Learn about the search ranking models that were updated in SharePoint Server 2013
If you use SharePoint Server 2013, the Cumulative Update package published in August contains an updated and expanded set of ranking models for search.
If you use SharePoint Online, you will have access to these ranking models automatically when the service is updated.
Summary of Changes
The TechNet article "Overview of search result ranking in SharePoint Server 2013" contains information on how search results are ordered via the use of ranking models. The article also lists the set of ranking models that are included in SharePoint Server 2013. In our continuing efforts to provide relevant results for search users, we have refreshed the set of General Purpose ranking models:
Default Search ModelWe’ve introduced a new Dynamic ranking feature that boosts search results with anchor text pointing to them that matches the complete query text. Along with a newly introduced boost for the Site result type, all other factors being equal, top-level sites are expected to be ranked higher than the other file types (for example, docx, pdf, etc.). In addition, we noticed a deficiency of the old default ranking model in environments with many documents in different languages. When the query language does not match the main language in the documents, the previous default ranking model boosted short documents for which the language couldn’t be detected. We have updated the language matching feature in the new Default Search Model to avoid this.
O15 Default Search Model For backward compatibility reasons, we retained a copy of the old Default Search Model. If you don’t explicitly specify a ranking model, the new default ranking model will automatically be used after the update. If you prefer to continue using the previous default ranking model, you can override the default behavior for either all or some queries to use the O15 Default Search Model.
Search Ranking Model with Two Linear StagesAs indicated in a previous blog post about changing the ranking of search results in SharePoint 2013, we expect that using one of the standard ranking models in combination with query rules will provide a good relevance experience in many scenarios. However, some scenarios will still require a custom ranking model. With a neural network for its second stage, we expect the Default Search Model to provide better ranking of results in a variety of environments. But, it is inherently difficult to customize. The newly introduced Search Ranking Model with Two Linear Stages contains a linear model for its second stage that contains the same set of features as the Default Search Model. Being a linear model (i.e., the rank score is a weighted sum of the ranking features), this model is a more amenable starting point for producing custom ranking models. Read Customizing ranking models to improve relevance in SharePoint 2013 on MSDN to learn more about the most important ranking features in the Default Search Model. That article also explains how to deploy a custom ranking model.
The documentation that we have published on MSDN and TechNet has been updated to provide the keen reader with sufficient information to interpret the intended effect of our updates. Along with gathering more information about the usage of our search product, feedback from our current users is a critical ingredient in the continual improving of our product.
Do leave us a comment on this blog if you have any additional questions, or indeed if you want to provide us with some encouraging feedback!
Contributor Vishwa Vinay is an Applied Researcher on the SharePoint Search team