Decide which text metrics can be calculated on-the-fly and which can be calculated dynamically

[phHartl]

Currently, following metrics a considered to be static for specific corpora:

• Topic modeling (needs a lot of processing resources)
• Judgment classification (needs to be trained on whole corpus) - but can probably provide a model so it can be dynamically called in the end
• Intertextuality (but probably won't make it because that is super hard - network analysis also needs a lot of processing power) (- Sentiment analysis should work dynamically depending on approach used) In contrast, the following metrics can be called dynamically:
• Tokens (e.g. "The", "sentence", "is", "short" ".")
• Sentences (e.g. "The sentence is short.")
• Part of speech tags (e.g. "The" (DET), "sentence" (NOUN), "is" (VERB), "short" (ADJ) "." (PUNCT)) (https://universaldependencies.org/u/pos/)
• Lemmata (e.g. "sentences" -> "sentence")
• Document similarity (based on word embeddings - per-document basis)
• Named entities (aka. 'the court' etc.)
• N-grams (collocations) (e.g. "the court" (2/bi-gram) currently only per-document basis)
• Readability score (using https://en.wikipedia.org/wiki/Flesch%E2%80%93Kincaid_readability_tests#Flesch_reading_ease)
• Most frequent words Still in progress:
• Sentiment
• Keyword extraction
• Co-occurrences on corpus basis (probably tf-idf)
• Diachronic analysis ideas (which metrics should be measured over time?)
• Classification