Search Results Ranking
HERE Search uses a machine learning system to rank search results, as do most major location-based search engines. The system "learns" from real-world queries randomly sampled from search logs. Human annotators look at the results provided by the system and rate them by quality. These ratings are fed back into the system so that it can learn from its mistakes and constantly improve over time, adapting to changing query types, new application releases and data updates, and the growing geographic distribution of users.
- The quality of the string match between the query and the result, including the amount of spell correction required
- The distance between the user and the result, e.g. actual distance, whether the result is in the same city as the user
- The category or type of the result, e.g. address, city, restaurant, hotel
- The popularity of the result, e.g. number of clicks, number of times it has been saved as a "favorite"
- The population for results such as cities or states
- Certain types of categories, such as restaurants or hotels, are ranked via a "recommendations-style" algorithm where measures of popularity or quality, such as number of stars or reviews, are taken into account.
- Other types of categories, such as gas stations, are ranked purely by distance.
- Chain stores that are recognised as such by the system are also ranked purely by distance. See the following section.
Chain ranking
A chain is a business with multiple locations that all provide the same or similar merchandise or services. Most chains are retail stores.
HERE Search handles many chain searches by checking whether the query consists of a predefined chain search string along with an optional location, such as "bank of america" or "bank of america manhattan". See Chain-like Query Terms for the list of predefined strings. The list covers many popular chain searches and consists of strings that unambiguously refer to chains for which search has high-quality coverage. It includes common misspellings ("banck of america") and aliases ("b of a"). When a query matches a term on the list, Search returns chain locations within 200km of the query center , sorted by distance.
- The query does not match a predefined chain search string.
- Search determines that a group of the search results are likely to belong to the same chain based on commonalities in their names, URLs, and other attributes.
- One of the results in that group is the top result.
Search then modifies the results to emphasize the chain locations by reranking the chain locations by distance. The ranks of any other results are unaffected. Search may also return more results than usual in order to allow more chain locations to be returned.
For automotive chains, search also recognizes phrases such as Ford Dealer and Ford Service. All of these phrases return dealerships or service locations identified by HERE as belonging to the given automotive chain. The complete list of recognized phrases is included within Chain-like Query Terms.
