11 V May 2023

https://doi.org/10.22214/ijraset.2023.52930
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue V May 2023- Available at www.ijraset.com

Social Media Spam Comments Detection Analysis

using Machine Learning
Vaibhav Ambhore1, Adwait Garud2, Harshal Lande3, Pratap Rede4
D. Y. Patil College of Engineering Ambi, Pune-410507

Abstract: The results published by Google's famous video distribution platform Social Media are attracting more and more users
every day. However, this success has also attracted many malicious users who try to promote their videos or infect them with
viruses and malware. As we know Social Media has a tool to measure comments so spam is increasing very fast which is why the
owner's comment section is closed. Creating a classification system for automatic spam filtering is very difficult because words
are short and often contain slang words, symbols, and abbreviations. In this article, we consider several best practices for
detecting and analyzing spam messages.
Analysis of the results showed that decision trees, logistic regression, Bernoulli Naive Bayes, Random Forest, Linear and
Gaussian SVMs were equal to the highest values, with 99.9% confidence. That's why it's important to find a way to check video
comments and ads before they're viewed by unsuspecting users.
Keywords: Machine learning, linear regression, random forests, prediction.

I. INTRODUCTION
In an era dominated by social media, platforms such as Facebook, Instagram, Twitter and YouTube have become hotspots where
users communicate, discuss and express themselves. However, the growing popularity of these platforms has also attracted many
spam comments, making it difficult to maintain a positive and engaging online presence. In order to solve this problem, the use of
machine learning techniques in the detection of spam messages has attracted attention.
Machine learning algorithms have proven useful in many areas, including natural language processing (NLP) and pattern
recognition. Using the power of machine learning, researchers and developers have sought to create efficient and accurate models
that can identify and filter spam on social media platforms.
The purpose of this analysis is to explore the use of machine learning techniques in detecting spam comments. By analysing the
characteristics and patterns of spam messages, we aim to build a model that can effectively analyse and filter out spam content.
Analyzing spam comments involves processing a lot of user-generated data, which presents unique challenges. Legislation based on
methods or concepts is briefly compared in relation to the evolution of spam technology. Machine learning, on the other hand, is
more flexible and scalable.
In this review, we will explore various machine learning algorithms such as support vector machine (SVM), decision trees, random
forests, and deep learning models to generate the proposed spam detection. We will leverage the power of NLP techniques,
including content extraction, text classification and sentiment analysis, to train and evaluate our models.
Data used for training and assessment may contain comments, including spam and suggestions from social media platforms. Using a
suitable pre-processing method, we will clean the raw data and transform the model into a suitable one for training.
Evaluation of the design will be based on criteria such as accuracy, precision, recall and F1 score to evaluate their performance in
distinguishing spam from legitimate messages.
As spam analysis seems to be in the minority, we will also explore strategies for dealing with data inconsistencies.
The results of this analysis will give you a better idea of the performance of machine learning algorithms in detecting spam on social
media platforms. By creating the right model, we can help platform administrators and moderators get spam filters, thereby
improving user experience and reducing the occurrence of inappropriate content language.
Overall, this review aims to contribute to research and techniques that use machine learning to combat spam messages on social
media platforms. By leveraging the power of Naive Bayes algorithms and NLP technology, we strive to create a safe and effective
online experience for our users.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 5852
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue V May 2023- Available at www.ijraset.com

II. EXISTING SYSTEM

Over the years, machine learning has been at the forefront of a lot of research on online spam or spammer or spam account
detection.
Researchers used different techniques when designing spam detection methods using machine learning techniques. While
researchers work to detect spam, spammers work hard to avoid detection. Spam comments are defined as comments that contain
advertisements or appear to be unrelated to the content of a video. Many algorithms have been used before to identify spam, for
example CNN (Convolutional Neural Network).
We use Naive Bayes classifier and Support Vector Machine classifier (SVM) in the planning process. Naive Bayes is used to
calculate the number of possibilities, which means that using spam money as an example results in spam money. SVMs is used for
data classification.

III. PROBLEM STATEMENT

To develop a system which is capable of performing following actions:
1) Training System on a Test Dataset in order to Recognize spam Comments: The training data is the largest (in terms of size)
subset of the original data used to train or fit the machine learning model. First, the training data is fed into machine learning
algorithms, allowing them to learn how to predict a task.
2) Test System for Test Data: After training the model with the training data, it's time to evaluate the model with the test data. This
information evaluates the performance of the model and enables the model to adapt to new or unknown data. Test data is
another primary dataset that is independent of the training data.
3) Test Based on Real Spam Files: The active spam testing system receives spam specifications from testers, which will be pulled
into the search engine for testing. This test can generate information to help track the signs of spam to be reported.

IV. IMPLEMENTATION STRATEGY

1) Create an infographic using 4 out of 5 Social Media files.
2) Since the dataset is a data file, attributes must be provided by the (CONTENT) field.
3) It is important to bring properties from the document only to the "CONTENT" field in the document.
4) Remove unwanted special characters () from the response.
5) Build a spam dictionary of words commonly found in spam by looking at spam patterns.
6) Use the spam dictionary to count spam messages in messages.
7) To check if the comment contains strings "http “,"www" or ".com" string which represent promotions and could be SPAM and
set IS_HTTP=1 else
8) To calculate the ratio of spam to the number of words in the message, first remove the English words "STOPWORDS" such as
(I, me, etc.) from the replies
9) Find out the length of the message, because long messages are usually spam9. Count the number of words in the review after
removing the stop words.
10) Calculate the ratio of spam messages to all messages.
11) Use Naive Bayes classifier on test data and validate with test data

V. PROPOSED SYSTEM
We try to detect spam messages using some traditional heuristics, such as the traditional machine learning algorithm Naive Bayes,
as well as N-Grams, which has proven effective in searching and after fighting spam. We have collected and created five databases
containing real, public and uncensored data directly from Social Media via API7. We've selected five of the ten most-watched
YouTube videos. Each sample represents the word spoken in the speech of each selected video. No previous procedures were
applied.
Each sample is then tagged as spam or legal (raw) using a common tool called tagging created for this purpose. The model is
associated with metadata such as the stored name and publication date.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 5853
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue V May 2023- Available at www.ijraset.com

VI. USE CASE VIEW

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 5854
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue V May 2023- Available at www.ijraset.com

VII. ALGORITHMS USED

A. Naive Bayes
Naive Bayes is a classification algorithm among machine learning algorithms. It is used only to solve the problem of classification
of texts with high educational content. It is a probability-based classifier. Probability includes the probability that an event will occur
based on experience. The algorithm is known for its simplicity and efficiency.
Bayes’ theorem, also known as Bayes’ rule or Bayes’ rule, is used to determine the probability of a previously informed hypothesis.
Depends on what it is.

The formula for Bayes' theorem is given as:

P(A) is Prior Probability: Probability of hypothesis before observing the evidence.
P(B) is Marginal Probability: Probability of Evidence.

B. C4.5
C4.5 is a pedigree machine learning algorithm. The algorithm uses data entropy to build a tree from the given data. We deliberately
analyze and classify data from large collections and yield similar results.
C4.5 is an extension of the ID3 algorithm. The algorithm uses the obtained data to divide the sample data into a series of subsets,
choosing the data attributes for each node, and the highest value is determined. This process reverts to the smallest sublist on the
other hand, if you are dealing with environmental data such as weather data or historical crop data, you can use RNN-based models
such as Long Short Term Memory (LSTM) or Gated Recurrent Unit. (GRU)) may be more appropriate. RNNs are designed to
capture trends in data, making them ideal for time series analysis tasks.

VIII. CONCLUSION
Social media has become popular, creating an opportunity for spammers to spread unwanted messages.
Previously, some machine learning algorithms were used for this detection. In the planning process, we also use advanced machine
learning algorithms with advanced features and use them to compare the results of various algorithms. We create features based on
features derived from user profiles and the content they share. Based on the test results, it can be assumed that existing products
widely used in the data mining community can use these features to identify spammers.

IX. PROJECT SCOPE AND FEASIBILITY

1) Social Media is a large video sharing network with a large user base. It is not possible to follow such a large-scale project at our
level. However, the project was carried out on some limits. This work can be extended to more users and larger data.
2) A feasibility study is an evaluation of the effectiveness of a proposal or process.
The project needs information that can be completed on time. The information needed by the Theis project has been posted in the
public domain. The project is possible because all the components required for the project are available

REFERENCES
[1] K. S. Adewole, N. B. Anuar, A. Kamsin, K. D. Varathan and S. A. Razak, “Malicious accounts: Dark of the social networks”, Elsevier, 2017, pp. 41-67.
[2] A. Gupta and R. Kaushal, “Improving Spam Detection in Online Social Networks”, IEEE, 2015.
[3] M. Verma, Divya, S. Sofat, “Techniques to Detect Spammers in Twitter – A Survey”, International Journal of Computer Applications, January 2014, Vol. 85,
No. 10, pp. 27-32.
[4] C. Chen, Y. Wang, J. Zhang, Y. Xiang, W. Zhou and G. Min, “Statistical Features- Based Real- Time Detection of Drifted Twitter Spam”, IEEE Transactions,
April 2017, pp.914-925

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 5855