While standard search-ranking approaches can rank results derived from a single corpus of data (e.g., the Internet or an intranet), such approaches cannot rank results derived from multiple disjoint corpora, e.g., the Internet and on-device data. This disclosure describes ranking techniques that dynamically incorporate scores for results from on-device and cloud sources to generate a composite, more accurate ranking of search results. Ranks of results from cloud-based data sources (e.g., the Internet, application stores, etc.) are compared with the ranks of on-device data sources (e.g., messages, on-device settings, application data, etc.) to generate a multi-part ranking across the two data sources. The techniques include mixing or combining results at different stages of the ranking pipeline. On-device results are generated and ranked locally with no transfer of data outside the device.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.