Defining the similarity of two strings is a large problem that must be fixed to perfect natural language processing models on Scratch, and especially to perfect fuzzy logic chatbots. For a long time, I used a simple system which checked if the nth character of both strings was the same, but small variations in spelling or even a simple offset by a space character being placed before a string can completely zero the similarity of two strings that look nearly identical. This algorithm searches for occurrences of all substrings of a string in another string, and per each occurrence increases its similarity score. Thus, it does not only check the similarity of individual characters, but also words, clauses, and sentences. Larger substrings that match increase the score proportionally to their size (substring of 1 character increases the score by 1, substring of 5 characters increases it by 5) because of how uncommon they increasingly become to find. This algorithm is extremely similar to that likely used by Cleverbot, a chatbot that works in the same way as modern Scratch chatbots but performs noticeably more naturally. I will soon be implementing it in a new chatbot alongside other features to drastically increase the speed of the chatbot's training.
