Page MenuHomePhabricator

Determine how to use the data to provide recommendations
Closed, DeclinedPublic

Description

With the different sources of recommendations [1], there are many options for how to utilize them before providing recommendations to the client.

The actual method to merge the 2 recommendation sources is a topic we are discussing in these days and for the moment in not completely clear.

[1] where sources of information are as defined in T159227, pasted below for easier access:

  • Sources for the recommendations
    1. Article category
      • Needs a table with the mapping between the category name and the list of sections to recommend
      • <category, [sections], probability>
    2. Current list of sections
      • Use output of an association rules algorithm
      • Table with a list of sections mapped to a possible section to add
      • <[sections], [missing_section], probability>

Event Timeline

@schana what do you mean by "different sources of recommendations" in the current task description?

@schana what do you mean by "different sources of recommendations" in the current task description?

As described in the parent task, T159227

ok, thanks! I pasted it as part of the description.

Now my question is: do you have access to the data in item 1 and 2 under "Sources for the recommendation" captured in the description? If yes, does the question boil down to choosing one of (1) or (2)?

@leila Yes, we have access to both sets of data (see T159705 and T159706), but are working through performance issues for the 2nd set.

does the question boil down to choosing one of (1) or (2)?

I'm not sure what the best method for using the data to provide recommendations is. We'll probably start off with some article that needs expanding, and from there can use both sets in some way to determine what sections to recommend.

@schana I will take this to Bob and Tiziano to discuss and converge on which one to choose for this first iteration and get back to you.

@schana confirming that we are still working on the algorithm that should provide the input for your code.

Diego's been working on an alternative solution for section recommendations.