Hello,
I've read many review papers about which is the best classifier of Deep Learning in "text classification", some researchers prove that LSTM is the best, Some say CNN is the best, and some stated that hybrids such as LSTM/GRU, or BiLSTM is the best. There are studies indicate that when the size of the data set is large or the sentences are long, it is preferable to use the LSTM, some research indicate that the use of the CNN is faster and has wonderful results and is used with a short sentences. I am really confused, I don't know which classifier based Deep Learning technique should I use with text classification? Could anyone expert in this field give me the answer to my question or even a suggestion? supported by research or evidence.
Many thanks in advance.
If you need immediate implementation or evaluation, you can use Keras https://keras.io/examples/.
BERT, HuggingFaces, and GPT are the ones to go to.
:)
There're no any specific algorithms to claim the best. However, if we can identify the problem domain and nature of work, we can claim the best. Specifically, it depend on the problem. For example, it the sequence of words matter in the sentence, then LSTM/RNN or any sequence-based DL methods work better than CNN.
The best one is the LSTM-RNN ( Long Short Term Memory - Recurrent Neural network) since it deals with the sequence of data like the text rather than the CNN that deals commonly with images
The main two advantages of the LSTM are :
1) it is recurrent in its work that connect the previous input with the next one and complete the process untill the end of the text
2) it contains set of gates that preserve the content and the text information during the classification process
Dear Suhaib Kh. Hamed ,
Deep learning is a set of algorithms and techniques inspired by how the human brain works, called neural networks. ... The two main deep learning architectures for text classification are Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN).
https://monkeylearn.com/text-classification/
Regards,
Shafagat
Dear Suhaib Kh. Hamed
RNN-LSTM or Random Forest( I thnik it is best classification algo)
#Opinion
The answer by Chiranjibi Sitaula is the most accurate. If the order of works matters then RNN and LSTM should be best. If the order of words does not matters the simple DNN - Dense neural networks with or without CNN (convolution) on top of them, should be best. Still, the last layer must be dense nn with proper activation (such as binary or categorical cross-entropy).
Hello Suhaib Kh. Hamed,
you could try looking at state-of-the-art text classification methods:
https://paperswithcode.com/task/text-classification .
Bear in mind that most of these approaches are benchmarked on large datasets and may not work directly out-of-the-box if you were to decide on one.
Aside from that, I agree with Chiranjibi Sitaula and Barak Avisar .
Cheers,
Raoul
No single classifier is solely effective for every type of data at hand. Some of the classifiers have already been tested and recommended for particular types of databases, however., one may choose one's as according to the ground truths of their own research questions. A list of some popular classifiers is here : https://www.greycampus.com/opencampus/machine-learning/different-types-of-classifiers
I didn't get what he means by saying classifying text with CNN? is it image or text?
Based on some of the research papers, Hierarchical attention neural networks seem to be best for text classification in grasping the meaning of sentences and performing the task.
I think in Machine learning the only way to find the best classifiers is try-and-error
you can assume few classification techniques and make your experiments then report
your accurate database will play a key role in the results
GIGO Garbage In, Garbage Out
QIQO Quality In, Quality Out
- Badges
- Science topic