Social bots detection in Brazilian presidential elections using natural language processing

Ferreira, Gabriel Estavaringo; Santos, Bianca Lima; Ó, Marcelo Torres do; Braz, Rafael Rodrigues; Digiampietri, Luciano Antônio

doi:10.1145/3466933.3466991

Cited by 5 publications

(3 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Revista Brasileira de Sistemas de Informac ¸ão (iSys: Brazilian Journal of Information Systems) https://sol.sbc.org.br/journals/index.php/isys 12:18 was superior to the use of each technique individually, shown in Table 6. When compared to the work which this one extends [Ferreira et al 2021], the performance improvement of the MLP classifier is noticeable, which could not detect any bots, but now has a recall of 0.38 and accuracy of 0.54. It is noteworthy that MLP needs a lot of calibration of hyperparameters to reach its maximum performance, a task not performed in the scope of this work, since the intention was to compare the different algorithms and feature sets in a balanced way.…”

Section: Combined Featuresmentioning

confidence: 72%

“…This paper used a training set of 800 Twitter profiles manually labeled as bot or human to train five different classifiers using 10-folds cross-validation. In [Ferreira et al 2021], we included some features extracted from textual representations of user posts to improve the classification. The present work is an extension of this previous work that used NLP, by including features extracted from a pre-trained BERT model, the state of the art in a textual representation.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Comparison of natural language processing techniques in social bot detection on Twitter during Brazilian presidential elections

Santos

Ferreira

et al. 2022

iSys

View full text Add to dashboard Cite

Currently, there are thousands of social bots acting on different online social networks. Identifying them automatically is a computational challenge.This work uses different natural language processing methods to extract features from tweets collected during the 2018 Brazilian presidential election period in order to make the bot detection process more precise. The developed solution uses artificial intelligence techniques, combining feature selection and classification algorithms.The authors obtained the best results through a union of all the extracted features using the Random Forest classifier, achieving an precision of 0.86 for the bot class and AUC of 0.86.

show abstract

Section: Combined Featuresmentioning

confidence: 72%

Section: Related Workmentioning

confidence: 99%

Comparison of natural language processing techniques in social bot detection on Twitter during Brazilian presidential elections

Santos

Ferreira

et al. 2022

iSys

View full text Add to dashboard Cite

show abstract

“…Pada salah satu penelitian dijelaskan bahwa algoritma random forest classifier dapat digunakan sebagai pendeteksi bot. Hasil menunjukkan bahwa algoritma random forest yang dikombinasikan dengan word2vec memiliki kinerja sebesar 81% [12]. Dalam penelitian lainnya kombinasi algoritma random forest dengan word2vec digunakan untuk menganalisa ulasan para pengguna maskapai guna meningkatkan pendapatan maskapai.…”

Section: Pendahuluanunclassified

Analisa Kinerja Algoritma Random Forest Classifier dengan Mutual Information dan Skip-Gram pada Klasifikasi Jurnal INIS

Karimah,

Hindasyah,

Taryo

2023

JIMP

View full text Add to dashboard Cite

Abstrak-Supervised learning adalah teknik yang bergantung pada masukan berlabel untuk mempelajari suatu fungsi dan menghasilkan keluaran yang sesuai apabila diberi data baru tanpa label. Penggunaan algoritma supervised learning sering dibutuhkan dalam berbagai kondisi, salah satunya yakni mengklasifikasikan dokumen. INIS adalah salah satu perpustakaan digital yang dianggap masih melakukan pengklasifikasian dokumen secara manual dan membutuhkan pengotomatisasian klasifikasi dokumen. Hal ini mengakibatkan proses pengelompokkan memakan waktu yang relatif lama dan terdapat banyak kendala karena banyaknya jumlah dokumen. Penelitian ini memiliki tujuan utama untuk menentukan algoritma yang memiliki kinerja dan akurasi terbaik agar dapat diimplementasikan dalam proses pengklasifikasian dokumen ilmiah. Penelitian ini menggunakan kombinasi algoritma Random Forest (RF) dengan Skip-Gram (SG) dan Mutual Information (MI) sebagai metode ekstraksi fitur dan metode seleksi fitur. Hasil menunjukkan bahwa kumpulan data yang digunakan dapat mempengaruhi kinerja suatu algoritma. Selain itu pada penelitian ini menunjukkan bahwa presentase kinerja dari algoritma Random Forest dan Skip-Gram lebih baik jika dikombinasikan dengan seleksi fitur daripada tidak menggunakan seleksi fitur. Penggunaan seleksi fitur pada Random Forest dan Skip-Gram dalam penelitian ini juga menunjukkan presentase kinerja lebih stabil jika dibandingkan dengan kinerja Random Forest dan Skip-Gram tanpa fitur seleksi. Kata kunci : Text Mining; Klasifikasi Dokumen; Pengklasifikasian Random Forest; Skip-gram; Mutual Information Abstrak-Supervised learning is a technique that relies on labeled inputs to learn a function and produces the appropriate output when given new data without labels. The use of supervised learning algorithms is often needed in various conditions, one of which is classifying documents. INIS is one of the digital libraries that is considered to still perform manual document classification and requires automating document classification. This causes the grouping process to take a relatively long time and there are many obstacles due to the large number of documents. This research has the main objective to determine the algorithm that has the best performance and accuracy so that it can be implemented in the process of classifying scientific documents. This study uses a combination of the Random Forest (RF) algorithm with Skip-Gram (SG) and Mutual Information (MI) as a feature extraction method and feature selection method. The results show that the data set used can affect the performance of an algorithm. In addition, this study shows that the performance percentage of the Random Forest andSkip-Gram algorithms is better when combined with feature selection than not using feature selection. The use of feature selection in Random Forest and Skip-Gram in this study also shows a more stable percentage of performance when compared to the performance of Random Forest and Skip-Gram without feature selection.

show abstract

Political communication on social media in Latin America: unequal use of Twitter by members of parliament

Haman

2023

EPI

View full text Add to dashboard Cite

This article focuses on the under-researched topic of the use of Twitter by members of parliament (MPs) in Latin America. There have not yet been any thorough comparative studies on this topic, and the majority of publications on the region focus on single case studies. Previous studies have primarily concentrated on presidents, or in the case of MPs, on specific case studies. A total of 2,353,138 tweets were accessed via the Twitter API, and 3,215 MPs were examined. Regression models and correlations were used to answer research questions, and the main variables examined concerned individual characteristics of MPs (gender and age) and socioeconomic indicators of the country (number of people on Twitter, internet access, Human Development Index - HDI). As a result, this paper offers a report on how MPs in Latin America are currently utilizing Twitter. While his social network is used by more than 90% of MPs in some countries (Argentina, Colombia, Ecuador, El Salvador, Peru, and Uruguay), there are countries where fewer than half of MPs use it (Bolivia, Honduras, and Nicaragua). The results show that female MPs are more likely than male MPs to use Twitter. In addition, Twitter is being adopted more by younger MPs. Other results show that country characteristics such as internet penetration, Twitter population, and the HDI are significant predictors regarding the adoption and use of Twitter by MPs. These results are consistent with assumptions based on cost-benefit calculus. Thus, it does not make as much sense for politicians to adopt Twitter in countries where there are less people on Twitter and low internet penetration. In particular, if politicians want to be elected or inform citizens about their activities, they have an opportunity to reach voters through Twitter. However, this is only true if Twitter is used in their countries.

show abstract

Social bots detection in Brazilian presidential elections using natural language processing

Cited by 5 publications

References 11 publications

Comparison of natural language processing techniques in social bot detection on Twitter during Brazilian presidential elections

Comparison of natural language processing techniques in social bot detection on Twitter during Brazilian presidential elections

Analisa Kinerja Algoritma Random Forest Classifier dengan Mutual Information dan Skip-Gram pada Klasifikasi Jurnal INIS

Political communication on social media in Latin America: unequal use of Twitter by members of parliament

Contact Info

Product

Resources

About