Comments & Twitter accounts gRPC classification service.
There are two distinct classification services a) account classification and b) comment classification.
AccouncClassifier uses a pre-trained model created by the Civic Information Office from 2016 until 17/07/2021, with almost 12,000 handpicked accounts in the open-source version and more than 25,000 users in the production version, of which 7,000 (fake/genuine) accounts from Botometer.
In our approach we investigated the behavioural characteristics that differentiates normal accounts with "amplifiers" without addressing the binary logic -bot or not- instead, classifing accounts as Influencers, Active, Amplifier, and Unknown. We use the CatBoost Classifier to predict user classes Refer to Twitter Accounts - CatBoost Classifier notebook for more details.
CommentsClassifier uses multiple pre-trained models and can be extented with any fintuned model supported by Huggingface's transformers module. By default we use detoxify-original for english comments, detoxify-multilingual for italian, french, russian, portuguese, spanish and turkish and out own comments-el-toxic for greek comments.
docker run --rm -it --name rtaa-classifier -p 50052:50052 cvcio/rtaa-72-rtaa-classifier:latest# clone the repo
git clone [email protected]:cvcio/rtaa-classifier.git
cd rtaa-classifier
# create the virtual environment (ex. with cobda)
conda create --name rtaa-classifier python=3.9
# install poetry package manager
pip install poetry
# install dependencies
poetry install
# run the service
make serveIt is recommented not to re-generate thh stubs if you want to have the latest version. Alternatevely you must clone the rtaa-72 repo and set the PROTO_PATH environment variable to that path. To build the protos, we use buf. Please refer to buf's documentation on how to install it, alternatevely you can run:
# install buf
make buf-install
# build protocol buffers
# the command will download the latest version of github.com/cvcio/proto,
# generate the stubs, and finally clean up the leftovers.
make proto If you're new to contributing to Open Source on Github, this guide can help you get started. Please check out the contribution guide for more details on how issues and pull requests work. Before contributing be sure to review the code of conduct.
In general, we are making this software publicly available for broad, noncommercial public use, including academics, journalists, policymakers, researchers and the public in general.
If you use this service, please let us know at [email protected].
See our LICENSE for the full terms of use for this software.