Suggestion - How to improve OSE metrics for DA & PA

[James77]

I am sure everyone is aware at Moz, that although the Moz link metrics ( primarily I am talking about DA & PA) are good, there is a lot of room for improvement, and that there are a lot of areas where the metric values given to some types of site are well out of whack with what their "real" values should be.

Some examples
www.somuch.com  (Link Directory) - DA 72
www.articlesbase.com (Article Directory) - DA 89
www.ezinearticles.com (Article Directory) - DA 91

I'm sure everyone would agree that links from these domains are not as powerful (if of any value at all), as their DA would suggest, and therefore by definition of how moz metrics work, the sites these have links from such sites are also inflated - thus they throw the whole link graph out of whack.

I have 2 suggestions which could be used to singularly or in conjunction (and obviously with other factors that Moz use to calculate DA and PA) which could help move these values to what they should more realistically be.

1/. Incorporate rank values.
This is effectively using rank values to reverse engine what google (or other engines) as a "value" on a website. This could be achieved (if moz were not to build the data gathering system itself), by intergrating with a company that already provides this data - eg searchmetrics, semrush etc. As an example you would take a domian and pull in some rank values eg http://www.semrush.com/info/somuch.com?db=us  - where you could use traffic, traffic price, traffic history as a metric as part of the overall Moz scoring alogrithm. As you can see from my example according to SEMRush the amount of traffic and traffic price is extreamly low for what you would expect of a website that has a DA of 72.  Likewise you will find this for the other two sites and similarly to pretty much any other site you will test. This is essentially because your tapping into Googles own ranking factors, and thereby more inline with what real values (according to Google) are with respect to the quality of a website. Therefore if you were to incorporate these values, I believe you could improve the Moz metrics.

2/. Social Sharing Value
Another strong indicator of quality the amount of social sharing of a document or website as a whole, and again you will find as with my examples, that pages on these sites have low social metrics in comparison to what you would normally associate with sites of these DA values. Obviously to do this you would need to pull social metrics of all the pages in your link DB. Or if this we to tech intense to achieve, again work with a partner such as searchmetrics, which provide "Total Social Interations" on a domain level basis. Divide this value by the number of Moz crawled pages and you would have a crude value of the overall average social scorability of a webpage on a given site.

Obviously both the above, do have their flaws if you looked at them in complete isolation, however in combination they could provide a robust metric to use in any alogrithm, and in combination with current moz values used in the alogrithm I believe you could make big strides into improving overall Moz metrics.

[Dr-Pete]

Thanks - happy to pass that along. We're actually in the middle of a long-term spam detection project to help notify people when a site seems to be suspicious or is likely to be penalized by Google. Eventually, this may find its way into DA/PA. We don't want to use ranking and Google's own numbers, as it creates a bit of a problematic data dependency for us (especially long-term).

[James77]

Cheers Pete.

I totally understand the data dependency. One thing you could do, which would not require data dependency (long term), and also help with the spam detection your building is to take a single snapshot of "Ranking" - then use this as a data set to pattern match spam sites. EG if you managed to pull say 100,000's of ranking scores (say traffic scores from SEMRush), then match that with Moz's current scoring on that domain, then bucket the sites into groups that have higher or lower ranking scores than DA would predict, then try and reverse engineer the link or other patterns Moz use which are common to those buckets.

[Dr-Pete]

I'm not directly involved in the project, but I think that's actually part of what they're doing - using Google de-indexation and obvious penalties to train the system, but trying to avoid a system that would have to go look up the site on Google every time it needed to make a prediction.