Graph Interaction

Randomizer

Graph Randomizer Demo

This demo shows the graph updating randomly when the randomizer button is clicked.



Search Interaction

Graph Search Demo

This demo shows the graph updating based on clicking on the search bar queries.



Line/Word Interaction

Graph Line Interaction Demo

This demo shows the graph updating when words on a connecting line are clicked.

Calendar

Calendar Demo

There are multiple ways to save courses to calendar:

1. Saving via the search bar

Saving course via search bar demo

2. Saving via the information bar

Saving course via information bar demo

3. Saving via the quiz result's page

Saving course via the quiz result's page demo

Exporting Calendar

Exporting saved courses Demo

After filling out the 'Export saved courses' section, this information will be emailed to the user:

Emailed saved courses Demo

Now, assuming the user's calendar is cleared and after filling in the emailed input phrase,the user will get their selected courses back:

Importing saved courses Demo
Importing saved courses results Demo

Quiz

Quiz Demo

This demo recommends a course based on the user's background experience, preferred difficulty, and time of day. The background experience filter leverages department recommendations from our graph.

FAQ

How do we decide if a course is similar?

Once our course descriptions are vectorized as is described in "The Model" section, we are able to use cosine similarity to calculate the distance between each course. In other words, we can calculate relative distances, which hopefully indicate semantic/structural similarity in descriptions. If a pair of courses are similar enough to pass a threshold that we set, we will draw a line between each of their nodes on our graph.

What do the words on the lines mean?

Each course connection includes two words, each representing the most similar term from the other course’s description. These words are identified use a Word2Vec model, which, like the Doc2Vec model underlying our graph, captures semantic relationships. However, Word2Vec is specifically designed for individual words rather than entire documents. We also employ a LDA Topic Model to extract keywords from each description, enhancing the scores of specific words in each similarity comparison. This feature provides a simplified representation of how our Doc2Vec model functions, making its behavior easier to interpret. Additionally, upon hovering over the words on a line, if one of the words is in the description currently opened, that word will be highlighted.