International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056
Volume: 12 Issue: 02 | Feb 2025 www.irjet.net p-ISSN:2395-0072
Deception Detection Using a Passive-Aggressive Classifier: A Novel Approach to Identify Deceptive Communication
1Rehamat Bee, 2Diya N, 3 Prof. L Sreenivasa Perumal
1,2 UG student, Dept. of CSE-AIML, AMC Engineering College, KARNATAKA, INDIA. 3Assistant professor, Dept. of CSE-AIML, AMC Engineering College, KARNATAKA, INDIA
ABSTRACT - This studypresents a passive-aggressive classifier for detecting fake news, leveraging its adaptability for real-time misinformation detection. Trained on labeled datasets, it identifies deceptive patternsefficientlywithoutfullretraining.Resultsshow its effectiveness in curbing misinformation and enhancingtrustindigitalcommunication.
Keywords: Fake news detection, Passive- aggressive classifier, Machine learning, Text classification, Misinformationdetection.
1. INTRODUCTION
The way we consume information has changed significantly in the fast-paced digital world of today. Social network 29. For millions of people, platforms, news websites, and online discussion boards are becomingtheirmainnewssources.Althoughthereare advantages to this accessibility, there has also been a concerning increase in False or misleading material that is passed off as real news with the intention of misleadingorcontrollingreadersisreferredtoasfake news. This phenomenon has the potential to have detrimental effects on public opinion, cause fear in times of crisis, and undermine confidence in reliable informationsources.
Thechallengeofidentifyingfakenewsiscompounded by the sheer volume of content generated every day. Traditionalmethodsoffact-checking,whichoftenrely on manual verification, are not only time-consuming but also impractical in the face of such overwhelming data. Automated solutions that can effectively and precisely identify misleading content in real time are thereforedesperatelyneeded.
This project, titled "Deception Detection Using a Passive-Aggressive Classifier," by leveraging machine learning techniques. At the heart of our approach is thePassive-AggressiveClassifier,apowerfulalgorithm designed for text classification tasks. This model effectively distinguishes between authentic and fraudulent news stories because it is especially wellsuitedforfindingpatternsintextualdata.
We allow the system to understand the nuanced differencesbetweentrueandmisleadingnewsarticles by training the passive-aggressive classifier on labelled datasets that comprise both types of articles. This classifier's capacity to adjust to new information without requiring a lot of retraining is one of its most notable properties, which makes it a reliable answer for the constantly changing world of internet disinformation.
2. PROBLEM STATEMENT:
The rapid spread of misleading and deceptive information, particularly through digital platforms, posesasignificantchallengetoindividualsandsociety. Traditional misinformation detection methods often require extensive retraining and struggle to adapt to evolving deceptive strategies. There is a need for an efficient and adaptive approach to identify deceptive communication in real time. This study addresses the problem by utilizing a passive-aggressive classifier, a machine learning model capable of handling largescale text classification tasks while dynamically adjusting to new data. The proposed solution aims to enhance the accuracy and efficiency of deception detection, thereby improving trust and reliability in digitalcommunication.
3. METHODOLOGY
Outline the methodology employed in our project, "Deception Detection Using a Passive-Aggressive Classifier." Our approach is structured around three main components: the passive-aggressive classifier, datacollectionandpre-processing,andmodeltraining and evaluation. Each of these components ensuring the effectiveness and reliability of our fake news detectionsystem.
A. Passive-Aggressive Classifier:
A potent linear model created especially for extensive text classification tasks is the PassiveAggressive Classifier. Its distinct learning strategy,
International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056
Volume: 12 Issue: 02 | Feb 2025 www.irjet.net p-ISSN:2395-0072
whichfocusesonupdatingitsweightssolelyforthose instances that are misclassified, is what distinguishes thisclassifier.Thisimpliesthatratherthancompletely reworking its learning process, the model makes a focusedchangetoincrease accuracywhenitdetectsa mistake. Rapid convergence, or the ability to swiftly learn from mistakes and adjust to new information, is madepossiblebythisfeatureofthepassive-aggressive classifier. This is particularly beneficial in the context of fake news detection, where the landscape of informationisconstantlyevolving.Asnewarticlesare published, the model must be able to adjust its understanding of what constitutes real versus fake news. The classifier's ability to perform online learning updatingitsparametersinreal-timeasnew data comes in makes it an ideal choice for this project.
B. Data Collection and Pre-processing
Thefoundationofanymachinelearningprojectliesin the quality of the data used. For our project, we sourced publicly available datasets that contain labeled news articles categorized as either "real" or "fake." These datasets were carefully selected to ensureadiverserepresentationofnewstopics,styles, and formats, which is crucial for training a robust model.
Once the data was collected, it underwent a series of pre-processing steps to prepare it for analysis. This stage is essential since the model's performance is directlyimpactedbythecalibreoftheinputdata.
Thepre-processingstepsincluded:
Text Cleaning: The first step involved removing any unwantedcharacters,punctuation,numbers,andURLs from the articles. This cleaning process ensures that the text is free from noise, enabling the model to concentrate on the important content.
Tokenization: The text was cleaned and then can analyse,tokenizationisnecessary.Weallowthemodel to comprehend the structure and meaning of the informationbysegmentingthetextintosmallerparts.
Stopword Removal: Words like "and," "the," and "is” don't significantly contribute meaning to the text, wereeliminated.Byreducingthedimensionalityofthe data, this stage enables the model to focus on more pertinenttermsthataid inthecategorizationprocess.
Vectorization: The final pre-processing phase involved transforming the cleaned and tokenized approach.Astatistical metric calledTF-IDF assessesa word's significance in a document in relation to a corpus,orgroupofdocuments.Weallowthemodelto efficientlyassessthe data andgenerate well-informed predictions by converting the language into a structuredformat.
C. Model Training and Evaluation
Afterpreparingthepre-processeddata,weproceeded to the passive-aggressive classifier's training phase. Teaching patterns linked to news stories is a crucial step in this process. The dataset was divided into two subsets at the start of the training process: one for testing and one for training. Generally, we employed an 80-20 split, in which the model was trained using 80% of the data and evaluation was done with the remaining20%.
BLOCK DIAGRAM
Fig-1: BlockDiagramofDeceptionDetectionusinga PassiveAggressiveAlgorithm.
Throughout the training process, the passiveaggressive classifier acquired the ability to recognize the characteristics that differentiate authentic news pieces from fraudulent ones. Through iterative training, the model's weights were modified in response to the misclassified cases it came across. With the help of this focused learning strategy, the model wasabletoincreaseitsaccuracyovertimeand becomemoreskilledatidentifyingmisleadingcontent.
We used a different testing dataset to assess the model's performance after training was finished. We usedanumberofimportantmetrics,suchasaccuracy, precision,recall,andF1-score,toevaluateitsefficacy.
International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056
Volume: 12 Issue: 02 | Feb 2025 www.irjet.net p-ISSN:2395-0072
• Accuracy quantifies the proportion of true findings (includingtruepositivesandtruenegatives)acrossall cases analysed, demonstrating the overall soundness of the model's predictions. • Precision evaluates how accurate the optimistic forecastswere.
4. IMPLEMENTATION:
Data Collection: Gather a dataset that contains deceptive and truthful communication. You could use text datasets like those from LIAR dataset or Fake News Detection datasets.
Pre-processing: Clean and pre-process the text data by removing stop words, stemming, or lemmatizingthetext.
Feature Extraction: Convert the text data into numerical features using methods like TF-IDF or Word2Vec
Modeling: Train the Passive-Aggressive Classifier toclassifythetextaseitherdeceptiveortruthful.
Evaluation: Evaluate the performance using standard classification metrics such as accuracy, precision,recall,andF1-score.
5. RESULTS:
Ourtestsofthepassive-aggressiveclassifiershowthat it is capable of identifying false information with a high degree of accuracy. The model continuously maintained an accuracy rate above 90% during the testing phase, demonstrating its remarkable capacity to discriminate between authentic and fraudulent news pieces. In the case of disinformation, when misclassification can have serious repercussions, this highdegreeofaccuracyisessential.
We assessed the model using a number of performancecriteria,includingasprecisionandrecall, inaddition tototal accuracy.Additionally,precision ametricthatquantifiesthepercentageoftruepositive predictions among all of the model's positive predictions surpassed90%.Thisindicatesthatwhen theclassifierlabelsanarticleasfake,itishighlylikely to be correct, thereby minimizing the risk of false positives. Similarly, recall, which assesses the model's ability to identify all relevant instances (in this case, fake news articles), also demonstrated impressive results, reinforcing the classifier's reliability in detectingdeceptivecontent.
OneofthestandoutfeaturesofthePassive-Aggressive Classifier is its lightweight nature, which contributes to the system's scalability. This characteristic allows themodeltoefficientlyprocesslargedatasetswithout significant delays, making it suitable for real-time applications. In an era where information spreads rapidly,theabilitytoanalyseandclassifynewsarticles in real-time is essential for combating the dissemination of fake news. Our experiments confirmed that the classifier could handle substantial volumesofdata,
Furthermore, the adaptability of the passiveaggressive classifier to evolving patterns of misinformation was evident during our testing. As new articles were introduced, the model demonstrated its capacity to adjust its predictions based on the latest data, highlighting its potential for continuous learning. This adaptability is particularly important in the fast-paced digital landscape, where the nature of fake news can change rapidly, and requiring detection systems to remain agile and responsive.
All things considered the outcomes of our tests demonstrate how well the passive-aggressive classifierworksasareliableinstrumentforidentifying false information. Its high accuracy, exceptional precision and recall, scalability, and adaptability position it as a significant instrument in the ongoing fight against misinformation. As we move forward, thesefindingspavethewayforfurtherenhancements and applications of the model, ultimately contributing to a more trustworthy digital communication environment.
Fig-2:Dashboard
International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056
Volume: 12 Issue: 02 | Feb 2025 www.irjet.net p-ISSN:2395-0072
Fig-3: PredictionPage
6. DISCUSSION
The results of our experiment demonstrate how important machine learning is in addressing the urgentproblemofdisinformationinthecurrentdigital environment.Theproblemofseparatingreliablenews from misleading content gets more difficult as the amount of information published online keeps growingatanexponentialrate.Weshowthatmachine learning can give a dependable and scalable solution to this problem by developing the Passive-Aggressive Classifier, which enables the automated detection of bogusnewswithremarkableaccuracy.
Thisdesign'salignmentwiththelargerfieldofNatural Language Processing (NLP) is among its most important accomplishments. Understanding and interpreting mortal language requires the use of natural language processing (NLP) ways, which are also pivotal for efficiently analyzing textual material. Our technology is a useful tool for businesses looking to save the integrity of information participated with the public since it can fete minor verbal patterns that distinguish false news from true news by exercising thesestrategies.
More importantly, our model's inflexibility in responding to changing patterns of intimation is relatively remarkable. A discovery system with realtimeliteracyandadaptationcapabilitiesisessentialin a world where false news styles are ever- evolving. This point guarantees that the model stays applicable in the face of arising difficulties while exploration supports the nonstop enterprise to promote trust in digital communication by offering a scalable and effective result for fake news discovery. Tools that
increasethetrustabilityofinformationismorepivotal than ever in a time when intimation can have major impacts,frompublicfeartothebreakdownofpopular processes.
In conclusion, the findings of this design not only emphasize the significance of machine literacy. As we continue to upgrade and expand our model, we hope to play a part in creating a more informed and sapient public, able of navigating the complications of the information age with confidence.
7.CONCLUSION
In conclusion, the" Deception Discovery Using a Passive- Aggressive Classifier" design successfully illustrates the important part that machine literacy wayscanplayinaddressingthepervasiveissueoffake newsinourdigitalsociety.Throughthecarefuldesign and perpetration of our system, we've created a scalable and effective result able of relating deceptive content in real time, thereby fostering lesser trust in digital communication. The emotional delicacy and rigidity of the unresitant-aggressive classifier punctuate its eventuality as a robust tool for combating misinformation, making it an inestimable asset for social media platforms, news agencies, and governmental associations likewise. As welook to the future, there are multitudinous avenues for improvement that could further bolster the system's capabilities. Exploring multilingual support would enable the model to operate effectively through different verbal surrounds, while the integration of advanced deep literacy models could enhance its understanding of complex language patterns and nuances.
8.REFERENCES
1) Choudhury, A., & Gupta, R.( 2023). Relating False News Using Enhanced Passive- Aggressive Classifier styles.IEEEAccess.Https//ieeexplore.ieee.org/docume nt/10489711
2) Khan, S., & Kumar, R. (2023). Fake News Discovery Using Recent Machine Learning Algorithms. IEEE Access. Https// ieeexplore.ieee.org/document/10403584
3) Sharma, A., & Singh, V. (2020). Fake News Discovery.UsingPassive-AggressiveClassifierandTFIDFVectorization
International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056
Volume: 12 Issue: 02 | Feb 2025 www.irjet.net p-ISSN:2395-0072
(4) Pang, M., & Lee, L.( 2023). A check of Fake News Discovery on Social Media Recent Advances and unborn Directions. ACM Computing checks, 55( 6), Composition124.
5) Sahu, P., & Padhy, N. P.( 2023). Fake News Discovery Using Deep Learning A Review. Journal of Information and Data Management, 14( 1), 1- 12
6) Alam, F., & Shabbir, J.( 2022). Using Natural Language Processing and Machine Learning for Fake News Discovery A Survey. Journal of Information Science,48(5),616-635.
7) Singh, R., & Gupta, N.( 2022). Fake News Discovery Using Machine Learning A Review and exploration Directions.ComputersandSecurity,119,103799.
8) Mishra, P., & Yadav, S.( 2022). A Comprehensive Study on Fake News Discovery ways Using Machine literacy. Expert Systems with Applications, 203, 117622.
BIOGRAPHIES
REHAMAT BEE student pursuing BE in CSE AIML branch 4th year in the esteem college AMC Engineering College Bengaluru Karnataka . Interned at DRDO CAIR.
DIYA N studentpursuing BE in CSE AIML branch 4th year in the esteem college AMC Engineering College Bengaluru. InternedatDRDOCAIR.
L.SREENIVASAPERUMAL ASSISTANTPROFESSOR AMCENGINEERINGCOLLEGE