I’ve gotten a few questions about how people results are ranked in SharePoint/SharePoint On-line so I thought I would give a brief overview of two of the people ranking models available and how they work. I might be off on some points as decoding the models is not all that trivial as they have been created using machine learning and they also have multiple levels and use neural networks. If that sounded Greek or geek to you, then join the club :)
I’ll present my simplified interpretation and you can take it at face value.
For each model I will list the managed properties and associated user profile properties used in the rank profile, and list them in ranking importance from highest to lowest with my relative weights to put it into perspective.
User profile properties not listed in the tables will not influence how results are ranked.
The default model – People Search application ranking model
Id: d9bfb1a1-9036-4627-83b2-bbd9983ac8a1
| Managed Property | User Profile Property | Mikael's relative weight |
| JobTitle | Job title | 20 |
| RankingWeightHigh | Department | 20 |
| Responsibilities | Ask Me About | 15 |
| PreferredName | Name | 10 |
| RankingWeightName | User name Work email + generated permutations of first name and last name | 5 |
| Memberships | Quick links | 4 |
| RankingWeightLow | Skills Interests | 3 |
| ContentsHidden | Office Location Office Past projects | 2 |
| Pronunciations | First name Last name Name | 0.5 |
In addition to this there’s a ranking feature using the managed property LevelsToTop. This ranking feature counts more than the textual matching, and effectively means that if you have linked up everyone in your organization in a hierarchy with whom everyone reports to, people higher up in the chain will get priority ranking.
People Search social distance model
Id: 5df7ba10-55d6-4da1-b55f-896f7bcb486b
| Managed Property | User Profile Property | Mikael's relative weight |
| PreferredName | Name | 10 |
| CombinedName | generated permutations of first name and last name | 5 |
| AccountName | Account name | 1 |
| UserName | User name | 1 |
| WorkEmail | Work email | 1 |
| SipAddress | SIP Address | 1 |
| Pronunciations | First name Last name Name | 0.5 |
LevelsToTop also counts in this rank profile, but what counts a lot more, is how close to you people are in the organizational chart. There are two managed properties called FirstLevelColleagues and SecondLevelColleagues which trumps everything in this profile. These are people reporting the same manager as you, and one level beyond that as well. If those properties are not populated, and I have seen this in environments, then social distance will not work.
