Investigate feasibility of Interest model fitting

NEW
Assigned to

Status

4 years ago
3 years ago

People

(Reporter: mzhilyaev, Assigned: mruttley)

Tracking

(Blocks: 1 bug)

Firefox Tracking Flags

(Not tracked)

Details

(Whiteboard: .?)

(Reporter)

Description

4 years ago
Recall that rule application may result in multiple assignments of categories to page in history.  Since, rule produces probabilities for category(s), it's valid to consider an algorithm that trains a probabilistic model on all of user history so that category assignments by page classifier become cohesive with interest classification.

For example, a page classifier may assign [Politics: 0.5, Biology: 0.9] to a page, but since the user did not demonstrate any other evidences of interest in Biology, then Politics should be chosen over Biology.
(Reporter)

Updated

4 years ago
Whiteboard: .?
(Reporter)

Updated

4 years ago
Blocks: 1104322
Points: --- → 13

Updated

3 years ago
Assignee: nobody → mruttley
You need to log in before you can comment on or make changes to this bug.