Citations are generally not semantically important for text summarization, though they may be useful for weight analysis, relational values, identifying legal tests, etc. Citations are important when they are used as labels for legal tests (eg, "Grant test", "Stinchcombe obligation", "WD analysis", and so forth).
This issue will be fully resolved once I've trained a model that can distinguish these citation references from their comparatively superfluous counterparts. An interim (and possibly good enough) solution will simply detect citations inside texts and exclude them from the text that ultimately gets submitted to GPT for summarization.
GPT-3.5's inexpensiveness makes removing citations less a priority than it used to be. The program is currently set up to run certain legal analyses based on whether it detects certain citations.
Citations are generally not semantically important for text summarization, though they may be useful for weight analysis, relational values, identifying legal tests, etc. Citations are important when they are used as labels for legal tests (eg, "Grant test", "Stinchcombe obligation", "WD analysis", and so forth).
This issue will be fully resolved once I've trained a model that can distinguish these citation references from their comparatively superfluous counterparts. An interim (and possibly good enough) solution will simply detect citations inside texts and exclude them from the text that ultimately gets submitted to GPT for summarization.