The trending page needs to get fixed finally

[steemchiller]

All botted and overvalued posts must be excluded completely or at least downrated substantially.
Discussion is welcome :)

[only-dev-time]

Right! But it is an issue of hivemind. The Condenser receives the posts for the trending page from Hivemind.
I am currently testing the extension of the score calculation with interaction components.

My approach is this:

• There is a combination with the previous score and the new interactions score.
• The interactions score takes into account the number of reblogs and the number of comments.
• littel Votes weight,
• higher Interactions weight,
• the interaction components contains the reblog weight and comment weight.

I try out different factors to get a better balance.

The inclusion of unique comments would be ideal. However, without changes to the database schema one or more database queries would be necessary.

[blazeapps007]

Would be interesting to USE AI to filter posts before it gets Delivered to the frontend from hivemind, like a Middleware, that can categorize content based on Different factors like tags / body / title / upvotes.

[steemchiller]

@only-dev-time Okay, sounds good so far. I think we should keep the old trending algorithm available in the API and add a method for retrieving the posts according to the new logic. In condenser we then could switch to the new method by default and maybe make it configurable later. Changes to the database will highly likely be required to make it run performantly.

[only-dev-time]

@steemchiller

I think we should keep the old trending algorithm available in the API and add a method for retrieving the posts according to the new logic.

This is difficult because the scores are stored in the hivemind database and the condenser receives the posts sorted by scores. There are no score calculations during retrieval. The score is recalculated with every post action (vote, resteem) and stored in the database.
The scores for trending and hot only exist once in the database. Either with the old or the new logic. If the Condenser should be able to request two different score methods, we would have to save two variants in the database.

@blazeapps007
In order for an AI to analyse the posts for the trending page, we would have to specify which posts should be returned as a basis. I would therefore assume that hivemind would have to return (at least) all posts from the last 7 days, which the AI would then have to analyse. I don't currently think that this can be implemented efficiently.

[blazeapps007]

Yes, that would be a challenge, but it would to fun to take it

[steemchiller]

@only-dev-time

If the Condenser should be able to request two different score methods, we would have to save two variants in the database.

I see. Yes, I did the same for SDS. I have an extra table there for active posts though. One score field for each trending type would be needed. Maybe we skip that for now in Hivemind and just use the existing fields then.