|
Signal Quality
The quality of expertise identification is entirely reliant on the quality of signals the host platform can provide the Xperscore engine. While implicit feedback, such as views, likes, or flags are sufficient, significant improvements in quality can be seen when enabling more explicit signals such as category subscriptions, ratings, societies, and other social elements. |
|
Views
Treated as a secondary signal, notifying the engine that a user has viewed certain content allows assumptions to be validated, and behavioural correlations to be reinforced. While often as straightforward as whether a user has loaded a page, views can become a significant source of intelligence when the signal reflects item by item real-time loading, or a pause in scrolling while the content is being shown. |
|
|
Likes
Designed as the primary signal for lower resolution platforms, a like is a clear indicator of a user’s opinion of a content item. Care is taken to extract maximum intelligence from this semi-binary signal. Often, the implication is that nearby content is disliked, or can be assumed to be at least inferior to the promoted item. Coupled with ‘view’ signals, probabilities of dislikes versus non-signals can be extracted. |
|
Flags
Users flagging content as spam, inappropriate, or simply false, can reveal more about a user’s involvement with the topic, and particularly their mental engagement. Users don’t often flag content as false in areas they are impartial to; therefore declaring another user wrong typically implies some degree of mastery of the topic, from the user’s point of view. |
|
|
Ratings
The ideal explicit signal of user preference, Xperscore treats a Rating as a definitive data point. Prior beliefs are adjusted while the effect of a solid signal ripple outward from the content involved. The degree to which a rating is congruent with the ratings of other users for the same item, weighted by their estimated expertise on the topic, often implies strong intelligence on whether the given user is genuinely an expert. Exceptions abound! If the content is highly controversial, a rating may not be reflective of the target’s perceived expertise, but instead a staunch disagreement with their expressed position. Filtering on certain metrics can assist in preventing these misleading signals from propagating far beyond the initial content item. |
|
Societies
If the host platform includes a concept of subscribing to a category, or joining a society related to a category, this contributes to the internal measure of user interest in a topic. This does not, in itself, affect their expertise, but rather is helpful in informing weightings of other signals, as well as preventing saturation by ‘over-involved’ members of a given category. |
|
|
Social Elements
Developed as a result of the research and development performed by akatoo in a Q&A environment, Xperscore is trailing signals describing whether two users are officially ‘connected’ in certain contexts. Incorporating this new stream of data enables the engine to balance against skew caused by friends reinforcing each others’ ratings. |
