The greatest challenge for any search engine is in ensuring that the results shown to the end user are the best results available. Over time the methods by which results are ranked have evolved many times over.
Search Engines have ordered pages by the date they were crawled, alphabetically, based on site owner included meta tags, keyword density, implicit and explicit search actions by users...and the list goes on.
Some of the improvements made to result quality calculations have been about understanding the content itself but much of the effort has been spent trying to outsmart marketers and malicious actors who attempt to game the results and ensure their product/service/site is shown at the top of the rankings.
We believe The Trust Network can significantly improve the quality of results while decreasing the spamminess of results. How? We are so glad you asked!
Phoebe uses the aggregate individual user's implicit and explicit feedback on results quality accounts to attribute value to any given site regarding a topic. If we stopped here we'd be susceptible to manipulation just like other search engines...so we don't stop there, we add the trust network layer.
Phoebe provides each user (excluding anonymous users) with an authority score. This starts quite low and can be increased over time through several methods:
A trusted user explicitly trusts your user account.
You provide substantive contributions over time to Phoebe (explicitly and/or implicitly, with the former carrying significantly more weight than the latter).
You verify your identity in ways that ensure you are an actual individual making legitimate contributions.
You verify your credentials on a specific topic.
If we stop here we are still susceptible to manipulation, individuals and/or machines can open accounts and increase their authority by granting trust to one another. That is why we take things yet a step further:
When an individual account ("grantor") trusts another account ("grantee"), the grantor's account reputation becomes linked to the grantee's account reputation.
If the grantor account trusts a spammy grantee account the grantor will take a penalty.
The negative penalty increases exponentially - that is, initially it may only be a few points off the grantor's authority score but as the number of grantee accounts performing detrimental actions increases we will use a multiplier to rapidly deduct from the grantor's authority.
This linkage need not be only first level trusts. For example:
If JohnDoe (grantor) trusts JaneDoe (grantee) and JaneDoe (grantor) trusts IAmSpammy (grantee) those penalized will be: IAmSpammy (most penalized), JaneDoe (moderately penalized), JohnDoe (lightly penalized).
This doesn't make us impervious to manipulation but it will make manipulation much harder. In order to gain authority one has to provide value and build relationships. To build relationships others must trust you to an extent where they are willing to suffer a penalty if you behave poorly.
It is likely that malicious activity will still occur through account compromises, but using the trust network we should be able to quickly isolate the impact of such manipulation and (hopefully) restore account access to the appropriate individual. That is, when a highly trusted user begins performing or being linked to negative behavior we can lockdown the account temporarily and work with the account owner to restore secure access to their account.