https://www.tidytextmining.com/tfidf.html

https://www.tidytextmining.com/tfidf.html

Dear Nerella,

That's correct while working with textual data, it's difficult to find generic data set for training and performance evaluation.

I would suggest you to 1) evaluate your model with existing dataset that you have (though domain specific), 2) prepare your own corpus of the domain you work for, and 3) compare cross domain performance of your model (training set of one domain, test set from other and vice-versa). I hope this woud provide you clear picture of your domain and model.

Dear Mahoto,

Thanks for the reply. Unfortunately my test set has no labels assigned.

This is an overview of my task.

The goal of the task is to predict readability level of an article.

I have around 4000 story books(total) in 5 classes.

I extracted lexical, syntactical features and achieved 86% accuracy. Now I wanted to test this classifier on news data and unfortunately I donot have labels for.this.

When I used my classifier on news most of the articles are in level 4 and 5 but in reality they are not of that level. So to refine my features I though of using adverbs, adjectives as my features for this I thought of using tfidf but this tfidf is a corpus specific.

For this reason I am trying to find alternative solution using tfidf on cross domain.

I hope you have got my point.

Thanks again

Udaysimha Nerella

Hello Mr. Sobhan,

Thanks for your reply could you kindly check my very recent reply and could you provide your thought on it.

thanks

Udaysimha Nerella