This repository is focused on using BERT (Bidirectional Encoder Representations from Transformers) for bot detection on Twitter data. BERT is a deep learning model that is commonly used in natural language processing tasks such as text classification, question answering, and language translation.
The repository includes code for training the BERT model, and evaluating its performance. The code is written in Python and uses popular machine learning PyTorch library.
Overall, the Git repository provides a useful resource for anyone interested in using BERT for bot detection on Twitter data, with code, explanations, and results all available in one place.
In this approach, we use the popular T5 model by google and frame the Bot classification problem as a generative task. We maek use of the popular library for instruction tuned T5 modelling (InstructABSA) which has a T5Classifer inbuilt API.
As part of the T5Classifier, we insturction tune the model by adding the definition of task followed by 2 positive, negative and neutral examples to every sample in the training set to provide more contextual information.
The notebook can be found here.
In this approach, we use two models. Logisitc Regression and SVM models. We use TF-IDF to vectorize text data and then convert it into a supervised classification task. The notebook can be found under SVMClassification.
Libraries needed to run this are present in requirements.txt
| Model Name | Accuracy |
|---|---|
| Bert Classifier | 89.73% |
| TkInsturct Classifier | 51.56% |
| SVM Classifier | 70.90% |
[1] Scaria, K., Gupta, H., Goyal, S., Sawant, S. A., Mishra, S., & Baral, C. (2023). InstructABSA: Instruction Learning for Aspect Based Sentiment Analysis. arXiv preprint arXiv:2302.08624.
[2] Cheng-Caverlee-Lee September 2009 - January 2010 twitter scrape . Internet Archive. (2010, November 7). Retrieved from https://archive.org/details/twitter_cikm_2010
[3] Dukić, D., Keča, D., & Stipić, D. (n.d.). Are you human? detecting bots on Twitter using Bert | IEEE conference ... Retrieved from https://ieeexplore.ieee.org/document/9260074/
[4] Feng, S., Wan, H., Wang, N., Luo, M., & Li, J. (2021, August 27). TwiBot-20: A comprehensive twitter bot detection benchmark - arxiv. Retrieved from https://arxiv.org/pdf/2106.13088v4.pdf
[5] Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks (Wang et al., EMNLP 2022)
[6] BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (Devlin et al., NAACL 2019)