I am pleased to announce some important improvements to the cloud search service which are releasing to coincide with our Fall Cloud release but are applicable for all cloud search customers.

 

The search team has a set of improvements in the hopper and they are being deployed in phases. This first phase is releasing this weekend and includes:

 

Structured outcomes

 

When a content object is associated with an outcome, we will update the rank of the object accordingly. Objects marked as "outdated" will be ranked lower, and objects marked with other outcomes will be ranked higher -- among which objects marked as "official" and "final" will get the strongest boost. Leveraging these outcomes will make it much easier for users to find content that the company cares about most, and it is one of the most requested features by customers around search. Of course, this improvement only applies to our cloud customers or to those who upgrade to Jive 7 to get the structured outcome functionality.

 

Recency boosting

 

As content ages (as determined by last modification date), we will reduce its relevance score. Since it is a very new feature, the decay function is relatively moderate. This is another most requested features by customers around search.

 

Tokenizing tags by underscores

 

When a tag contains multiple terms, the terms are usually concatenated by underscores. However, we didn't tokenize tags by underscore (e.g., searching for term "search" wouldn't match tag "cloud_search"), and so tags were not very useful in search relevance. This should make tagging much more effective and help with search relevance!

 

Diacritics

 

Accents are removed during search, e.g., searching for "protege" will get protégé search test . For now, we only applied it on English content. We will be investigating other languages.

 

Thai Analyzer

 

Thai becomes an official supported language. We have full support on Thai now!

 

Setting ranking factors per content type

 

Different content types have different characteristics. Blogs can't be marked as "final". Status updates decay fast. Polls are very short. Using a universal algorithm + configuration for all content types is apparently less ideal. Starting from this update, we will treat each individual content type somewhat differently. The difference is minor now, but we expect to have more and more special treatment per content type as it makes sense.

 

That's the list for phase 1 - look for updates on phase 2 in the coming months!