DEVLOG 2: (TD;DR: built the base of b4uSend’s local rule based NLP system with 19 communication signals and context aware scoring) . Started working on my NLP system that powers b4usend . So essentially today i built the base for a rule based NLP text analysis system . My analyzer works by using 19 communication signals which include politeness, warmth , friendliness, warmth , frustration , hostility , pressure , urgency , blame and so on and the way this works is by using context aware scoring so it doesn’t simply look for keywords but rather it looks around a phrase to see if its being negated , hedged , used in a conditional way , quoted and so on . So for negation handling, words such as not ,never , can’t , don’t etc reduce the strength of the detected signal . For hedging detection words or phrases such as “maybe” “i guess” “probably” lower the certainty . Detecting conditionals such as “if” “when” “could” and so on can affect the confidence of interpretation. Also there is signal specific context where for instance hostility to yourself is treated as different form hostility to others and things like blame need evidence of a target , things like sarcasm should get special treatment as positive words can be used sarcastically and polite wording can be downgraded if the surrounding text seems sarcastic , to detect intensity we use indicators like multiple occurrences of “!” or “?” or strong words like “absolutely” or “totally” etc . Formality detection is done by categorizing words like “furthermore” , “respectfully” etc into formal and words like “yo” , “gonna” , “wanna” etc as casual and thus finally gives us a confidence score based on the number of patterns found , message length and contextual cues . The main benefit of this is that its completely local and doesn’t send your messages to an ai model elsewhere and all processing is done locally . I’ve just started and have a very basic version right now but will update it very frequently
