This is Chatbot CHELSEA (CHat Emotion Logic SEnse Automator)

CHELSEA is directly based off of my other chatbot, OPHELIA. Note that the Math Logic is still not fully tested, but everything else seems to be working the way it should. If choosing to start with almost blank memory, copy the 3 files from 'starter_memory_files' to the directory of the python scripts. If choosing to use my own chatbot's memory (Inherited from OPHELIA), copy the 3 files from 'my_memory_files' to the directory of the python scripts.

IMPORTANT: If you are viewing this repo for the first time and want to jump into working with this chatbot, it would be best to just download the highest 'BETA#' version directory, it has everything you need and is current. Some things in the highest version may still be experimental, so there may be some uncaught bugs, just to warn you.

This is a video demonstration of talking to CHELSEA: https://youtu.be/8tlWXGgrxCo?si=wmqwHlWXvzgXSySX

Note: The directories with 2 chatbots talking to each other use the old memory file format, they are not compatible with the newer version of CHELSEA, and vice-versa. That being said, everything that's included in the directory works fine for those double chats. I will be making an updated version of the double chat for CHELSEA and her copy, Dasha, that works with the new json memory file format.

Here is the algortihm for how CHELSEA deals with the user's message:

Chatbot CHELSEA message handling algorithm

1. User enters message

2. Detect if math question, use external script to handle and give solution to question (Highly experimental)

3. Detect if asking what she is or is not

4. Detect if asking what user is or is not

5. Detect if telling what she is or is not, learn accordingly Also detect if contradiction to what she already knows

6. Detect if current user describing what they are or are not Learn accordingly, detect if contradiction to what she already knows

7. Split message into words, remove punctuation Determine mood of reply, add counts to CHELSEA mood as well

8. Mark unkown words in message

9. Add to counts for each word, adjust the tied emotion to them accordingly

10. Take each word and associate all other (non-neutral) words in message with them (Either start or add to count) Comes into play in matching single associated word with highest count, or select one randomly from list of highest

11. Add to running counts of all words in conversation Get topic(s) of conversation by the maximum value

12. Keep track of matching words from 3 previous message/response pairs Get list of these 'depth words'

13. Mark unknown words in dictionary for 'what is/are __' unanswered questions

14. Check for answer to previous 'what is/are question'

15. Check if user is asking What Question, give possible answer response if so

16. Check if user asking for clarification on previous response, try to find matching message containing 2 or more words from previous response

17. Check if asking what CHELSEA feels about ___, respond according to the tied emotions with the words in ___

18. Check if asking 'do you like/dislike ___' question respond according to tied emotions to words in ___, agreement or disagreement depends on emotion and like/dislike word used

19. Check if asking 'which is better, 1 or 2' Used emotions tied to the words to decide which CHELSEA likes better, or indifferent if neither is tied to happy overall

20. Check if asking 'why is' question, try to find match involving 'because' words a possible answer

21. Check for 'most emotion' question, respond with biggest word(s) tied to emotion

22. Check for an exact match of message, give linked response if so

23. Check if message matches as part of a message in memory, give linked reponse if so

24. If certain or greater messages and words in dictionary, PRNG to determine if attempting topic match, or 'depth match' (Coin flip between two)

25. If certain or greater messages and words in dictionary, PRNG to determine if doing single term match. If so, coin flip to determine whether trying to match single word from message as part of message in memory, or single word associated with word from message. Repond with linked response if so

26. No match, add response to list in message memory or learn brand new message/response pair

27. 1/3 chance asking unanswered question, else respond with random response from memory to keep conversation going