0
$\begingroup$

I have twitter data which is labelled with the sentiment(Postive, Negative, Neutral) and I have evaluated the performance of Tf-Idf and Doc2Vec feature extractor using the KNN algorithm and logistic regression.

The output for using KNN is

enter image description here

enter image description here

The output for using logistic regression is enter image description here enter image description here

From the above output, I could see that TF-IDF is better than Doc2Vec in logistic regression and Doc2Vec is better than TF-IDF in KNN. Then in KNN, TF-IDF is performing poorly because of recall score.

Is there any way that one can say which feature exactor is better based on the above output and what is the meaning if TF-IDF has high precision but poor recall score. I need some conclusion based on the above output. Any insight is much appreciated. Thanks in advance!

$\endgroup$
0
$\begingroup$

The first important part about your problem is its an imbalanced multi-class classification. In these cases, it's a valid point to tune the threshold of classification which is by default = 0.5. Moving to the important part is that since both of your classifiers - KNN & Logistic Regression aren't perfectly tuned it's not sound to compare the classifiers or the method of feature extraction.

Secondly, your different classifiers are responsible for different decision boundary - LR makes a linear unlike KNN with non-linear boundaries.

Thirdly, usually I have seen doc2vec to be better but it depends how good is your data. If documents are really small then doc2vec doesn't provide the additional benefit hence not much difference might be observed in TF-IDF & doc2vec.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.