Location-based recommendations using nearest neighbors in a locality sensitive hashing (LSH) index
Software for a website hosting short-text services creates an index of buckets for locality sensitive hashing (LSH). The software stores the index in an in-memory database of key-value pairs. The software creates, on a mobile device, a cache backed by the in-memory database. The software then uses a short text to create a query embedding. The software map the query embedding to corresponding buckets in the index and determines which of the corresponding buckets are nearest neighbors to the query embedding using a similarity measure. The software displays location types associated with each of the buckets that are nearest neighbors in a view in a graphical user interface(GUI) on the mobile device and receives a user selection as to one of the location types. Then the software displays the entities for the selected location type in a GUI view on the mobile device.
1. A method, comprising:
creating an index of a plurality of buckets for locality sensitive hashing (LSH), wherein a bucket of the plurality of buckets includes one or more word or phrase embeddings derived from a corpus of documents that describe entities associated with geographic locations;
mapping a query embedding to corresponding buckets in the index and determining which of the corresponding buckets are nearest neighbors to the query embedding using a similarity measure;
displaying one or more location types associated with one or more buckets that are nearest neighbors in a view in a graphical user interface (GUI) on a mobile device;
receiving a user selection of a location type of the one or more location types; and
displaying one or more entities associated with the location type in a GUI view on the mobile device.
2. The method of claim 1 , comprising ranking the one or more entities based on geographical proximity to the mobile device.
3. The method of claim 2 , comprising using the ranking to determine prominence when displaying the one or more entities.
4. The method of claim 2 , wherein the ranking is performed using a mapping app.
5. The method of claim 4 , wherein the mapping app uses at least one of a geo-location or a geo-position for the mobile device.
6. The method of claim 1 , wherein the one or more word or phrase embeddings are derived from the corpus using a continuous distribution model.
7. The method of claim 6 , wherein the continuous distribution model is at least one of a continuous bag-of-words model or a continuous skip-gram model.
8. The method of claim 1 , wherein a geographic location of the geographic locations is at least one of a geo-location or geo-position associated with the mobile device.
9. The method of claim 1 , wherein the similarity measure uses at least one of cosine similarity, city-block similarity, or Euclidian similarity.
10. One or more non-transitory computer-readable media persistently storing instructions that, when executed by a processor, perform operations comprising:
creating an index of a plurality of buckets, wherein a bucket of the plurality of buckets includes one or more word or phrase embeddings derived from a corpus of documents that describe entities associated with geographic locations;
map a query embedding to corresponding buckets in the index and determine which of the corresponding buckets are nearest neighbors to the query embedding using a similarity measure;
display one or more location types associated with one or more buckets that are nearest neighbors in a view in a graphical user interface (GUI) on a device;
receive a user selection of a location type of the one or more location types; and
display one or more entities associated with the location type in a GUI view on the device.
11. The non-transitory computer-readable media of claim 10 , the operations comprising ranking the one or more entities based on geographical proximity to the device.
12. The non-transitory computer-readable media of claim 11 , comprising using the ranking to determine prominence when displaying the one or more entities.
13. The non-transitory computer-readable media of claim 11 , wherein the ranking is performed using a mapping app.
14. The non-transitory computer-readable media of claim 13 , wherein the mapping app uses at least one of a geo-location or a geo-position for the device.
15. The non-transitory computer-readable media of claim 10 , wherein the one or more word or phrase embeddings are derived from the corpus using a continuous distribution model.
16. The non-transitory computer-readable media of claim 15 , wherein the continuous distribution model is a continuous bag-of-words model or a continuous skip-gram model.
17. The non-transitory computer-readable media of claim 10 , wherein a geographic location of the geographic locations is at least one of a geo-location or a geo-position associated with the device.
18. The non-transitory computer-readable media of claim 10 , wherein the similarity measure uses at least one of cosine similarity, city-block similarity, or Euclidian similarity.
19. A method, comprising:
creating an index of a plurality of buckets, wherein a bucket of the plurality of buckets includes one or more word or phrase embeddings derived from a corpus of documents that describe entities associated with geographic locations;
mapping a query embedding to corresponding buckets in the index and determining which of the corresponding buckets are nearest neighbors to the query embedding;
displaying one or more location types associated with one or more buckets that are nearest neighbors in a view in a graphical user interface (GUI) on a device;
receiving a user selection of a location type of the one or more location types; and
displaying one or more entities associated with the location type in a GUI view on the device.
20. The method of claim 19 , comprising ranking the one or more entities based on geographical proximity to the device.