Abstract: The technology relates to methods for detecting and classifying emotions in textual communication, and using this information to suggest graphical indicia such as emoji, stickers or GIFs to a user. Two main types of models are fully supervised models and few-shot models. In addition to fully supervised and few-shot models, other types of models focusing on the back-end (server) side or client (on-device) side may also be employed. Server-side models are larger-scale models that can enable higher degrees of accuracy, such as for use cases where models can be hosted on cloud servers where computational and storage resources are relatively abundant. On-device models are smaller-scale models, which enable use on resource-constrained devices such as mobile phones, smart watches or other wearables (e.g., head mounted displays), in-home devices, embedded devices, etc.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, re-ranking resources for categorical queries. In one aspect, a method includes receiving queries, and for each received query: receiving data indicating resources identified by a search operation as being responsive to the query and ranked according to a first order, each resource having corresponding search score by which the resources are ranked in responsiveness to the query and determining whether a proper subset meets a quality condition based on a quality measure that is indicative of the quality of the resources in the proper subset and independent of search scores of the resources for received query. For each query for which the proper subset meets the quality condition, determining a quality score for each resource in the proper subset and re-ranking the resources in the proper subset according to their respective quality scores.

Abstract: An analytics engine for determining analytic relationships in data queries based on responsive data sets includes a memory for storing data and a processor in communication with the memory. The processor is configured to identify a data query for analysis from a query repository, retrieve a plurality of interaction data associated with the data query, wherein the interaction data represents interactions between a plurality of user systems and a query result previously generated based on the data query, wherein the query result includes a plurality of links, identify a link selection count for each of the plurality of links based on the plurality of interaction data, classify the data query as one of a content targeting query and a data-creator targeting query based upon the plurality of link selection counts, and generate a query characteristic analysis based upon the classified data query and the plurality of link selection counts.

Abstract: Systems, methods, and other embodiments associated with document relevancy are described. One example method includes receiving one or more query terms from a query on stored documents. A relevancy operator is run on a document clump that applies more than one type of matching operation between the query terms and the document clump in a single pass. The relevancy operator may also apply at least one heuristic on the document clump in a single pass. A document's relevancy to the query is predicted based, at least in part, on an output of the relevancy operator.