CODEWITHSUNDEEP
machine-learningscikit-learnnlpopennlpcreme
siddhartha chakraborty
siddhartha chakraborty

Reputation: 475

How to implement incremental learning in NLP

We are building a system wherein, we would have a initial very small amount of trained data to start with. The job is to Classify the incoming data(Document, for our case) into 2 categories: Category A & B. Data is document , so the user needs to classify the Document to belong to Category A or B. So, with the limited amount of data, we create the trained data set and we start predicting the Category of next Document using the trained data set.

Now if the prediction is correct user moves to the next Document. But if the prediction is incorrect , the user inputs the correct Category (Lets Say Category A was predicted by the system, wherein the correct assignment to the data should be Category B). So now the system should use this learning(Category B instead of Category A) to enrich(learn) itself in near real time.

It should train only the added data and not the complete dataset , which is already trained. So it should be incremental learning. For classification we would be applying Naive Bayes Classification.

Now the question is :

Upvotes: 5

Views: 1141

Answers (1)

Boring Guy
Boring Guy

Reputation: 83

Disclaimer: I am one of creme main developper.

Creme seems to be the right solution to your problem. One of the goals of creme is precisely to train observation by observation without training on the whole dataset every time.

If you want an example of using creme where the dataset is initially small and grows over time, you can find one here.

Upvotes: 3

Related Questions