border border


Team members

Chong Yin Yi Christy (ASD), Looi Siao Si (ASD), Pei Jinling (ESD), Ang Jian Yu Kenny (ESD), Yang Peng (ESD), Ng Ming Bing (ISTD), Xu Licheng (ISTD)


Bige Tunçer, Matthieu De Mari, Yue Mu

Writing Instructors:

Rashmi Kumar

Teaching Assistant:

Ataman Cem

Explore Architecture like never before.

CONNECTARCH is an AI-enabled website that empowers anyone, particularly students and professionals in the architecture industry, to discover architecture by connecting iconic works through a unique graph interface. CONNECTARCH features an industry-specific recommendation system to assist in efficient and relevant precedent recommendations. 



What is a precedent?

There is no list of facts that students can memorise to learn architecture. Instead, students learn by example, and this includes studying buildings and projects designed in the past. These example projects are called precedents.






What do architects and students need?

Architecture lacks structured documentation for precedents due to the subjectivity of the field. To address this, we created a platform for architecture students that recommends precedents within the architecture context.


Needs Finding

1. Solely Architectural

Existing platforms mix visual content of various kinds and are distracting for architecture students who wish to focus on architecture.

2. Curated content

Students benefit from being able to learn from precedents from the established canon of architecture or are otherwise considered excellent examples.

3. Easily accessible & Seamless UI

Students need to be able to reference precedents with ease to make the learning process smooth and enjoyable.



What is a Knowledge Graph?

A knowledge graph is a series of nodes linked with edges to represent a set of information or data.

Each precedent is connected to key information about that precedent, such as the architect(s), location, or materials involved.

In this graph, we have two precedents, Fallingwater House and Maison Des Fondateurs. They are connected to each other by the principle of integration with the site. From this, the platform can decipher that a student interested in Fallingwater House would also be interested in Maison Des Fondateurs.

asset 2




Recommendation System

Using a cutting-edge AI engine, our recommendation system combines a content-based approach and a collaborative filtering approach. These two approaches combine expertise-driven knowledge embedded in the graph with the wisdom of the crowd to make optimal recommendations.


Initial Weights

In a knowledge graph, each edge connecting two nodes is assigned a weight representing how closely those nodes are connected. To get more accurate recommendations, these weights are set based on user interests.

In this example, the user is looking at Biomuseo. If the user is most interested in context, a good recommendation might be the Bilbao Guggenheim, as both are designed by architect Frank Gehry. However, if the user is looking at Biomuseo because it is a museum, its programme, the best recommendation might be Castelvecchio Museum.

Similarly, the user might be interested in V&A Spiral or Denver Airport if s/he is interested in the concept behind Biomuseo or the formal attributes of Biomuseo.





Content-based Approach

This approach considers the nodes and edges of the knowledge graph. The platform calculates the distance between two precedent nodes based on the weights of all the edges between those nodes and recommends the nodes closest to the current node.


Collaborative Filtering Approach

This approach considers recommendations specific to each user. The system tries to predict your rating of a precedent by taking into account your rating history and the ratings of that precedent by other users.


Combined Recommendation System

The platform gives recommendations based on a combination of collaborative filtering and content-based recommendation.

This is achieved with an algorithm that is responsive to the amount of user data we have, relying on the content-based system when there is little data and more on collaborative filtering when there is more data.


Weights Updating

Updating weights improves the state of the graph, allowing recommendations to become better and better, as the weights more accurately reflect how closely connected precedents are.

When a user gives a low rating to a precedent, the weights of connecting edges is decreased, and when a user gives a high rating, the weights of connecting edges are increased.

asset 6

Reinforcement Learning

The knowledge graph is part of a larger system, known as a state-action-reward model which seamlessly connects the graph, the recommendation system, and learning processes.

1. State

The knowledge graph encodes architectural information, representing the architectural understanding that the platform has.

2. Action

The system makes a recommendation to the user based on this information.

3. Reward

Depending on how good the user feedback is, the system receives a reward. This allows it to update its understanding to learn to give better recommendations.

Web Application Features

Our web application has 4 key features which help to aid learning in the architectural community by recommending relevant nodes.

1. Interactive and Dynamic Knowledge Graph

Discover applicable precedents through a rich browsing experience. Our graph transforms to your every need.

2. Split Screen View

The ultimate split-screen experience. Our knowledge graph keeps you connected to the big picture while you discover more about your preferred precedents.

3. Bookmarks and Projects

Your preferred precedents are one click away thanks to our projects and bookmark features. Create projects and receive dedicated recommendations.

4. Device Agnostic

Our web application is available on desktop and mobile user interfaces, making it easy for architecture students to browse even on the go!

asset 15mdpi

Industry Partner



student Chong Yin Yi Christy Architecture and Sustainable Design
student Looi Siao Si Architecture and Sustainable Design
student Pei Jinling Engineering Systems and Design
student Ang Jian Yu Kenny Engineering Systems and Design
student Yang Peng Engineering Systems and Design
student Ng Ming Bing Information Systems Technology and Design
student Xu Licheng Information Systems Technology and Design
border border