1. General
Category
SDG 3: Good Health and Well-being
SDG 4: Quality Education
Category
Schools/Education
2. Project Details
Company or Institution
Omdena
Project
Save the Children
General description of the AI solution
OVERVIEW
Children across the world encounter many issues while surfing online and get exposed to cyber crimes. Sometimes, they meet and interact with adults and face child grooming and experience other unwanted behaviors from the interacting adult. From a previous survey, it is evident that children are shy to discuss their online experience with parents and other relatives. This way they lack proper guidance regarding the online interactions and fall prey to the predators. To mitigate this problem and to foster safe online communities, an Artificial Intelligence (AI) powered educational bot was developed. Since the bot is not a human, it is expected that children may feel more safe and comfortable in sharing their experience with the bot and get proper guidance. Our solution has direct positive impact on SDG 3 and SDG 4. Our partner Save the Children is thrilled with the final solution.
METHODOLOGY
The educational bot comprises two main modules- the training module and the graphical user interface (GUI). It also has a corpus of intents written in the JavaScript Object Notification (JSON) format. The bot is first trained on the corpus of intents and the GUI is used to get user input. The user input is then matched with proper context and a previously defined list of intents to fetch the appropriate answer for the question in hand.
DATA SOURCES
For developing the corpus of intents the website eSafety Commissioner (https://www.esafety.gov.au/young-people) has been used extensively. The website has a guideline for different topics including cyberbullying, gaming, online abuse, unwanted contact, sending nudes, etc. For developing the corpus of intents the information provided by the website has been arranged in a question-answer format with an appropriate context. In addition, the educational bot has some information integrated from the game reviews dataset and can be extended to incorporate the social media and news articles dataset.
METHODS & TOOLS
To train the educational bot a deep neural network model has been used as described below:
Model: Artificial Neural Network (ANN)
Hidden layers: 2
No. of units in the hidden layer: 128, 64
activation function: ReLu
Optimizer: SGD
Loss: Categorical_cross_entropy
Output activation: Softmax
NLTK package has been used for pre-processing (e.g. lemmatization) and tokenization. Artificial Neural Network (ANN) and Pattern matching have been used to match the extracted intent with the saved intent in the corpus. The model has been trained for 200 epochs with a batch size of 5 and a dropout rate of 50%. The vocab size is 102. The graphical user interface (GUI) has been developed using the Tkinter package from python.
RESULTS & INSIGHTS
The chatbot was designed, developed and deployed in a 8-week period by 54 collaborators from across the world with a single minded purpose of curbing Online violence against children. The model achieved a training accuracy of 96.83% for intent matching. The fetched response corresponding to a question makes sense in most cases.
Website
Organisation
Omdena
3. Aspects
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
Online Violence Against Children (OVAC) is a problem that is becoming more and more common in modern society. In a period of 8-weeks, a team of 43 collaborators from 18 countries partnered with Save the Children to use data to investigate violence against children in digital spaces. OVAC is a broad term, defined differently by various authors. Several attempts exist to categorize OVAC into different classes such as cyberbullying, cybergrooming, sexting, child pornography, and more.
We investigated multiple data sources viz. online forums and research articles, newspaper articles and social media. Using a combination of Natural Language Processing (NLP) and other Machine Learning (ML) and Deep Learning (DL) techniques together with text visualization, analysis of different data sources (social media, newspaper articles and research papers, among others) we generated insights on the causes and actors of online violence against children, as well as identify behaviors of predators online.
Our solution was very innovative and one of its kind in the world. After 2 months of intense work, we took all the analyses and modeling results and created:
a) a tool to predict predatory behavior during online chats and
b) a chatbot to educate minors about online risks as they occur.
Select tasks performed included:
– Scraping real-life narratives of victims from various platforms including Reddit, Newspapers, Survivor Forums and Google Searches.
– Analyzing Social media and game forums where predators commonly approach children
– Manually labeling narratives to provide insights into complex themes, such as the behaviour of predators and victims prior to and throughout the online interactions, with the goal of answering a series of questions.
– Identifying popular platforms for predators and the most vulnerable age groups of victims.
– Applying NLP techniques such as ngrams, name entity recognition and keyword extraction to analyze these stories and features.
– Graphing findings based on label frequency.
We looked at multiple AI driven approaches for building the chatbot. The chosen approach became a ground-up build of a sequence-to-sequence model to handle the word association.
In summary, the team collaborated extensively, did a lot of research, analyzed tens of thousands of data points from numerous media sources, and used various classification models to come up with the proposed solution that helps vulnerable children stay safe against online violence.
The partner Save the Children was extremely pleased with the solution we offered and is more than happy to give a testimonial on the efficacy of our solution.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
We strongly believe the work we did positively impacted 2 SDGs, viz. Improved mental and physical health and well being of children by staying safe from online violence [SDG 3] that allows them to focus on their studies and education (SDG 4].
Our proposal consists of objectively verifiable indicators of the project outcome. The feedback and advice for appropriate action from the chatbots is real-time and the outcomes for children are measurable. It has a direct impact on the mental well being of children.
The proposal provides a easy to use and a clear technical solution by way of utilizing technology that children in middle and high school use on a daily basis (eg. a phone or computer). It can be easily integrated with the online platform where the children participate and keep them safe from any type of perceived threats or violence against them.
The global impact will be high if a) educational institutions, b) parents and c) social platforms adopt the AI technology that was developed through this solution. Even with a 1% adoption rate, we can touch 5M+ school going children across the world.
Impacting children through our AI driven solution will drive a safe and healthy community for children with a positive outlook and bright future prospects (SDG 3, 4].
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
The solution was delivered for our partner Save the Children for usage by their stakeholders and communities. Our solution is easily scalable across regions as it can be easily layered on a desktop or mobile application for real time advice. Below is a quick summary of the chatbot with demo videos.
EDUCATIONAL CHATBOT AND PREDATOR BEHAVIOR ANALYSIS
Children encounter many issues while surfing online and get exposed to cyber crimes. Sometimes, they meet and interact with adults and face child grooming and experience other unwanted behaviors from the interacting adult. From a previous survey, it is evident that children are shy to discuss their online experience with parents and other relatives. This way they lack proper guidance regarding the online interactions and fall prey to the predators. To mitigate this problem in this task an Artificial Intelligence (AI) enabled educational bot has been developed. Since the bot is not a human, it is expected that children may feel more comfortable in sharing their experience with the bot and get proper guidance.
Demo Video of Educational Chatbot: https://drive.google.com/file/d/1LXcmaSIPL8mSfpaoRB_wr8513REeWdPW/view?usp=sharing
Demo Video of Predator Chatbot: https://drive.google.com/file/d/1oLxyhR-5x8C35qSseSf_06QwVr4JhyQX/view?usp=sharing
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
The project analysis followed the Ethical Guidelines for Statistical Practice.
All data used for this project was available publicly in different formats, which was consolidated and preprocessed by the team.