Fake News Detection Using NLP

ABSTRACT
The proliferation of fake news across online platforms poses a significant threat to
public discourse and societal well-being. The rapid and widespread dissemination of
misleading or false information can influence public opinion, incite social unrest, and erode
trust in credible sources. Addressing this challenge necessitates the development of effective
automated methods for identifying and mitigating the spread of fake news. Natural Language
Processing (NLP) techniques offer a promising avenue for tackling this complex problem by
enabling the analysis of textual content to discern patterns and linguistic cues indicative of
misinformation.
This abstract explores the application of various NLP methodologies in the domain
of fake news detection. We delve into feature engineering approaches, which involve
extracting relevant linguistic features from text, such as sentiment, subjectivity, writing style,
and the presence of specific keywords or rhetorical devices often associated with fake news.
Furthermore, we examine the role of machine learning algorithms, including traditional
methods like Naive Bayes and Support Vector Machines, as well as more advanced deep
learning architectures like Recurrent Neural Networks (RNNs) and Transformer models, in
classifying news articles as either credible or fake based on these extracted features.
The abstract also considers the importance of incorporating contextual information
and external knowledge sources to enhance detection accuracy. This includes leveraging
social network analysis to study the propagation patterns of news and identifying potentially
suspicious sources or user behaviors. Additionally, the integration of knowledge graphs and
fact-checking databases can provide valuable external validation for the claims made within
news articles, thereby improving the reliability of fake news detection systems.
Challenges and future directions in the field are also discussed. These include the
evolving nature of fake news, the sophistication of adversarial attacks aimed at evading
detection, and the need for robust and interpretable models that can explain their reasoning.
Addressing these challenges requires ongoing research into more nuanced feature
representations, the development of more resilient and adaptive models, and the exploration
of explainable AI techniques to build user trust and facilitate human oversight.

1
Department of CSE(AI&ML)
 Fake News Detection Using NLP

CHAPTER - 1: Introduction

1.1 Introduction to Fake News Detection

The digital age has ushered in an unprecedented era of information accessibility, yet
this abundance is accompanied by the pervasive challenge of fake news. The ease with which
information can be created and disseminated online has unfortunately been exploited to
spread misleading, inaccurate, or entirely fabricated content. This phenomenon poses a
significant threat to individuals and society as a whole, impacting public opinion, political
processes, and even personal well-being. The ability to discern credible information from
falsehoods has become a crucial skill in navigating the modern information landscape.
The rapid and widespread propagation of fake news is facilitated by social media
platforms and online news outlets, often amplified by algorithmic curation and echo
chambers. Unlike traditional forms of misinformation, which might have been confined to
specific channels, fake news can quickly reach vast audiences, transcending geographical
boundaries and societal segments. This velocity and reach make it particularly challenging
to counteract the influence of false narratives once they gain traction. Consequently, there is
an urgent need for effective mechanisms to identify and mitigate the spread of fake news in
real-time.
Addressing this complex problem requires a multi-faceted approach, involving
technological solutions, media literacy initiatives, and policy interventions. Among the
technological solutions, Natural Language Processing (NLP) has emerged as a powerful tool.
By enabling the automated analysis of textual content, NLP techniques can identify subtle
linguistic patterns, stylistic anomalies, and semantic inconsistencies that may indicate the
presence of fake news. This capability offers the potential to develop scalable and efficient
systems for detecting and flagging misinformation before it causes significant harm.
This paper delves into the application of NLP methodologies for fake news detection.
We will explore various techniques, ranging from traditional feature engineering approaches
to advanced deep learning models, that can be employed to analyze the textual characteristics
of news articles.

2
Department of CSE(AI&ML)
 Fake News Detection Using NLP

1.2 Importance of Fake News Detection

The ability to effectively detect fake news holds paramount importance in today's
interconnected world for a multitude of reasons. Firstly, the unchecked proliferation of false
information can severely erode public trust in legitimate news sources and institutions. When
individuals struggle to distinguish between credible reporting and fabricated stories, it
undermines the foundation of an informed citizenry, hindering sound decision-making on
critical societal issues. This erosion of trust can have far-reaching consequences, impacting
everything from public health to democratic processes.
Secondly, fake news can have tangible and detrimental real-world impacts.
Misinformation surrounding health crises can lead to harmful behaviors and hinder effective
public health responses. In the political sphere, fabricated stories can sway public opinion,
manipulate elections, and exacerbate social divisions. Economically, false rumors can
destabilize markets and damage the reputations of businesses. The potential for harm
underscores the urgency of developing reliable methods for identifying and mitigating the
spread of fake news before it can inflict significant damage.
Furthermore, the sheer volume of information generated online makes manual fact-
checking an unsustainable solution. The speed at which fake news can spread across social
media platforms necessitates automated systems that can rapidly identify and flag potentially
false content. NLP-powered detection mechanisms offer a scalable and efficient approach to
address this challenge, providing a crucial layer of defense against the overwhelming tide of
misinformation. These systems can act as an initial filter, allowing human fact-checkers to
focus their efforts on more complex or ambiguous cases.
In essence, robust fake news detection capabilities are vital for safeguarding the
integrity of the information ecosystem. By developing and deploying effective NLP-based
tools, we can empower individuals to make informed decisions, protect society from the
harmful consequences of misinformation, and preserve trust in credible sources of
information. This is not merely a technological challenge but a societal imperative in
navigating the complexities of the digital age and ensuring a more informed and resilient
future.

3
Department of CSE(AI&ML)
 Fake News Detection Using NLP

CHAPTER - 2: Literature Survey

2.1 Existing Systems

The challenge of fake news detection has spurred significant research and the
development of various systems employing diverse approaches. Early systems often relied
on manual fact-checking, where human experts would investigate the veracity of claims.
While accurate, this method is inherently slow and struggles to keep pace with the rapid
dissemination of online information. This limitation highlighted the need for automated
solutions.

The advent of Natural Language Processing (NLP) and Machine Learning (ML)
techniques paved the way for more scalable and efficient fake news detection systems. Many
early automated systems focused on feature-based approaches. These systems involved
extracting linguistic features from news articles, such as sentiment, subjectivity, writing
style, presence of specific keywords (e.g., emotionally charged language, clickbait terms),
and grammatical correctness. Machine learning classifiers like Naive Bayes, Support Vector
Machines (SVMs), and Logistic Regression were then trained on these features to distinguish
between credible and fake news. For instance, systems analyzed the frequency of intensifiers
or the use of hyperbolic language as indicators of potentially biased or fabricated content.

Beyond textual content, researchers also explored source-based features. These

systems analyzed the characteristics of the news source itself, such as its reputation, domain
registration information, website design, and history of publishing accurate or inaccurate
information. Network analysis techniques were also employed to study the spread of news
and identify potentially unreliable sources based on their propagation patterns and
connections within social networks. Systems incorporating source credibility often
maintained blacklists or whitelists of known unreliable or reliable publishers.

More recently, the rise of deep learning has led to the development of more
sophisticated fake news detection systems. Recurrent Neural Networks (RNNs), particularly
LSTMs and GRUs, have been used to capture sequential dependencies in text, allowing
models to understand the context and flow of information within an article. Convolutional
Neural Networks (CNNs) have also been applied to extract hierarchical features from text.

4
Department of CSE(AI&ML)
 Fake News Detection Using NLP

2.2 Proposed System

Our proposed system aims to leverage the strengths of recent advancements in

Natural Language Processing and Machine Learning to create a robust and effective fake
news detection framework. The core of our system will be a deep learning model specifically
tailored for understanding the nuanced linguistic characteristics of both genuine and
fabricated news articles. We propose utilizing a Transformer-based architecture, such as
a pre-trained model like BERT or RoBERTa , fine-tuned on a carefully curated and balanced
dataset of real and fake news. The inherent ability of Transformer networks to capture long-
range dependencies and contextual information within text makes them particularly well-
suited for discerning subtle indicators of misinformation.

key aspects of the proposed fake news detection system:

1. Enhanced Contextual Understanding through Transformer Networks:

 Capturing Long-Range Dependencies: Transformer models excel at understanding

the relationships between words that are far apart in a sentence or even across
paragraphs, which is crucial for identifying subtle manipulations of context often
found in fake news.
 Contextualized Word Embeddings: Unlike traditional word embeddings,
Transformers generate embeddings that are specific to the context in which a word
appears, allowing the model to differentiate between different meanings and identify
nuanced language use indicative of deception.
 Fine-tuning on Specialized Datasets: The pre-trained Transformer model will be
fine-tuned on a diverse and well-labeled dataset of both genuine and fake news,
specifically curated to capture the linguistic patterns characteristic of misinformation
across various domains.

2. Leveraging Meta-Level Information for Holistic Analysis:

 Source Reliability Assessment: Analyzing the domain age, registration details, and
past publishing history of the news source can provide valuable cues about its

5
Department of CSE(AI&ML)
 Fake News Detection Using NLP

credibility. Integration with reputation scoring systems or community feedback

mechanisms could further enhance this aspect.
 Social Media Signal Analysis: Examining how a news article is being shared and
discussed on social media, including the speed of propagation, the types of users
sharing it, and the sentiment expressed in comments, can reveal suspicious patterns
indicative of coordinated disinformation campaigns.
 User Engagement Metrics: Analyzing metrics like the click-through rate, time
spent on the page, and bounce rate might offer indirect signals about the perceived
credibility or engagement level with the content.

3. Integration of External Knowledge for Factual Verification:

 Querying Fact-Checking APIs: Automatically querying established fact-checking

APIs (like Snopes, PolitiFact) with key claims from the news article can provide
direct evidence of whether the information has been previously verified or debunked.
 Knowledge Graph Integration: Utilizing knowledge graphs (structured databases
of facts and relationships) can enable the system to verify factual claims by checking
for consistency with established knowledge. For instance, verifying dates, locations,
and relationships between entities mentioned in the article.
 Identifying Contradictions and Inconsistencies: By comparing the information in
the news article with external knowledge sources, the system can identify
contradictions or logical inconsistencies that might suggest the content is fabricated.

4. Focus on Explainability and Interpretability:

 Attention Visualization: Transformer models' attention mechanisms can be

visualized to highlight the words and phrases that the model deems most important
for its classification decision, providing insights into its reasoning.
 Feature Importance Analysis: Techniques can be employed to assess the
contribution of different input features (textual, source-based, social) to the model's
prediction, helping to understand which factors are most influential.
 Generating Rule-Based Explanations: In some cases, the model's decision-making
process can be distilled into a set of understandable rules or patterns, making its
predictions more transparent to users.

6
Department of CSE(AI&ML)
 Fake News Detection Using NLP

CHAPTER – 3: Analysis and Design

3.1 System Architecture

Figure-3.1: System Architecture for Fake News Detection

Supervised Learning Process

Supervised learning involves training a model on labeled data to make predictions

on new, unseen data. The process can be broken down into the following key stages:

1.Training Data and Labels

In the initial phase, the Machine Learning Algorithm is provided with Training Text,
Documents, Images, etc. This training data consists of various examples relevant to the task
at hand. Crucially, each piece of training data is associated with a corresponding Labels.
These labels represent the correct or desired output for each input example. For instance, in

7
Department of CSE(AI&ML)
 Fake News Detection Using NLP

a fake news detection task, the training text would be news articles, and the labels would
indicate whether each article is "Fake" or "Real."

2.Feature Vector Generation

Before the training data can be fed into the machine learning algorithm, it needs to
be transformed into a numerical representation called Feature Vectors. This step involves
extracting relevant features from the raw input data. For text data, these features could
include word frequencies, sentiment scores, or more complex embeddings. For images,
features might involve pixel intensities or texture patterns. The goal is to convert each data
point into a format that the algorithm can understand and process mathematically.

3.Training the Machine Learning Algorithm

The Machine Learning Algorithm takes the Feature Vectors derived from the training
data and their corresponding Labels as input. During the training process, the algorithm
learns the underlying relationships and patterns between the features and the labels. It adjusts
its internal parameters to build a Predictive Model that can map input features to the correct
output labels. Various machine learning algorithms can be used depending on the nature of
the task, such as logistic regression, support vector machines, decision trees, or neural
networks.

4.Prediction on New Data

Once the Predictive Model is trained, it can be used to make predictions on New
Text, Document, Image, etc. This new, unseen data undergoes the same Feature Vector
generation process as the training data. The resulting feature vector is then fed into the
trained Predictive Model.

5.Generating the Expected Label

The Predictive Model processes the feature vector of the new data point and outputs
a prediction, which is the Expected Label. This label represents the model's best guess for
the correct output based on the patterns it learned during the training phase. The accuracy of
this predicted label depends on the quality and quantity of the training data, the effectiveness
of the feature engineering, and the suitability of the chosen machine learning algorithm.

8
Department of CSE(AI&ML)
 Fake News Detection Using NLP

CHAPTER – 4: Implementation

Implementing a fake news detection system involves several key stages, from data
collection and preprocessing to model training and evaluation. Here's a breakdown of the
implementation process with side headings:

4.1 Data Collection and Preparation

 Dataset Acquisition: The first crucial step is to gather a diverse and representative
dataset of both real and fake news articles. This may involve collecting data from
various sources, including reputable news websites, social media platforms (with
appropriate ethical considerations and API access), and publicly available fake news
datasets (e.g., LIAR dataset, Fake News Net).

 Data Balancing: It's essential to ensure a balanced representation of real and fake
news samples in the dataset to prevent the model from being biased towards the
majority class. Techniques like oversampling the minority class or under sampling
the majority class might be necessary.

 Data Cleaning: The collected data needs to be cleaned by removing irrelevant

information such as HTML tags, special characters, and noise. Handling missing
values and ensuring consistent formatting are also important.

 Data Splitting: The dataset is typically split into three subsets:

o Training Set: Used to train the machine learning model.

o Validation Set: Used to tune the model's hyperparameters and prevent

overfitting during training.

o Testing Set: Used to evaluate the final performance of the trained model on
unseen data.

9
Department of CSE(AI&ML)
 Fake News Detection Using NLP

4.2 Feature Engineering and Selection

 Textual Feature Extraction: As discussed earlier, various NLP techniques are

applied to extract meaningful features from the text of the news articles. This can
include:

o Basic Features: Word counts, character counts, average word length.

o Lexical Features: Presence of specific keywords or phrases associated with

fake news (e.g., clickbait terms, emotionally charged language).

o Syntactic Features: Part-of-speech tagging, grammatical error detection.

o Semantic Features: TF-IDF vectors, word embeddings (Word2Vec, GloVe),

contextual embeddings (from Transformer models like BERT).

 Meta-level Feature Extraction: Implementing the collection and processing of

source-based, social media-based, and user engagement features as outlined in the
proposed system architecture. This often involves interacting with APIs and parsing
structured data.

 Feature Selection/Reduction: To improve model performance and reduce

dimensionality, feature selection techniques (e.g., chi-squared test, information gain)
or dimensionality reduction techniques (e.g., Principal Component Analysis - PCA)
can be applied to select the most relevant features.

4.3 Model Selection and Training

 Choosing the Model Architecture: Based on the chosen features and the complexity
of the task, an appropriate machine learning model is selected. This could range from
traditional classifiers (e.g., Logistic Regression, Naive Bayes, SVM) to more
advanced deep learning models (e.g., RNNs, CNNs, Transformer networks).

 Implementing the Model: The chosen model architecture is implemented using a

suitable machine learning library (e.g., scikit-learn, TensorFlow, PyTorch).

10
Department of CSE(AI&ML)
 Fake News Detection Using NLP

 Model Training: The model is trained using the training dataset and the extracted
features. This involves feeding the data into the model and adjusting its internal
parameters to minimize the prediction error based on the provided labels.

 Hyperparameter Tuning: The model's hyperparameters (e.g., learning rate, number

of layers, regularization strength) are tuned using the validation set to optimize its
performance and prevent overfitting. Techniques like grid search or random search
can be employed for this purpose.

4.4 Knowledge Integration Implementation

 API Integration: Implementing the functionality to interact with fact-checking

APIs. This involves handling API requests, parsing responses, and extracting
relevant verification information.

 Knowledge Base Connection: Setting up and querying a knowledge base (if used).
This might involve using graph databases or other knowledge representation
systems.

 Feature Engineering with External Knowledge: Incorporating the information

retrieved from fact-checking APIs and knowledge bases as additional features for the
model. This could involve binary flags indicating whether a claim has been debunked
or confidence scores from fact-checkers.

4.5 Explanation Generation Implementation

 Attention Mechanism Analysis (for Transformer models): Implementing code to

extract and visualize the attention weights of the Transformer model to understand
which parts of the input text were most influential in the prediction.

 Feature Importance Calculation: Using techniques provided by the machine

learning library (e.g., feature importance scores in tree-based models, permutation
importance) to determine the contribution of different features.

11
Department of CSE(AI&ML)
 Fake News Detection Using NLP

 Evidence Presentation Logic: Developing the logic to present the evidence

retrieved from fact-checking APIs and knowledge bases in a user-friendly and
informative way.

4.6 Model Evaluation and Deployment

 Model Evaluation: The trained and tuned model is evaluated on the unseen testing
set to assess its generalization performance. Various metrics are used, such as
accuracy, precision, recall, F1-score, and AUC.

 Performance Analysis: Analyzing the evaluation results to identify strengths and

weaknesses of the model. This might involve examining confusion matrices and
identifying common types of misclassifications.

 Deployment: Once a satisfactory level of performance is achieved, the model can be

deployed as a web application, a browser extension, or integrated into other systems
to detect fake news in real-time. This involves building an interface for users to input
news articles and receive predictions and explanations.

 Monitoring and Maintenance: After deployment, the system needs to be

continuously monitored for performance degradation. Retraining the model with new
data and adapting to evolving patterns of fake news are crucial for maintaining its
effectiveness.

12
Department of CSE(AI&ML)
 Fake News Detection Using NLP

4.5 Source Code

Fake News.py

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from tqdm import tqdm
import re
import nltk
from nltk.corpus import stopwords
from nltk.stem.porter import PorterStemmer
from wordcloud import WordCloud
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score
from sklearn.tree import DecisionTreeClassifier
from sklearn import metrics

# Importing Libraries and Datasets

data = pd.read_csv('News.csv', index_col=0)

# Data preprocessing
data = data.drop(["title", "subject", "date"], axis=1)
data.dropna(inplace=True)
data = data.sample(frac=1).reset_index(drop=True)
# Preprocessing and analysis of News column
nltk.download('punkt', quiet=True)

13
Department of CSE(AI&ML)
 Fake News Detection Using NLP

nltk.download('stopwords', quiet=True)
stop_words = set(stopwords.words('english'))
porter = PorterStemmer()

def preprocess_text(text_data):
preprocessed_text = []
for sentence in tqdm(text_data):
sentence = re.sub(r'[^\w\s]', '', sentence)
tokens = word_tokenize(sentence.lower())
tokens = [porter.stem(word) for word in tokens if word not in stop_words]
preprocessed_text.append(' '.join(tokens))
return preprocessed_text

preprocessed_review = preprocess_text(data['text'].values)
data['text'] = preprocessed_review

# WordCloud for real news

consolidated_real = ' '.join(word for word in data['text'][data['class'] == 1].astype(str))
wordCloud_real = WordCloud(width=1600, height=800, random_state=21,
max_font_size=110, collocations=False)
plt.figure(figsize=(15, 10))
plt.imshow(wordCloud_real.generate(consolidated_real), interpolation='bilinear')
plt.axis('off')
plt.title('WordCloud for Real News')
plt.show()

# WordCloud for fake news

consolidated_fake = ' '.join(word for word in data['text'][data['class'] == 0].astype(str))
wordCloud_fake = WordCloud(width=1600, height=800, random_state=21,
max_font_size=110, collocations=False)

14
Department of CSE(AI&ML)
 Fake News Detection Using NLP

plt.figure(figsize=(15, 10))
plt.imshow(wordCloud_fake.generate(consolidated_fake), interpolation='bilinear')
plt.axis('off')
plt.title('WordCloud for Fake News')
plt.show()

# Bargraph of the top 20 most frequent words

def get_top_n_words(corpus, n=None):
vec = CountVectorizer().fit(corpus)
bag_of_words = vec.transform(corpus)
sum_words = bag_of_words.sum(axis=0)
words_freq = [(word, sum_words[0, idx]) for word, idx in vec.vocabulary_.items()]
words_freq = sorted(words_freq, key=lambda x: x[1], reverse=True)
return words_freq[:n]

common_words = get_top_n_words(data['text'], 20)

df1 = pd.DataFrame(common_words, columns=['Review', 'count'])
df1.groupby('Review').sum()['count'].sort_values(ascending=False).plot(
kind='bar', figsize=(10, 6), xlabel="Top Words", ylabel="Count", title="Bar Chart of
Top Words Frequency")
plt.show()

# Converting text into Vectors

x_train, x_test, y_train, y_test = train_test_split(data['text'], data['class'], test_size=0.25,
random_state=42)

vectorization = TfidfVectorizer()
x_train = vectorization.fit_transform(x_train)
x_test = vectorization.transform(x_test)

15
Department of CSE(AI&ML)
 Fake News Detection Using NLP

# Model training, Evaluation, and Prediction - Logistic Regression

model_lr = LogisticRegression()
model_lr.fit(x_train, y_train)
print("Logistic Regression Accuracy:")
print(f"Train Accuracy: {accuracy_score(y_train, model_lr.predict(x_train))}")
print(f"Test Accuracy: {accuracy_score(y_test, model_lr.predict(x_test))}")

# Model training, Evaluation, and Prediction - Decision Tree Classifier

model_dt = DecisionTreeClassifier()
model_dt.fit(x_train, y_train)
print("\nDecision Tree Classifier Accuracy:")
print(f"Train Accuracy: {accuracy_score(y_train, model_dt.predict(x_train))}")
print(f"Test Accuracy: {accuracy_score(y_test, model_dt.predict(x_test))}")

# Confusion matrix of Results from Decision Tree classification

cm = metrics.confusion_matrix(y_test, model_dt.predict(x_test))
cm_display = metrics.ConfusionMatrixDisplay(confusion_matrix=cm,
display_labels=[False, True])
cm_display.plot(cmap=plt.cm.Blues)
plt.title('Confusion Matrix for Decision Tree Classifier')
plt.show()

16
Department of CSE(AI&ML)
 Fake News Detection Using NLP

CHAPTER – 5: Results

Figure-5.1: Display few rows of Dataset

Figure-5.2: Display data prepocessing

17
Department of CSE(AI&ML)
 Fake News Detection Using NLP

Figure-5.3 Visualize the WordCloud for fake and real news Separately

Figure-5.4 Bar chart of Top words Frequency

18
Department of CSE(AI&ML)
 Fake News Detection Using NLP

Figure-5.5 confusion matrix for Decision Tree Classifier

19
Department of CSE(AI&ML)
 Fake News Detection Using NLP

CHAPTER – 6: Conclusion and Future Scope

CONCLUSION:

The widespread dissemination of fake news poses a significant threat to societal well-
being and the integrity of information ecosystems. This study explored the application of
machine learning, particularly focusing on NLP techniques, to automatically identify and
classify fake news articles. We implemented and evaluated several approaches, including
traditional machine learning models like Logistic Regression and Decision Trees, after
preprocessing and vectorizing the textual content of a substantial news dataset.

The results obtained from the implemented models demonstrate the potential of
machine learning in tackling the fake news problem. The Decision Tree Classifier, in
particular, achieved a high level of accuracy on the test dataset, indicating its ability to learn
complex patterns and effectively distinguish between real and fake news based on the textual
features. The visualization of word clouds for both real and fake news provided qualitative
insights into the distinct vocabulary and emphasis often found in these categories.
Furthermore, the analysis of the most frequent words offered a quantitative perspective on
the linguistic differences.

FUTURE SCOPE:

The future research in fake news detection should prioritize exploring advanced deep
learning architectures like Transformers and RNNs with attention mechanisms to enhance
contextual understanding. Integrating multi-modal information, including images, videos,
and social media context, will be crucial for a more holistic analysis. Leveraging external
knowledge from fact-checking databases and knowledge graphs can provide a stronger basis
for verification. Furthermore, focusing on explainability and interpretability of models is
essential for building user trust and enabling human oversight. Addressing the growing threat
of adversarial attacks and developing real-time detection and intervention strategies are also
critical areas for future work, alongside tackling the challenges of cross-lingual fake news
detection and exploring personalized approaches.

20
Department of CSE(AI&ML)
 Fake News Detection Using NLP

REFERENCES:

[1]. Parikh, S. B., & Atrey, P. K. (2018, April). Media-Rich Fake News Detection: A Survey.
In 2018 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR) (pp.
436-441). IEEE.

[2]. Conroy, N. J., Rubin, V. L., & Chen, Y. (2015, November). Automatic deception
detection: Methods for finding fake news. In Proceedings of the 78th ASIS&T Annual
Meeting: Information Science with Impact: Research in and for the Community (p. 82).
American Society for Information Science.

[3]. Helmstetter, S., & Paulheim, H. (2018, August). Weakly supervised learning for fake
news detection on Twitter. In 2018 IEEE/ACM International Conference on Advances in
Social Networks Analysis and Mining (ASONAM) (pp. 274-277). IEEE.

[4]. Stahl, K. (2018). Fake News Detection in Social Media.

[5]. Della Vedova, M. L., Tacchini, E., Moret, S., Ballarin, G., DiPierro, M., & de Alfaro, L.
(2018, May). Automatic Online Fake News Detection Combining Content and Social
Signals. In 2018 22nd Conference of Open Innovations Association (FRUCT) (pp. 272-279).
IEEE.

[6] Tacchini, E., Ballarin, G., Della Vedova, M. L., Moret, S., & de Alfaro, L. (2017). Some
like it hoax: Automated fake news detection in social networks. arXiv preprint
arXiv:1704.07506.

[7]. Shao, C., Ciampaglia, G. L., Varol, O., Flammini, A., & Menczer, F. (2017). The spread
of fake news by social bots. arXiv preprint arXiv:1707.07592, 96-104.

[8]. Chen, Y., Conroy, N. J., & Rubin, V. L. (2015, November). Misleading online content:
Recognizing clickbait as false news. In Proceedings of the 2015 ACM on Workshop on
Multimodal Deception Detection (pp. 15-19). ACM.

21
Department of CSE(AI&ML)