References - Chatbot - The Chatbot Explained

The most essential thing is a dictionary of 142 694 words. The dictionary contains nouns, verbs, adjectives, auxiliaries, pronouns, etc. Each word have a type (nouns, verbs, adjectives), a kind (masculine, feminine), an amount (plural, singular) and a verb tense (present, past, futur) in the table. When the user submit a sentence, it get splited at each dot couting all sentences and then splited at each space counting the words. Each word get it's data fetch from the dictionary. Then the adjective words are matched inside the synonymous table to get an extra word for later. All the words are stored inside a new response array. The user's verbs and pronouns are transformed from "I am" to "You are" and from "You are" to "I am".

The chatbot table contains for each row each type (verbs, adjectives, nouns, auxiliaries, pronouns...) for the columns, a pattern column and a question column. Each type column contain a serialized object of the user's words. The pattern column is a syntax string of the user sentence for additional functionalities only. The question column is used to store a chatbot question and match it to the next user's sentence. If for example some the user's sentence nouns matches, with an OR condition, some of the nouns in the keywords column and doing so for the other keywords type using an AND condition together, the chatbot will return the human column sentence which is an old user's transformed sentence. The users verbs are matched with all the other tenses of the dictionnary before being matched into the bot memory (using an OR condition). The nouns are matched with a dictionnary of synonymous for each nouns (using again an OR condition). The query will find short memory sentences for short user input and long memory sentences for long user input.

If words aren't matched the chatbot will create a new sentence with all the user's words using another type of sentence builder with a new syntax. Words will be analysed with their singular, plural, feminine or masculine and their types (adj, nouns, verbs, etc) granted to validate the sentence structure. There are a lot of sentence syntax. The response is then returned to the user.

One subject cannot be repeated multiple times. Once a subject is written the chatbot will store the user's sentence keywords in a session and avoid it. But after a few sentences the counter will be resetted and the sentence will be allowed again. This same temporary session is also used to find more precise results when querying the memory. The more user input there is and the more the context will be precise.

When the chatbot find a word which is already stored in the temporary session it will search randomly in the rows for a new subject but will still avoid using the old rows.

When the chatbot decided to use a question mark in its output it will learn the next user sentence after being reformulated by the chatbot. The chatbot will store this user sentence with an array of keywords which are being accumulated each user input (short term memory). If the user change subject and no other word is found from the previous answer the short term memory will be flush to start a complete new array with different subject. The chatbot will also learn sentence of the user input which contains possessive adjectives.

If the Google Natural Language API is enabled the chatbot will look for the proper nouns in the user input which are not matched in the dictionnary and have a specific meaning (for example a company name). This noun will be matched in wikipedia database and the description will be stored inside the database. If the user input contains specific words such as: "do you know" or  "what is" the chatbot will return the wikipedia description to the user. This stored wikipedia description will be used for the cron task functionality to help the chatbot submit wikipedia description to itself and make independent associations each minutes. This process will build the memory on its own but needs a minimum of 30 rows of memory to make significant links.

If the user input is negative after a new row is added to the memory, this same new row will be deleted because it's invalid. This is an extra security feature that avoid storing bad result in the database.

The chatbot is able to make small calculus. There is a conversation table (for liter, milliliter, ounces, centimeters, meters, millimiters, kilometers, feets, inches, squares and cubes units) to allow the chatbot calculating with different units. It can make additions, multiplications, substractions, divisions by firstly converting the equation units to the same units. It can also return a specific unit in another unit; for example centimeters to cubes.

The chatbot face consist of a small python app which can generate shapes such as oval, squares, triangle using the integer 1 on a grid filled with 0 integer. This feature also have a few faces already built-in with emotion which are matched to specific keywords. A small animation is played when a keyword is stored in the short term memory. Animations will be played each answer until the subject is flushed.

This chatbot is custom built.

Source : learningbot

Price: $ 00.00 CAD to access learningbot extensions, forum support, chatbox support and update support