This document is intended to illustrate key GL sections in Autocomplete Relevance rating by way of practical examples.
Query: globus bauma
Viewport: Stale
User location: 51.793428, 9.244054
The user issued the query in a rural environment. There are no possible suggestions in the immediate vicinity of the user. Consequently, we are more lenient in terms of distance than in urban environments with numerous possible suggestions closer by.
As we can see, there are three Globus Baumarkt locations roughly at the same . . . .
distance from the user, all within a range of 55 km plus/minus a few kilometers. In
this scenario, we do not want to demote any of them for distance; they are Excellent.
We can also see a distribution pattern that lets us define the Good and Acceptable zones. The Good zone includes the locations in Goslar, Salzgitter, and Peine to the NE of the user. The stores in Unna and Bergkamen to the SW, as well as the branch in Braunschweig to the NE are in the transition zone between Good and Acceptable. The store in Wolfsburg is clearly in the Acceptable zone.
Since there are no fixed distances or number of results in distance-driven rating, it can be helpful to visualize such patterns in these scenarios and group possible suggestions into such clusters.
In contrast to rural environments with few possible suggestions near the location intent, we are less lenient in terms of distance in urban environments with many possible suggestions nearby.
The user is looking for supermarkets near them. The BILLA store on Conrad-von- Hötzendorf-Straße 2 is clearly closer than any other possible suggestion. In an urban environment, our Excellent zone can be fairly small depending on the respective scenario, and taking into account the number and distribution of possible suggestions.
Just as in the previous example, we can now establish our Good and Acceptable zones by grouping together possible suggestions at similar distances from the user.
Query: seftigenstrasse
Viewport: Fresh, over Bern
User location: 46.940108, 7.450049 inside viewport
Suggestion: Kindergarten Seftigenstrasse, Seftigenstrasse 14, Bern, Schweiz
Rating: Good, Acceptable
According to GL 4.1.1. General Match, the query string can directly match the suggestion or it can match the most significant part of the suggestion.
Let’s have a look at the Target San Diego example. The GL explain that the store is no match for the query [san die] because the use of “San Diego” in the business name is extra information and not a part of the official branding for this business chain. And there we have the difference to our example, where the street name is part of the official branding. Consequently, we do not rate it Bad for no match, but instead demote it for user intent to Acceptable or Good. In other words, we do not consider the suggestion a random POI in such AC scenarios.
Query: Oberperfuss
Viewport: Stale
User location: 46.940108, 7.450049 inside viewport
Suggestion: Bergbahnen Oberperfuss, Peter-Anich-Weg 11, Oberperfuss, Österreich
Rating: Good, Acceptable
Here as well, the location modifier is part of the official POI branding. In addition, the suggestion is a transit suggestion for the locality.
Query: Potsdam
Viewport: Missing
User location: 47.312711, 12.794269
Suggestion: Schloss Sansoucci, Maulbeerallee, Potsdam, Deutschland
Rating: Good
The prominence of a feature refers to its popularity, including the number of people visiting and media sources referencing it. […] Suggestions with high prominence may still be relevant even when far away from the user, GL 4.3.3. Prominence.
The user issued the query in Zell am See in Austria. Our suggestion does not directly match the query. However, the suggestion is a highly prominent and internationally known POI associated with Potsdam, Germany, which is a likely intent for the query string. While not a direct match, this can be considered a prominence match.
It might not be immediately clear whether a suggestion is internationally prominent. If you are not familiar with the respective POI, the number of reviews for it are a good indicator for prominence. Schloss Sansoucci, for example, has more than 21000 reviews on Google. There may also be other indicators, such as status as UNESCO World Heritage, for example. Thus, if a suggestion does not match the query at first glance, conduct research to see if it might qualify as prominence match.
Query: hagenweg 22 5
Viewport: Stale
User location: 52.025783, 8.537711
Suggestion: Hagenweg 22, Coppenbrügge, Deutschland
Rating: Bad
The suggestion does not account for all of the information contained in the query string. The user appears to be looking for a street address in a postal code area starting with 5. The suggestion is for a street address in 31863 Coppenbrügge, which does not match the requested postal code area. To accurately identify potential user intents, it’s important to include all of the information included in the query string when rating Relevance.
Query and category suggestions are rated by considering the user intent and the possible suggestions within the viewport. For a query/category suggestion to be valid, it needs to provide some use to the user. Simply matching the user query is not enough on its own. (Source: GL 4.13. Query and Category Suggestions)
Allans Ann
Query: Apo
Viewport: Fresh, over Basel, Schweiz
User location: 47.55627, 7.58102 inside viewport
Suggestion: Aponeo
Rating: Acceptable, Bad
For the incomplete query, there are possible, likely suggestions, such as Apotheke, Apostolische Kirche, etc. Our suggestion yields only few matches far away from the user. At best, this suggestion is Acceptable since it’s neither a widespread search category, nor is it helpful to the user.
Relevance is always rated independently of any data issues the suggestion may have. This includes suggestions for addresses that do not exist in real life.
When assessing the distance for valid, existing address suggestions, they should be evaluated only in relation to other existing addresses, including any potential unreturned valid addresses. Non-existent address suggestions should be disregarded, even if closer to the user/VP location.
However, when assessing the distance for non-existent address suggestions, it should be evaluated only in relation to valid address suggestions, including any potential unreturned valid addresses. (Source: GL 8.1.1. Address does not Exist)
Query: Roggenstraße 49
Viewport: Fresh, within Duisburg, Deutschland
User location: 51.353108, 6.568204 outside viewport
Suggestion 1: Roggenstraße 49, Oberhausen, Deutschland
Suggestion 2: Roggenstraße 49, Essen, Deutschland
Suggestion 3: Roggenstraße 49, Weeze, Deutschland
Suggestion 4: Roggenstraße 49, Goch, Deutschland
Suggestion 1 rating: Excellent
Suggestion 2 rating: Excellent
Suggestion 3 rating: Excellent
Suggestion 4 rating: Excellent
Suggestion 1 is the closest suggestion; however, research shows that it does notexist. We still rate Relevance as if the address actually existed. Since it is closer than the nearest possible valid address, it’s rated Excellent.
Suggestion 2 is second-closest to the viewport, but it doesn’t exist, as well. When rating Relevance, we disregard other addresses that do not exist, even if they are closer to the location intent. Distance demotions in this scenario are solely based on existing addresses. Because this address is also closer than the nearest possible valid
address, it’s rated Excellent, as well.
The same considerations apply to suggestion 3. The address does not exist, but is closer to the location intent than the nearest possible valid address, which makes it Excellent.
Suggestion 4 is for the nearest possible valid address in the real world. Consequently, it’s Excellent.