FAKE NEWS DETECTION USING

MACHINE LEARNING

Mr.M.ALOK KUMAR, M.Tech.,

Assistant Professor
Department of Computer Science and Engineering
Government College of Engineering, Dharmapuri

PRIYADHARSHINI A SIVANANTHINI S.H

Department of Computer Science & Engineering Department of Computer Science & Engineering
Government College of Engineering, Dharmapuri Government College of Engineering,
Dharmapuri

SUJITHA G
Department of Computer Science & Engineering
Government College of Engineering, Dharmapuri

ABSTRACT

Long-Term Memory Network (LSTM) has performed well in sentiment analysis work. The usual way is to use
LSTMs to combine word embeddings for text representation.However, word embeddings contain more semantic
information rather than emotion information. Only It is wrong to use word embeddings to represent words in sentiment
analysis tasks. To solve this problem,We propose a vocabulary-enhanced LSTM model. The model uses the first sentiment
lexicon as additional information.Pre-training a word sentiment classifier and then obtaining word sentiment embeddings
including words The word is not in the dictionary. Can create words by combining sentiment embedding and its word
embedding Representation more accurate. In addition, we define a new method to find the attention vector in general
Sentiment analysis without target which can improve LSTM capability in capturing global sentiment Information The
results of experiments on the English and Chinese datasets show that in our model Comparable or better results than
existing models

Keywords: LE-LSTM,Sentiment lexicon, sentiment embedding, word embedding, sentiment analysis.

 3. Classification of Fake News
1.INTRODUCTION 3.1. preprocessing
3.2. feature extraction
Recently, there has been a rapid and widespread increase in 3.3. LE-LSTM Classification
fake news, defined Misinformation spread with the goal of 4. Prediction of Fake News
fraud.Fake news is widely regarded as one of the most 5. Performance Analysis
serious threats globally commerce, journalism and
In Figure 3.1, we have represented the architecture of
democracy, with significant collateral damage. It is false
the proposed approach
information or misleading information that can be
masqueraded as facts Influencing a person's opinion. It is
difficult for human to trace the news manually, even with
existence of all the topics shown on social
media.Therefore, we need an efficient way to help us
differentiate the false. Information of true people posted on
social media. In this project, we propose a lexicon-
enhanced LSTM model For sentiment analysis, which uses
prior sentiment information of words as supplementary
information to improve word quality Representation. To
improve semantic composition capability of LSTM, we
define a new system to calculate Attention vector in
sentiment analysis without a Target and take two particular
situations as examples. To verify our methods, we conduct
experiments on several English datasets and Chinese
datasets. Result Experiments show that our methods are
Figure 3.1: Architecture of proposed system
effective. Furthermore, we find that LSTM has strong
potential In handling short text sequences. The attention
mechanism we use only works on long text sequences.The 1. Fake News Spotter UI
analysis experiments also prove the results. In this module we have developed web based GUI for
Fake News Classification system to classify tweets as
2.LITERATURE REVIEW fake or genuine. Integrate with Trainer and Tester
modules Developed with Python and Flask Framework
The problem of fake news is not new, as misinformation is
being circulated for a long time.Newspapers and radio. 2. Preparation and Exploration of the FN Dataset
2.1. import dataset
Import data from the application or upload an Excel
file. It will be used to train you Emotion analysis
Fake news spreads faster because of the Internet Social media
model.
and blogs. This type of information can be harmful. like this
we need Be able to differentiate between fake and real news
[1]. This article introduced a Fake news detection system i.e. 2.2. read dataset
NoFake API using DL. In the dll, we have Lexicon Enhanced Reads the text to publish the purpose or purpose of the
- Long Short Term Memory (LE-LSTM) model proposed text
Who has achieved the best performance[2]. In addition, for the .
proposed false news Detection System, a real-time system to 2.3. Explore Datasets
be used to detect fake news Facebook, Twitter, Instagram and Data visualization tool, which brings all the data
other platforms[3]. together Easy view to follow.

3.PROPOSED SYSTEM
3. Classification of Fake News
In the proposed system, we propose a vocabulary-enhanced 3.1. preprocessing
LSTM model (LE-LSTM) to integrate Sentiment Lexicon A data standardization process called text, in order to
in LSTM to get more sentiment information of words by obtain consistent results Default, which removes non-
using Sentiment Lexicon To train a word sentiment alphanumeric characters In the text, there are some
classifier, we can do Get the sentiment embedding of each techniques commonly used in NLP:
word. Serialized Embedding as word and its feeling
LSTM's input can perform high in sentiment analysis work. Stop words: We have removed the words in the
agreement that they do not contribute to the samples
The proposed system is sub-divided into the following The learning process behind the problem; For example,
modules: articles and Proposals.

1. Fake News Spotter UI Stemming: This technique was used to reduce words to
2. Preparation and Exploration of the FN Dataset their roots.
2.1. import dataset
2.2. read dataset Tokenization and padding: As usual in text
2.3. Explore Datasets processing tasks, we do tokenization And when
necessary, padding for the representation of words and Figure 4.1,Training and Validation
sentences accuracy

3.2. feature extraction

In Figure 4.2, the training and validation loss of
WordCloud, which provides the number of instances of Lexicon Enhanced is shown -LSTM algorithm
each word in the text corpus; Periodic frequency-inverse
document frequency (tf-idf), which provides a weighted
scale.The importance of each word in the text (according to
its frequency)Sentences;And pre-trained word embedding,
including words and semantic relationships They are
referred to as vectors.

3.3. LE-LSTM Classification

In this module, we have implemented the five-density
Lexicon upgrade LSDM model Layers with a block
normalization process for categorizing fake messages. LE-
LSTM: Lexicon layer, units (units) in hidden layers [this
model only Single hidden layer], kernel regulator (KR),
recurrent regulator (RR), Abandonment (D). Fake - 0, real -
Figure 4.2,Training and Validation loss
1 are classified into two types.

4. Prediction of Fake News

5.CONCLUSION AND FUTURE DEVELOPMENT
After training your model with some examples, you should
name it. Then you can Check it out with new text to see In this digital age where false news is everywhere Where
how it is classified.In this module, users can log in to the in digital platforms, there is an absolute need to detect
Fake News Spotter UI and post messages. And those fake news and this model serves its purpose by becoming
messages Dummy contents were tested with trained a necessity of our tool.Fake news on sensitive topics
dummy news databases. creates toxic environment on the web. Fake news
detection is the analysis of socially relevant data to
5. Performance Analysis determine whether it is genuine or not.to make a fool of.
Here in this paper we have compared different methods
The ratio of the sum of accuracy, true positive (TP) and
like BagOfWords (BoW), Ngrams, TF‐IDF, Naïve Bayes
true positive (TP) and false-positive cases (FP): Recall the
etc.LSTM is the most effective of all we use Various
ratio of the sum of true positive (TP) and true positive (TP) techniques like stop word removal, word embedding and
and false negative cases (FN):The F1-rating is the overall how LSTM can be used to achieve better results. The
rating and weighted average in terms of accuracy and model outlined in this paper is very effective, as well as
recall. Indicates "1" The best score is "F1", "0" is the worst compared to the existing system The model proposed
score. Accuracy is the ratio of the total number of samples here gives better results with an accuracy of 93% which
to the number of samples correctly classified. AUC (lower is very promising, we can further enhance the result by
part of the curve) is an evaluation indicator used for the increasing the training data.
binary classification model,Indicates that the positive
In the future we plan to implement recent multimodel-
model and the negative model are approximately selected based methods Pre-trained word embeddings to handle
4.RESULT: visuals (ie Elmo, XLNet, etc.) Information such as videos
and images.
In this result section, we have shown Developed a model
for detecting fake messages using the Lexicon-enhanced 6.REFERENCES
LSTM algorithm.
• N. R. de Oliveira, P. S. Pisa, M. A. Lopez, D. S. V. de
In Figure 4.1, the training and validation accuracy of the Medeiros, and D. M.F. Mattos, ``Identifying fake news
lexicon indicates Enhanced - LSTM algorithm. on social networks based on natural language processing:
Trends and challenges,'' Information, vol. 12,
no. 1, p.38, Jan. 2021.

• J. A. Vijay, H. A. Basha, and J. A. Nehru, ``A dynamic

approach for detecting the fake news using random forest
classifier and NLP,'' in Computational
Methods and Data Engineering. Springer, 2021, pp. 331
341.

• M. N. Nikiforos, S. Vergis, A. Stylidou, N. Augoustis,

K. L. Kermanidis, and M. Maragoudakis, ``Fake news
detection regarding the Hong Kong events from
tweets,'' in Proc. Int. Conf. Artif. Intell. Appl.
Innov. Greece: Springer, 2020, pp. 177-186.
• S. Kumar, R. Asthana, S. Upadhyay, N. Upreti, and M.
Akbar, ``Fake news detection using deep learning models: A
novel approach,'' Trans. Emerg. Telecommun.
Technol., vol. 31, no. 2, p. e3767, Feb. 2020.

• G. Sansonetti, F. Gasparetti, G. D'Aniello, and A.

Micarelli, ``Unreliable users detection in social media: Deep
learning techniques for automatic detection,''
IEEE Access, vol. 8, pp. 213154-213167, 2020.

• S. Ochoa, G. D. Mello, L. A. Silva, A. J. Gomes, A. M. R.

Fernandes, and V. R. Q. Leithardt, ``FakeChain: A
blockchain architecture to ensure trust in social media
networks,'' in Proc. Int. Conf. Qual. Inf. Commun.
Technol.Algarve, Portugal: Springer, 2019, pp. 105-118.

• Z. Shae and J. Tsai, ``AI blockchain platform for trusting

news,'' in Proc.IEEE 39th Int. Conf. Distrib.
Comput. Syst. (ICDCS), Jul. 2019, pp. 1610-1619.

• S. Paul, J. I. Joy, S. Sarker, H. Shakib, S. Ahmed, and A. K.

Das, ``Fake news detection in social media using
blockchain,'' in Proc. 7th Int. Conf.Smart Comput.
Commun. (ICSCC), Jun. 2019, pp. 1-5.

• J. Golbeck, B. Auxier, and V. Kori, ``Fake news vs satire: A

dataset and analysis,'' in Proc. 10th ACM Conf.
Web Sci., 2018, pp. 17-21.

• C.-S. Atodiresei, A. T n selea, and A. Iftene, ``Identifying

fake news and fake users on Twitter,'' Proc.
Comput. Sci., vol. 126, pp. 451-461, Jan. 2018, doi:
10.1016/j.procS.2018.07.279.

• M. Arquam, A. Singh, and R. Sharma, ``A blockchain based

secure and trusted framework for information propagation on
online social networks,''2018, arXiv:1812.10508.
[Online]. Available: http://arxiv.org/abs/1812.10508