How To Build Fake News Detection Model Using NLP

Rohan Volety -

Table of Contents

  1. Introduction
  2. Role of NLP in News Detection
  3. NLP Example
  4. Creating NLP Model
  5. Challenges in NLP
  6. Conclusion

Introduction

In today’s fast-paced digital world, information comes to us in every direction, thus shaping our beliefs and decisions. However, in between this flood of information lies a threat of fake news.

Fake news is basically news or different kinds of information that are presented as legitimate news on the web or on various platforms. It can be of different forms, like long stories, manipulated images, videos, etc.

Spreading of fake news has become a major concern in today’s digital world.

Natural Language Processing(NLP) is a powerful tool to combat this issue. It provides us with tools and techniques to analyze text and raise flags for any fake news.

Role Of NLP in News detection

NLP plays an important role in detecting fake news by using different techniques to analyze the text data. Different techniques used by NLP for fake news detection:

  1. Text Classification: NLP models are trained to classify news articles or social media posts as either real or fake based on patterns like word structure, sentence structure, or other features.
  2. Sentiment Analysis: NLP analyzes sentiments expressed in news articles or social media to detect any misleading content by examining the emotional tone and language used in the text. It can flag any deceptive or manipulative information.
  3. Named Entity Recognition (NER): NLP techniques can identify any named entities mentioned in the text like people’s names, organizations, locations, and dates. If any false information is found with the organization or the person's name across different sources, it would flag them.
  4. Semantic Analysis: NLP models can also understand the semantic meaning of text by understanding the relationship between different words and phrases. It tries to get information from the data in real time and flags any contextual clues that indicate misinformation.
  5. Topic Modeling: NLP techniques have the ability to tell us about the underlying topics or themes present in a collection of articles. It tells us about patterns of content manipulation or agenda-driven objectives used in fake news.
  6. Fact-Checking: NLP systems can automatically verify the accuracy of claims made in news articles by cross-referencing them with reliable sources. Fact-checking algorithms can easily identify inconsistencies and contradictions in new stories.
  7. Network Analysis: NLP can analyze the network structure of social media to identify accounts or bots spreading fake news. It would examine the patterns of communication, and information network analysis can detect fake news campaigns.

NLP Example

There is a news article getting viral claiming “Scientists Discover Coffee that can cure Cancer! ” with lots of images, videos and numerous comments.

Let’s see NLP in action:

1)  Content Analysis:

Topic Modeling: The system may flag the topic “Cancer Cure” as uncommon for a coffer related article.

Named Entity Recognition(NER): NER would verify the names of the scientists involved and also the organization behind this research.

Sentiment Analysis: In the post ,the language used can be positive and emotionally charged, which would raise a red flag for the news.

2) Stylistic Analysis:

Textual Features: NLP will examine all the words, vocabulary, sentence structure ,and also grammar. This would tell us about any unusual patterns compared to other reliable sources.

Clickbait Detection: NLP will look for any manipulative keywords or phrases in the headline of the news and flag them.

3) Fact-Checking and Verification:

Knowledge Graph: Cross checking the claims against liable medical sources and other scientific sources.

Social Network Analysis: Find the origin of the post and the pattern it is following like bots, suspicious accounts and posts , shared and liked by people in a particular region.

NLP techniques might raise multiple red flags, indicating the above news is fake. It would also prompt you to stop sharing the post and further investigate yourself by checking reputable news sources and medical journals.

Creating NLP Model

For the following tutorial dataset is downloaded from the kaggle website.

1. Import Required Packages:

import numpy as np
import pandas as pd


2. Load and View the dataset

fake_news = pd.read_csv('Fake.csv')
true_news = pd.read_csv('True.csv')

fake_news.head()

true_news.head()

3. Data Pre-processing:

# Add indicator of true or fake
fake_news['isTrue'] = 0
true_news['isTrue'] = 1
# Combine both the Datasets
df = pd.concat([fake_news, true_news], axis=0)
#Dropping unnecessary columns and
df = df.drop(['title', 'subject', 'date'], axis=1)
df.head()

4. Data Visualization

# Visualisation
label_counts = df['is_true'].value_counts()
plt.bar(label_counts.index, label_counts.values, color=['red', 'blue'])
plt.xlabel('Label')
plt.ylabel('Count')
plt.title('Distribution of Labels')
plt.xticks([0, 1], ['Fake', 'True'])
plt.show()

5. Cleaning text to remove any unwanted strings 

import re
import string
def preprocess_text(text):
text = text.lower()
text = re.sub(r'\[.*?\]|\W|https?://\S+|www\.\S+|<.*?>+|\n|\w*\d\w*', '', text)
return text
df["text"] = df["text"].apply(preprocess_text)

6. Prepare Data for Modeling

x = df["text"]
y = df["isTrue"]

7. Divide the dataset into 80:20 for training and testing

from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size = 0.20)

8. Before training model converting text data to vectors

from sklearn.feature_extraction.text import TfidfVectorizer
vectorization = TfidfVectorizer()
xv_train = vectorization.fit_transform(x_train)
xv_test = vectorization.transform(x_test)

9. Training a LogisticRegression Model

from sklearn.linear_model import LogisticRegression
lr = LogisticRegression()
lr.fit(xv_train,y_train)
pred_lr = lr.predict(xv_test)
from sklearn.metrics import classification_report
print(classification_report(y_test, pred_lr))
from sklearn.metrics import accuracy_score
accuracy_score(y_test, pred_lr)

Challenges in NLP

  1. Evolving Techniques: Fake news creators regularly update their tactics. They use sophisticated language manipulation and other fabrication techniques to misguide the model.
  2. Data Quality: NLP models require high quality and diverse training data. Biased or incomplete datasets can lead to low accuracy and hinder the performance of the NLP model.
  3. Limited Data: Training effective NLP models requires huge amounts of labeled data, but acquiring and labeling such data is expensive and time consuming.
  4. Multimodal Content: Fake news detection is not limited to text alone but also involves analyzing images ,videos and other content. Integrating multimodels adds complexity to NLP systems.

Conclusion

In conclusion, While fake news remains a major challenge in the digital age, Natural Language Processing(NLP) provides us with a promising solution for its detection and mitigation.

NLP models can analyze given data using techniques like sentiment analysis, semantic analysis to detect patterns and inconsistencies in the content. By combining the strengths of NLP with human expertise, we can detect fake news and further avoid sharing misleading information.

Frequently Asked Questions

Q1. What do you mean by NLP?

Natural language processing(NLP) is a branch of artificial intelligence(AI) that allows computers to understand, interpret and generate human language. NLP techniques allow machines to interact with humans in a natural way, enabling tasks like sentiment analysis, text summarization, and speech recognition.

Q2.How is NLP used in sentiment analysis?

Sentiment analysis is a technique used to determine whether data is positive, negative, or neutral. It is performed on textual data to monitor customer reviews of different products and brands.