With the rise of social media use among younger generations, it is important to ensure that the content they are exposed to is appropriate and free from harmful language.
KidFriendlySocial is a web application designed to promote safe online social interactions for children. With the increasing use of social media by children, it is essential to provide a platform that can filter out inappropriate language in their tweets, thereby promoting a healthy online environment.
KidFriendlySocial addresses this need by utilizing a predictive model and a generative model built using machine learning techniques to identify tweets that contain bad language content. The app flags these tweets, alerting the user of the potentially harmful content, and then offers recommendations on how to rephrase their tweets that are more kid-friendly or appropriate for children. This feature helps children become more aware of what is acceptable and unacceptable online behavior, making them better digital citizens.
Moreover, the recommendations aspect is particularly useful, as it helps children learn how to express themselves appropriately while still being creative with their language. This feature will help build their writing skills and promote a positive online environment for children.
- Backend development : Python for building the whole backend application
- Microservices: Used to separate the functionality of the application into smaller, more modular components, improving scalability and maintainability.
- Frontend development : HTML and CSS for the frontend
- NLP models: Distil-Roberta for bad language detection and GPT-3 for generating recommendations
- Deep learning : Pytorch framework for building the NLP deep learning model
- Web framework: Flask for building the web application
- APIs : Twitter api to obtain real time data
- Containerization: Docker for packaging the application and dependencies
- Container orchestration: Kubernetes for deploying and managing containers
- Cloud hosting: Digital Ocean for hosting the application in the cloud
- CI/CD : Github actions for continous integration and continous delivery
- Database: MYSQL as the database
- Version Control: Git for code version control and DVC for data and model version control
- Testing: Pytest for unit and integration testing
- Monitoring: Prometheus for monitoring metrics and Grafana for visualizing them
The model was built using the state-of-the-art natural language processing algorithm, DistilRoberta, to detect and classify bad language content. The app also leverages OpenAI's GPT-3 to generate recommendations for users to improve their tweets when bad language is detected.
The application was built using Flask, a lightweight Python web framework, and containerized using Docker to ensure that it can be deployed easily on any infrastructure. The app is hosted on DigitalOcean Kubernetes, a highly scalable and efficient platform that can handle a large volume of traffic. The application is also monitored using Prometheus and Grafana to ensure optimal performance, and all data is stored and managed in a MYSQL database also hosted on digital ocean
KidFriendlySocial also incorporates several MLOps best practices, including version control of models and data, continuous integration and continuous deployment (CI/CD), and automated testing. The project follows the agile methodology, allowing for continuous feedback and improvement.
Overall, KidFriendlySocial provides a valuable service to users who want to ensure that their tweets are appropriate for children, by leveraging cutting-edge machine learning algorithms and best practices in software engineering, the application is highly scalable, efficient, and reliable.
The data for this project was obtained from two distinct sources, firstly from available datasets on the internet, and from data extracted from the twitter api. The data available on the internet was cleaned and put into a format that is usable, and it was already labelled, whiles i used doccana, an open source data labelling platform to label all the data extracted from the twitter api. The data from twitter api is very relevant in order to obtain latest data from twitter to use to train the data. An example of the way i used doccana to label the dataset is shown below:
Relational databse, MYSQL, is the database of choice to store all user details related to authentications and user feedbacks in order to score the model in production and also to add actual tweets from users to use as training data to improve the model in the next version. The MYSQL database was hosted on digital ocean to ensure high availability.
Firstly, wanted to see the most common unique words(unigram) that appeared in tweets that contain bad language:
As an be seen there are some words that are more general such as want, human etc, these are included here because they were used in contexts of tweets that are not kid appropriate but the words themselves are appropriate but the words that are strictly for bad language can be seen in there such as fuck, stupid, niggers etc.
Also, it was ideal to visualize a word cloud of tweet that contain good language and ones that contain bad language, to see how different they are:
As can be seen above, there is a clear difference in the good and bad language tweets, which is useful for us to differentiate and build a model to try and classify tweets according to this.
The distil Roberta model which has been pretrained on a huge corpus of tweets(model name: vinai/bertweet-base) was used as the base model of choice for training, because the model already understands how tweets are constructed so it can easily generalize to our tweets, and also since its a distil roberta model, its a smaller model that still has excellent accuracy, and most importantly, its smaller in size, so it trains faster and the inference speed(latency speed) is smaller. 4 metrics were used ( accuracy, recall, precision and f1), but the most important metric here is recall, as the key is to identify as many potentially bad language tweets as possible to protect children. Here, the focus is on minimizing false negatives(the situation where the tweet contains bad language but the model predicts it to contain good language). Its okay to allow flag some good language as negative, but its not good to allow bad language to get out there, so reducing this situation(higher recall) is the most important here. However, precision is also important to ensure that the recommendations provided by the application are accurate and trustworthy. Below are the metrics obtained:
The performance seems good, with the recall being almost 95%, which suggests the model does well in catching bad language tweets. To visualize all these metrics, the confusion matrix was used, the result is below here:
As can be seen, a lot of bad language tweets were correctly predicted with only 376 out of 5483(false negatives) incorrectly predicted, also for good language tweets, only 336 were wrongly predicted as being bad language tweets. This shows the model would be robust in predicting if a tweet contains bad language or good language. They key here is to try to reduce the 376 false negatives to capture more bad language tweets, as thats the one we are most interested in predicting.
GPT3 was used for generating the recommendations when bad language is detected, the temperature(how random the generations should be) was set to 0.8, the max length to 256, to allow users to express themselves more, text davicni 003 which was the most advanced model at the time of building was used, and an excellent prompt written to generate the recommendations. The GPT3 was finetuned using 150 tweets of bad language-good language pairs to enable the model to understand the context a little better to generate relevant recommendations.
Pytest was used for performing all the tests in this project for both unit tests and integration tests. The testing is essential to catch any bugs or errors before pushing the app to production. It ensures a correct and robust app that works exactly how we want to and the users having a seamless experience. The tests written were in three parts, the authentication aspect, the data aspects and the model aspects. The authentication tests ensures all aspects of user authentication, such as signing up and logging in, works correctly without any issues. The data aspects tests for correctness and validity of the data to ensure the right format of data is being used and accessed. The model testing tests the models performance to ensure its given out correct predictions, quality predicttions and the predictive model's integration with the generative model is correct and its working well. Various tests in these catgories were written(in all about 12 tests) and added to the CI/CD pipeline, that ensures all the tests passes before any updates or changes are pushed to production.
The model was deployed using docker and kubernetes and hosted on digital ocean. The authentications(signup, login) was written using flask, and jwt authentication used to authenticate users to use the app and all the passwords hashed using bcrypt to protect user privacy. The app was made available to 11 diverse users for a week, to interact with the app, in order to get feedback on how to improve the app, and also to use their feedback to score the performance of the model and improve the model.
Below is a demo of how the app works. PS: i had to take the app down, since it costs money to host the app on digital ocean(the docker containers, the kubernetes clusters and the mysql database).
Demo of model predicting good language:
Demo of model predicting bad language:
As can be seen in the demos, the good language prediction is straight forward, the focus is on detecting the bad language, the model correctly detects that the tweet contains bad language that is not useful for kids, and generates appropriates recommendations for them to use to improve their tweets.
The image above shows the recommendations generated, when bad language is detected, so in the demo, the actual tweet was :
- "piss off, you fking b*ch, get the hell of here",
- "Go away, you awful person, get out of here"
- "Go away, you meanie, scram!"
- "Beat it, you nasty individual, leave now!"
As can be seen the recommendations contain the same meaning and context as the original tweet, but it is a more kidfriendly version and contains no bad language, which is what we seek to accomplish.
The performance of the model and the system was monitored using Prometheus and grafana and digital ocean internal monotring tool. System metrics such as server response time, CPU and memory metrics were monitored using digital ocean internal monitoring tool, and the model metrics monitored using Promethues and Grafana.
The performance of the model was based on user feedback, as they are the ones the app is for, so their feedback is the most important to know how well the model is doing. As already explained, the app was made available to 11 unique individuals to use who performed 43 predictions, and these were the metrics monitored:
-
Total number of feedbacks received: This can give an indication of the overall engagement with the app.
-
Number of feedbacks categorized as bad language: This helps to track the frequency of bad language .
-
Number of correct bad language predictions: This shows how effective the model is at identifying bad language.
-
Percentage of correct bad language predictions: This gives a more accurate representation of the model's effectiveness, taking into account the total number of bad language predictions.
-
Number of feedbacks categorized as good/normal language: This helps to understand the overall capacity of the model to determine good language
-
Number of correct good/normal language predictions: This can show how effective the model is at identifying appropriate language.
-
Percentage of correct good/normal language predictions: This gives a more accurate representation of the model's effectiveness, taking into account the total number of good/normal language predictions. Image below shows the monitoring of the model in production:
The model so far is very good at predicting bad languages, with only one instance where a tweet contained good language was predicted to contain bad language. This is where we are most interested in, catching all bad languages so bad tweets arent sent out there for kids to consume, as can be seen with the 95.2% performance of the model.
Good language tweets scored a little worse, even though our focus isnt solely on that, an improvement in that aspect is required. Because recall is the most important here, that is what was monitored.
This is the way model monitoring is done so far to keep track of the metrics of the model as users are interacting with the app, to get a sense of how the model is doing, and if the performance starts declining, appropriate measures are taken.
One potential avenue for future development is to incorporate KidFriendlySocial as a browser extension. This would allow users to easily flag inappropriate language and receive recommendations for kid-friendly alternatives while browsing Twitter or other social media platforms. By integrating KidFriendlySocial directly into the user's browser, we can provide a more seamless and user-friendly experience.
Another possible direction for future work is to integrate KidFriendlySocial directly into Twitter. This could take the form of a built-in language filter, which would automatically flag tweets containing inappropriate language and offer users suggestions for rephrasing their tweets. By partnering with Twitter, we can potentially reach a much larger audience and make a greater impact in promoting online safety for kids.
As with any machine learning model, there is always room for improvement. In future iterations of KidFriendlySocial, we would gather a lot of more diverse data that captures a lot of bad language and even in other languages in order to detect bad language no matter how its put out. One way is to keep rolling out the app to differnet and more diverse users and capturing their feedback and using it to add to training data. We may also explore different models or fine-tune our existing models to improve their accuracy and reliability. This could involve experimenting with different transformer models. Also, reducing the inference and latency speed of the predictions by employing techniques such as quantization and weight pruning to deliver quick results to users
One potential avenue for future development is to incorporate multilingual support into KidFriendlySocial. This would allow the application to flag and recommend improvements for potentially harmful tweets posted in languages other than English. By expanding the language capabilities of the app, we can make it more accessible to a wider range of users and help protect children on social media platforms beyond English-speaking ones.
Finally, we may consider integrating KidFriendlySocial with other social media platforms beyond Twitter. This could include popular platforms such as Facebook, Instagram, or TikTok, allowing us to expand our reach and provide even more comprehensive language filtering and recommendation services.
These are just a few examples of the potential future directions for KidFriendlySocial. By continuing to develop and improve this tool, we can work towards a safer and more positive online environment for kids and young people.
KidFriendlySocial is a web application that aims to make social media safer and more child-friendly. By using machine learning models(NLP), web technologies, container technologies and cloud hosting, we were able to create a minimal viable product(MVP) application that can detect and flag bad languages and provide recommendations to improve the language used in social media posts.
In the future, we plan to expand KidFriendlySocial by incorporating it as a browser extension and integrating it into popular social media platforms. We also plan to enhance the application by reducing latency speed and inference speed, obtaining and improving the dataset and adding a few more very useful features.
We hope that KidFriendlySocial can help parents and guardians feel more at ease with their children's social media usage and contribute to a safer and more positive online community.