nephased 0.0.3

A BERT-based text sentiment classification pipeline for Nepali

nephased

[!Warning] This section contains vulgar words

Nephased provides a BERT-based classification pipeline for detecting Nepali text sentiment

Installation

From TestPyPI:

pip install nephased

Or you can use the Nephased(finetune of distilbert-base-nepali) from huggingface

Usage

Import Nephased module using the following command.

from nephased import Nephased

Initialize Nephased

clf = Nephased()

• You can pass a single string:

>>> clf.predict("थुक्क पैसा मा बिकने हीजडा")
'PROFANITY_0'

• or, a list of string:

>>> clf.predict(["राडिको छोरोको शासन धेर दिन टिक्दैन |", "सुरु मा चाहिँ तैले यो देश छोडनु पर्यो |", "एसको घरमा आगो लाहिदे ।"])
['PROFANITY_1', 'GENERAL', 'VIOLENCE']

About Output

Nephased can distinguish between 4 categories:

• GENERAL : Instance without any profanity or violence.
• PROFANITY_0 : Instance including rude, bad or slander which are not very harsh but offensive words used on day-to-day lives in Nepal.
• PROFANITY_1 : Instance including swear or curse words which are very harsh
• VIOLENCE : Instance including physical assualt or rape and pyromaniac act.

The guidelines for segragating such sentiments are on NepsaGuidelines

[!NOTE] Nephased is trained on NepSa dataset
By default Nephased preprocesses the input:

• stemming using nepali-stemmer
• lowering case, punctuation and stopwords removal
  you can choose to not preprocess text when initializing Nephased

clf = nephased(preprocess_text = False)