Goals

• Determine interface design for game recommendation and expert training.
• Determine how to convert game attributes into a vector for machine learning. Decide what features do we select.
• Create consensus on terminology used for the machine learning component.

Discussion

At this point, there are two interfaces that need to be designed.

Recommendation Interface

The first interface is the users normal route to interact with our system. It is the recommendation interface. It takes in some user specified attributes and returns a list of recommendations.

Since many attributes that a user may specify will not be used in every query we do not want to clutter up the page with input fields. This sends the wrong message about fields being optional (or at least makes it harder to indicate), it is ugly and it is distracting resulting in an interface the user will not enjoy using. This we want a way to allow the user to opt into filling out parts of the form (or select sub-forms). Chris came up with two ideas for this. Both ideas simply provide a graphical way for the user to indicate which attributes they want to specify.

1. Accordion – Display each attribute as a section in the accordion. Allow the user to open multiple sections. An option section indicates that the user will be filling in its fields.
2. Sidebar Buttons – Instead of unrolling sections, the user would hit buttons in the sidebar describe the attributes he or she wants to filter on. Each button would be stateful (i.e. buttons toggle between a pressed and unpressed state). When pressed, a form appears in the main column of the page for that attribute. Multilple pressed buttons means multiple forms.

In general, the team gravitated toward the sidebar idea. It makes the distinction between used and unused forms clearer.

Expert Interface

The second interface will be used to train SiGMa (Similar Game Machine) which is this projects machine learning component. We spent a while discussing the proper terminology to describe the data we are looking for. We decided that the term similar is not exactly what we want. Instead we are trying to find a relationship like “given game A how good of a recommendation is game B”. Our interface will reflect this, but there are no plans to change the name of the library.

The original interface proposed putting two games in front of the expert and a slider. The expert would then manipulate the slider to describe how similar the two games are. Besides changing the terminology used here from “how similar are these two games” to “how good of a recommendation is game A if someone likes game B” we also want to change the layout. Alex made the point that an expert might not know one of the two games we are asking them to prepare. Providing a skip feature would work, but it might mean the expert does a lot of skipping if they know 100 games out of the 1000 we are using in the system it could take awhile to find a pair of games that they know both of. Thus the excepted interface involves a two step process. The first step is a page that shows a group of around 10 games. The expert selects which of these games they know. The expert then continues to the next page which lists all two game combinations from their selections on the first page.

Results

• Decided on two discussed interface designs.

Outgoing Responsibilities

• David is to work on creating a tracer-bullet implementation of the first interface.
• Chris is to work on creating a tracer-bullet implementation of the second interface.

Attendance

• Ryan McGowan (scribe)
• Chris Powers
• David Albert
• Alex Burkhart

Location

Thompson Café.