For the relevance, I would argue that sentiment analysis could be one (but not the only one) factor.

Have a look at http://www.scirp.org/journal/PaperInformation.aspx?paperID=30811 , which deals with sentiment analysis and Thanksgiving (the event).

Also, ontlogies may support: http://www.academia.edu/5246336/Ontology-based_sentiment_analysis_of_twitter_posts

Refer also to the book "Social Media Modeling and Computing", where chapt 3.2 deals with novelty.

Technique behind is text mining.

Have a look on WEKA (Newseeland) or https://semantria.com/

For the relevance, I would argue that sentiment analysis could be one (but not the only one) factor.

Have a look at http://www.scirp.org/journal/PaperInformation.aspx?paperID=30811 , which deals with sentiment analysis and Thanksgiving (the event).

Also, ontlogies may support: http://www.academia.edu/5246336/Ontology-based_sentiment_analysis_of_twitter_posts

Refer also to the book "Social Media Modeling and Computing", where chapt 3.2 deals with novelty.

Technique behind is text mining.

Have a look on WEKA (Newseeland) or https://semantria.com/

This is clearly a case of Social Network Analysis and modelling. Research more in that direction and get ur answers. Granovetter and Valente are quite cast authors in this regard.

I hope Tinfy (Text Inhomogeneities Finder) can help you. The program should mark the novel tweets by the green colour.

milicka.cz/en/tinfi

It is based on the type-token ratio analysis. Tweets may be too short to apply this method, but I think it is worth trying.

Thanks for all the replies.

I'm exploring whether I can use a Deep Learning approach to predict relevance in an unsupervised way.

For the novelty, I think a possibility is the promorank framework (https://www.researchgate.net/publication/237532476_Promotion_Analysis_in_MultiDimensional_Space). You can set a small subspace and add the temporal info.

Article Promotion Analysis in MultiDimensional Space

For practical solutions' base take a look at NLTK

http://www.nltk.org/howto/semantics.html