Headlines Classifier

News headlines are some of the most volatile and consumed pieces of information on the planet. We have made a headlines classifier to help users determine when headlines are junking up their content consumption ecosystem. We have also made it open-source so users can see how it works and operates. Our dataset is open-source and freely available.

Check out our demo and/or codebook

Usage

Models available for testing purposes only https://raw.githubusercontent.com/CuratedNews/headlinesclassifier/main/headlinesclassifier.tflite & https://raw.githubusercontent.com/CuratedNews/headlinesclassifier/main/headlinesclassifier2.tflite

Construction in Python

install if not already installed

!pip install -q tflite-model-maker-nightly

import packages

import pandas as pd
from tflite_model_maker import model_spec
from tflite_model_maker import text_classifier
from tflite_model_maker.text_classifier import DataLoader

import dataset and check if dataset imported correctly

df = pd.read_csv("headlinesvolatilitydata.csv")
df.head(25)

check dataset for total counts of positive, negative, and neutral sentiment labels

df['titlesentimentoverall'].value_counts()

set model specifications

spec = model_spec.get('average_word_vec')

for more information on text classification specifications see tensorflow docs

set parameters for model and iterate 10 times

train_data = DataLoader.from_csv(
      filename='headlinesvolatilitydata.csv',
      text_column='title',
      label_column='titlesentimentoverall',
      model_spec=spec,
      is_training=True)
model = text_classifier.create(train_data, model_spec=spec, epochs=10)

check model summary

model.summary()

How it works?

Check our demo for a hands-on with explanations

Jupyter Notebook

Do it yourself

Want to know more?

Visit https://curatednews.xyz