Disclaimer: I haven't still used the domain datasets as they are fairly new.
I believe you can do both parts of your questions by using the VIDEO_ON_DEMAND datasets of Amazon Personalize.
For the first part of the problem, the VIDEO_ON_DEMAND dataset gives you the option to add metadata such as the CONTENT_OWNER.  Even if Personalize wouldn't know who you are following, it will know you are interacting with content from a user, and that should have a similar effect.
For the second part, handling the cold start problem, you can use incremental uploads, and Personalize will retrain the model every two hours . If you also use events for tracking user interactions, recommendations for new users will get relevant faster. 
 https://docs.aws.amazon.com/personalize/latest/dg/VIDEO-ON-DEMAND-datasets-and-schemas.html  https://docs.aws.amazon.com/personalize/latest/dg/importing-domain-dsg-data.html#incremental-import-domain-dsg  https://docs.aws.amazon.com/personalize/latest/dg/recording-events.html
Amazon Personalize- Exploration weight impact on recommendationsasked 3 months ago
How do we give recommendations when users create/post content? Like in YouTube, TikTok etcasked a year ago
Is there any way to download the recommended lists of users as bulk when using AWS-Personalize (specifically using user-personalization)asked a year ago
Can AWS Personalize be optimized/segmented for cachable recommendations?asked 9 months ago
Personalize Recommendations for Anonymous Users using SessionIdasked 3 months ago
How do you give negative feedback to amazon personalize? i.e. tell personalize that a user doesn't like a certain item?
AWS personalize get-recommendations bugasked 6 months ago
Amazon Personalize Data RequirementsAccepted Answerasked 2 years ago
How much interactions data do you need for decent/good recommendations in amazon personalize? Any suggestions?
Personalize filters - Users metadataasked 4 months ago