Content Discovery KPIs #2239

New issue

Open

opened 2019-01-29 02:14:30 +01:00 by tiger5226 · 0 comments

tiger5226 commented

2019-01-29 02:14:30 +01:00

(Migrated from github.com)

In order to train the AI we need feedback from users as well as to collect our KPIs:

Ratings: We want to measure the quality based on ratings. If we are providing interesting content the ratings ratio will show it.
Recommendation Clicks: The % clicked will tell us if we are providing good recommendations. As they improve the click % will increase.
Tagged Content Clicks: The % clicked will tell us if the tagged content is correctly tagged since this is based directly on what the user said they were interested in. Also it will show improvement as we refine the tags.

Below are some key items for this. These are data requests, how the UI is handled I will leave to the app team and others:

Ratings - I need a scaled rating from users. Ideally at least 1,2,3,4,5. This would call an API to post the rating, and a call to an api to grab the previous rating for a specific user who already rated( not editable at first).
Suggested Content - We need a way to present suggested content based on a specific user. This could be an addition to the homepage for a user. The homepage URL does not get called with a specific user in mind, so it will be a different authed API call. We can show 10 pieces of content.
Related Content - We can change the API for related content so we grab it from the AI to recommend content based on a similar piece of content. Maybe split the content list to be half search based results and half AI based results. The AI will return content based on "if you liked that, you would like this".
Recommendation Click - An API will be made available to record recommendation clicks. The recommendations come with a recommendation_id, when it is clicked and the page is viewed that counts as a click.
Recommendation View - The view API will be adjusted to have the recommendation_id passed as a parameter so we can notify the AI for train.

Tagged Content

Tagged Content View - The desktop app needs to present a user interface for selecting interesting tags for the user to choose from. An API will be made available for retrieving the list of tags to choose from. The user should be able to Add/Remove the tags they are interested in. Seem image below

Tagged Content Discovery - This should be added to the user specific suggested content page/content.

These are the items important for feedback to train our AI, in this order.

In order to [train the AI](https://github.com/lbryio/internal-apis/issues/674#issuecomment-459520806) we need feedback from users as well as to collect our KPIs: - **Ratings:** We want to measure the quality based on ratings. If we are providing interesting content the ratings ratio will show it. - **Recommendation Clicks:** The % clicked will tell us if we are providing good recommendations. As they improve the click % will increase. - **Tagged Content Clicks:** The % clicked will tell us if the tagged content is correctly tagged since this is based directly on what the user said they were interested in. Also it will show improvement as we refine the tags. Below are some key items for this. These are data requests, how the UI is handled I will leave to the app team and others: ## KPI Related ## - [ ] Ratings - I need a scaled rating from users. Ideally at least 1,2,3,4,5. This would call an API to post the rating, and a call to an api to grab the previous rating for a specific user who already rated( not editable at first). - [ ] Suggested Content - We need a way to present suggested content based on a specific user. This could be an addition to the homepage for a user. The homepage URL does not get called with a specific user in mind, so it will be a different authed API call. We can show 10 pieces of content. - [ ] Related Content - We can change the API for related content so we grab it from the AI to recommend content based on a similar piece of content. Maybe split the content list to be half search based results and half AI based results. The AI will return content based on "if you liked that, you would like this". - [ ] Recommendation Click - An API will be made available to record recommendation clicks. The recommendations come with a `recommendation_id`, when it is clicked and the page is viewed that counts as a click. - [ ] Recommendation View - The view API will be adjusted to have the `recommendation_id` passed as a parameter so we can notify the AI for train. ## Tagged Content ## - [ ] Tagged Content View - The desktop app needs to present a user interface for selecting interesting tags for the user to choose from. An API will be made available for retrieving the list of tags to choose from. The user should be able to Add/Remove the tags they are interested in. Seem image below ![image from ios](https://user-images.githubusercontent.com/3402064/53104087-c13e4f80-34fc-11e9-8168-abd285a3c16f.jpg) - [ ] Tagged Content Discovery - This should be added to the user specific suggested content page/content. These are the items important for feedback to train our AI, in this order.