Influence and Context in the Grapevine
The Grapevine calculates User Influence, one layer at a time.
The Grapevine enables you and your community to identify who is the most trustworthy, and in what context, so they can help you curate content, facts, and information.
To accomplish the above goal, users issue Trust Attestations which are then processed by the Grapevine into Influence Scores. Trust Attestations and Influence Scores are contextual, with Context defined using two dimensions: the action dimension and the category dimension. In this post, we review what this means and how it works in some detail.
Let us start with an example. Suppose Alice knows that Bob is an aficionado of all kinds of movies and typically makes good recommendations of what to watch. The corresponding Trust Attestation might be: Alice endorses Bob as being trustworthy to create recommendations (the action) relating to the topic of movies (the category). In this example, the Context would be the action: to create recommendations, and the category: movies. What about users whom Alice does not know directly, but are known by her community? Her Grapevine will gather all Trust Attestations like the one above, combine them to calculate a weighted average, and produce a Trust Score for each of those users. When Alice is deciding what movie she wants to watch some day, her Grapevine takes such Trust Scores into account to know whose movie ratings to listen to and whose to ignore.
When Alice creates any given Trust Attestation, she must select the appropriate Context. To do this, she selects from a list of actions and a list of categories, with each list maintained and organized into hierarchies by Alice and her Grapevine. If comedies, dramas, and westerns are subcategories under movies, then Alice’s attestation effectively endorses Bob for each of those subcategories (and their subcategories, and so on), without the need to create independent attestations for each and every subcategory. For this reason, in the early days of the Grapevine, users are going to focus on broad-Context Trust Attestations, with the most broad taking the form: Alice endorses Bob for all actions, in all categories. After the Grapevine has been sufficiently fleshed out at the base layer, users will begin the process of making attestations that are more specific.
Suppose, for example, that Alice thinks Bob to have excellent taste in all types of movies, with the sole exception of comedies, where for whatever reason he has poor taste. She could create an additional attestation to that effect, in which case her more specific attestation in the child category of comedies would override the more general attestation in the parent category of movies. But if she thinks he has great taste in all types of movies, there’s no need for her to create separate attestations for each sub-category. One attestation in the parent category is sufficient.
Some Contexts are “adjacent” to one another, with neither being the parent or the child of the other. For example, Alice’s endorsement of Bob’s ability to create recommendations about movies does not imply anything about his ability to curate Wikifreedia articles about movies, to write the screenplay for a movie, to produce a movie, etc. None of these layers would inherit trust regarding any of the others, and so each of these actions would require an independent attestation.
Note that all of this is pretty much the way our brains would handle similar information in real life.
As stated above, the most generic of all Trust Attestations would be for Alice to trust Bob in all actions and in all categories. The most generic Influence Score tells Alice how much trust to place in any given user in all actions and in all categories. In the absence of more specific information, her Grapevine will always fall back upon more generic Influence Scores, ultimately to this most generic score. The Influence Score can therefore be said to exist in layers, with the Influence Score in all actions and in all categories being the base layer of trust for the Grapevine, and more specific scores being layered on top of more generic layers.
Summary
The Trust Score is the Grapevine’s recommendation to Alice of how much trust to place in Bob for any given action and in any given category. Each Context defines a unique own trust layer, with broadly applicable, generic layers on the bottom and specific, niche layers on top. As the Grapevine ecosystem gains users and gets built out over time, we should expect the foundational layers to form first, followed by progressively higher layers that are as specific and niche as required to serve the needs of its users and the community.