of the first search query and the first keyword may be generated. A data store may be searched using the first string data. A first suggested search string associated with the first string data may be determined. In some examples, the first suggested search string may be displayed on a display in association with the search field.

Abstract: Systems and methods are described for analyzing one or more online shopping missions to determine an estimated stage of a user within the mission and to adjust content for display to the user based on the estimated stage of the user within the mission. The estimated stage of the user within the mission may be determined using a model, such as a machine learning model, that accepts as input signals corresponding to the online shopping mission. The signals may represent user interface interactions and/or other data relating to the user and the mission.

Abstract: Systems and methods are provided for electronic catalog user experience improvements based on extracted item attributes. An example method includes obtaining information identifying items viewed during a user browsing session, the items being included in an electronic catalog, and the electronic catalog being organized according to a hierarchy, with the hierarchy comprising a plurality of categories. Information identifying attributes associated with the identified items is identified, and the attributes are ranked according to attribute relevance score. Items are determined for recommendation to the user based on the ranked attribute relevance scores, with the determined items being selected from items included in the electronic catalog which are also associated with a same category as the identified items.

Abstract: The present disclosure is directed to filtering co-occurrence data. In one embodiment, a machine learning model can be trained. An output of an intermediate structure of the machine learning model (e.g., an output of an internal layer of a neural network) can be used as a representation of an event. Similarities between representations of events can be determined and used to generate, augment, or modify co-occurrence data.

Abstract: The present disclosure is directed to dynamically determining the breadth of user intent reflected in a given search query. For example, historical user behaviors are used to construct a mapping that links particular search queries (or their constituent keyword strings) to items that users interact with during a predetermined window of time after submitting the query. The number of categories or browse nodes associated with these items (or a most popular subset thereof) is used to determine whether the search intent is broad or narrow. For broad searches, the present disclosure causes output of a specialized user interface that facilitates browsing of items in the browse nodes or categories, in place of the typical search results user interface.

Abstract: Disclosed are various embodiments for classifying network sites into site categories. A network site is classified into at least one of a plurality of categories based at least in part on similarity. Query popularity, query competitiveness, and/or query importance may be used in determining similarity. The similarity is measured between a first plurality of search queries and a second plurality of search queries. The first plurality of search queries is one that produced first search results that include the network site. The second plurality of search queries is one that produced second search results that include other network sites. Each of the other network sites has a respective category. The query-similarity based scoring may be combined with more scoring based on crawling and processing network page or network site contents.