Sorry, my library seems a bit outdated on that side.

But the one from Wikipedia looks great at first sight. Ralph., Kimball (2004). The data warehouse ETL toolkit : practical techniques for extracting, cleaning, conforming, and delivering data

Seems that you are more interested on the crawling and ETL side.

Maybe you should look more into Data warehouse or Data lake literatur. Especially the shift in paradigm from ETL (extract, transform, load) to ELT (extract, load, transform) respectively schema-on-read.

Depends a bit on your skill level and what you want to achieve.

I started with the Introduction to Information Retrieval (2008) book, which was quite math-heavy back then. But I learned a lot and found it a good starting point.

You get the concept of decompounding, reverse index, ranking functions, etc.

Newer IR strategies involve word2vec methods for item representation instead of handcrafted ones or directly learning the search ranking function, which is a different beast compared to traditional search engines.

Thanks - I think I got it. Kind of new to me why language models use perplexity instead of log-likelihood which is a monotonic function of perplexity.

From Wikipedia it seems that perplexity is in unit "words" instead of "nats/bits", which might be more interpretable.

Are there other advantages I overlook?

Thanks! Hm, seems to be a measure of sharpness for the predicted words?

Yes, I know the concept, but where's the connection to the pruning approach here?

Sorry - What's meant by perplexity here?