“Empowering Knowledge” — Tech for Social Good
OpenEdu C Team awarded the most completed Data Science Solution at deploy(impact) 2022: An Annual Software Development Program for Social Good supporting women transitioning into Technology.
“Empowering Knowledge” was the motto of the data science solution created by the deploy(impact) team, OpenEdu C, awarded, in a very head-to-head competition, the most complete solution for the OpenEdu project in partnership with Wikimedia CH.
OpenEdu C team, composed of
- Carol Hsu, Claudia Porcellato, Daniela Pavia Santolamazza, Iwona Braun-Nowak, Kevin Zihlmann, and Marcela Rozo Velandia as the participants with a specific role to develop the solution in 6 weeks, and supported by Heba Hussein as the team satellite,
impressed jury members at deploy(impact)’s Closing Ceremony on Saturday, November 19, with a well-structured presentation, a clear message, and a technically strong and visual prototype.
So how did they pull it off? Which Project Management techniques did they use to deliver a solution that met the requirements of the Project Owner, Wikimedia CH?
During the first two sprints, many teams reported that the scope of the project was unclear, due to the many different possibilities teams could choose to solve this challenge. But what stood out in OpenEdu C’s final project presentation was how they approached the project by asking questions, a technique in project management that is correlated with successful deliveries. Asking questions right from the start allowed the team to have a better understanding of the problem and to create a solution that received the feedback of “impressive” from Ilario Valdelli, the project owner and also a jury member.
Before we dive into the details of how the winning data science team approached this project, let’s give you a bit of context by explaining what OpenEdu is and what were the expectations of the Project Owner, Wikimedia CH.
The Project, the PO and the Project’s Scope
OpenEdu is a platform launched in 2020 by Wikimedia CH to support, disseminate and promote projects, training tools, and news from the world of open education. It aims at giving educators a chance to improve their teaching techniques and knowledge.
deploy(impact) 2022’s project owner, Wikimedia CH, together with our women++ team, defined the scope of this project for 5 data science, distributed agile teams to work on for 6 weeks or sprints. The project scope: creating an ontology to organize and structure the large amount of data they have. Improving user usability and implementing crawlers as well as a moderation tool would be a plus for this project.
How did the OpenEdu Team C approach this Project?
At the beginning of the project presentation, Marcela Rozo, deputy project manager, explained that in the context of giving educators a chance to improve their teaching techniques and knowledge, the team assessed which problems OpenEdu users face when researching content on the platform. Three areas stood out and required improvement:
- Usability: users cannot find information quickly.
- User Experience: the website is overloaded with information creating a cognitive loading for the user.
- Value proposition: unclear value proposition message.
The next question that needed to be answered was: who is using the platform? After extensive research, using the current data as well as researching other Open education platforms, Iwona Braun-Nowak, project manager, presented the two personas they found and to whom they designed their solution:
After these initial assessments, the team then defined the project goal:
“Reduce the cognitive loading of the end user and improve the usability of existing resources, the website in itself, and the structure of the database”.
Structuring The Ontology: Where to Start?
With a clear goal it is time to get to work, right? Yes, but what is the starting point of a data science project like this?
Daniela Pavia Santolamazza, one of the team’s data scientists, explained that their first step was to look at the data pipeline and once again ask more questions:
- What type of users do we have?
- What type of data we are collecting and how?
- How are the users connected with the data?
- How is data stored?
Answering these questions allowed them to come up with an ontology that originated using the current OpenEdu structure, but adds improvements to account for the proposed changes, for example:
- the search feature,
- the related content feature,
- the overall user experience,
- as well as a simplification of the content uploading process.
The proposed ontology gives more structure to the OpenEdu database and makes it easier to manage data collection and future processes. To test the ontology and some of the features proposed, a dummy UI was programmed using Django python and connecting-in the backend to a PostgreSQL Database.
Searching and Finding Related Content with NLP
With an ontology structure defined, Claudia Porcellato and Kevin Zihlmann, the team’s other two data scientists, got to work.
At first, the team looked at OpenEdu’s website and asked themselves how they could use NLP so the user can find all the relevant materials they are searching for.
They realised that when a user entered a keyword on the search box, the current solution only looked at the title and subtitle of the content to see if the word matched the search. But OpenEdu C team went further and developed a solution that analyses the whole project description and shows the user all the projects that are semantically related to the researched keyword.
For the related content feature, Kevin Zihlmann explained that in the current solution, the content creator had to manually assign which topic the work was related to, but the team automated this process by comparing the descriptions of the uploaded projects on the website.
During the presentation, Claudia Porcellato was the one responsible for explaining the technical details of how the solution works. She started by explaining that the first thing they did was the pre-processing of the text which included lowercasing, removing stop words, lemmatization, and so on. After that, they loaded the pre-trained model, the SBERT model, which was a big advantage as they would not otherwise be able to train a model on such a large amount of data.
By using this model, the following step was then the computation of embeddings of the texts and the search key and at last the calculation of the cosine similarity scores, from which they were able to easily get all the projects related to a search.
Implementing The UX Mental Model
Carol Hsu was the backend and UI Design expert on the project and explained how the team used UX Mental Model and integrated it with the technical parts of their solution to solve the usability, user experience, and value proposition challenges that OpenEdu has.
The UX Mental Model is a perception of the user on how the User Experience (UX) should be, it’s formed from past interactions with other sites and applications. The model is helpful for designers to build experiences and interfaces that are intuitive, predictable, and therefore highly usable. With all the information from the team’s initial research, they took into account three aspects to create an User Interface (UI) that meets the user’s expectations and supports the effective functioning of the platform:
- First impressions matter: 94% of the users decide on the visual design whether they want to stay on the platform or not. Therefore the team, OpenEdu C, proposed a UI that reduces the cognitive bias of the user and minimises the information overload on the website.
- The majority of users use the search box: in websites where there is a lot of text, 90% of users use the search function. Being a source of shared Open Education resources means most OpenEdu users would appreciate having an enlarged search function that is easy to locate.
- Long processes are demotivating: the current content uploading process on OpenEdu is very long and hard for the user to follow up. The team’s goal was to simplify it to encourage users to complete all the steps.
All three aspects above were seen in practice as Carol Hsu presented OpenEdu C’s product demo and visual prototype. In the video, everyone present at the deploy(impact) closing ceremony, could see how their solution returned results that are semantically relevant to the keyword search and also a simplified, visually attractive, and intuitive content-uploading process that can be concluded in only four steps.
What is Next? The Solution’s Future Roadmap
The team did not only deliver the most complete solution, but they also created a roadmap with the next steps to help Wikimedia CH successfully implement the solution they created.
After the product demo, Carol Hsu explained that at that moment, in November 2022, the team already developed the ontology, created the data structure, loaded the NLP model, and created the visual prototype. In the next stage, planned for December 2022, comes the data collection stage, for which they already have the data structure, but will work on defining the data strategy and decide which type of data is meaningful and relevant for the teachers and educators.
After the data collection comes the automation step, which is planned for Feb 2023, when the team will implement more machine learning and NLP algorithms to optimise the whole process and be able to provide personalization and recommendations to the user.
In April 2023, once all the previous steps are completed the team, OpenEdu C, will have a ready-to-launch-to-market solution that achieves its goal of reducing the cognitive loading of the end user and improving the usability of existing resources, the website in itself, and the structure of the database.
The Team’s Individual and Collective Learning Experience
To close the OpenEdu C project presentation, Iwona Braun-Nowak, the team’s project manager, shared three of their combined learnings:
- Collaboration, open communication, and building trust are key to effectively working as a team in a project like this.
- Keeping the end-user’s needs, perspectives, and points of view in mind during the whole project is essential to deliver a solution that brings value and solves problems.
- Managing without authority, following the Agile methodology, was a new experience for Iwona and one that she described as
“a very good learning opportunity”.
Team OpenEdu C’s contribution, as well as the other 8 teams, made deploy(impact) 2022 a memorable learning experience for all the parties involved. As Iwona mentioned in her closing statement:
“thank you to all the support during the program from sponsors to the women++ team”,
we would also like to thank everyone involved for the great work towards making this initiative possible:
- mentors and sponsors: Accenture, Aiven, Microsoft, Migros, Constructor Learning, Thomson Reuters and Wikimedia CH,
- deploy(impact) tech leads, program support and core team (women++): Christopher Gwilliams, Giada Fallo, Luisa Contreras, Namrata Gurung, PhD., Maria Luisa Diaz Bedregal (Malu), Suntka Rinke.
- team satellites: Alexandra Bødtker, Beatrice Neacsu, Renu Y. Simek, Simo Paraschi, Amy Bellis, Gilda Fernandez Concha, Heba Hussein, Maria Giakoumelou and Siyana Gurova,
- jury members: Ilario Valdelli, Jennifer Duchetta, Jyotsna Budideti, Mathias Keller and Franziska Schait,
- and last but not least, the full women++ team of volunteers.
Stay tunned for the implementation of this project by Wikimedia CH, once decided which features from each solution created by 5 teams during deploy(impact) 2022 will be implemented.
About the author:
Elisangela Merlin is the Social Media Marketing Manager Volunteer, specialised in Digital Marketing with a background in E-commerce and the startup ecosystems. She believes that education is the most empowering force in the world, as it builds up confidence, brings down barriers, and opens opportunities to forge a better future for society. When not immersed in the social media world, Eli loves to spend time with her kids, cooking (mainly Brazilian food), going for long runs, and getting it all out in a CrossFit WOD.
Visit our website: https://www.womenplusplus.ch/
