Madhav-Kanda / Statistical-Language-Modelling-using-N-gram

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Check out the Space at https://huggingface.co/spaces/Shruhrid/Next_Word_Prediction

Probabilistic approach to Next Word Prediction

The task of predicting the next word using an n-gram model is equivalent to the probability of getting a specific word at the nth place given the previous n-1 words. The probability of the same is given by the Bayesformula,P(w |w₁,.....,w ₋₁)

Key Aspects

Laplace Smoothing

Laplace smoothing is a technique that helps when the dataset remains limited for a plrobabilistic approach. Essentially it helps to avoid the problem of zero probability. For a bigram approach:

image

Keyword Meaning
P* Probability of the Laplace Smoothed N-grams
Wi i’th word
𝑐(Wi,Wi-1) count of word sequence Wi-1,Wi)
V total number of words in vocabulary

Perplexity

Being confused ain't as bad as you may think! Perplexity is a measure of how certain the model was while making the predictions. It is the inverse of the probability of predicting the test set normalized by the number of words.

About


Languages

Language:Python 100.0%