The Altmetric Attention Score is influenced by two factors:
The quantity of posts mentioning an output
The quality of the post's source
The quantity is relatively straightforward: the more posts mentioning an output the higher its attention score. We measure quality in a few different ways. In general:
Higher profile posts are worth more than lower profile ones. An article in the Washington Post contributes more, in score terms, than a blog post. A blog post contributes more than a tweet.
Who authored each post is important. For posts on social media sites we typically fetch an author's list of followers, a list of their past posts and information about how often those posts were liked, retweeted or reshared. A tweet from a doctor followed by other doctors will contribute more than an automated tweet from a journal's press office.
A more detailed explanation of how the scoring algorithm works can be found here.
Important things to remember:
Altmetric measures attention, not quality. People pay attention to papers for all sorts of reasons, not all of them positive.
Altmetric only tracks public attention. Papers are discussed in private forums, offline in journal clubs and by email but we cannot track this.
Altmetric tracks direct attention, that is to say attention focused on a specific research paper or dataset. More specifically for a newspaper article or blog post etc. to be counted by Altmetric it must either contain a link to the publication (journal article, DOI, PMID, or institutional repository) or reach our text mining criteria. We have more information here about how we do English-language text mining for news stories and policy documents.
Altmetric provides you with a single metric per output so that you can quickly compare relative levels of attention but it only makes sense to use this when comparing apples with apples (e.g. within a single discipline). The norms for attention are very different for different scientific disciplines, just as the norms for citations are.