Background info Pod #1

This will give a quick breakdown of background info that will be helpful for each project.

How to create the Training Data: Deep learning

---

patterns (x)-

"Hi" "How are you?"

tag (y) ---> the tags for patterns

greeting

---

NLP Basics:

Tokenization: splitting a string into meangingful units(diff categories ex. words, numbers, punctuation)

Stemming: Generate the root form of the words. Crude heuristic that chops off the ends of words. ex: "organize", "organizes", "organizing" ----->["organ", "organ", "organ"] OR "University", "Universe" ----->["Univers", "Univers"]

To be able to get an accurate response you'd put the string of data (patterns) into a 'bag of words'. All single words within the string are put into an array. The array size will be the size of all the words.

    ["Hi", "How", "are", "you"]

"Hi"--->[1, 0, 0, 0] 0(greeting)

    X             |        Y (Y vector)

---

NLP Preprocessing Pipeline ex:

    "Is anyone there?"
        |
        | tokenize
        V
    ["is", "anyone", "there"]
        |
        | lower + stem
        V
    ["is", "anyone", "there", "?"]
        |
        | exclude punctuation char.
        V
    ["is", "anyone", "there"]
        |
        | bag of words
        V
    [0,0,0,1,0,1,0,1]

---

Feed Forward Neural Net:

X -----> Neural Net( # of patterns--->#of classes)--->Softmax--->Y

-Input is the bag of words -First layer of the neural net is 3 of patterns (layer for connected) -then hidden layer, another hidden layer -output layer

---