Pebble ppt

Creating PebbleAppTailor

for

Maureen Teyssier, Insight Fellow

without apps

with apps

useraction

Increasing User Engagement

Pebble doesn’t have App rating data!

database of usage

Pebble doesn’t have App rating data!

database of usage

Metric for Engagement:

Pebble doesn’t have App rating data!

Launch count per App per User

Predict usage of all Apps for a User, based on how they use the Apps they already

haveRecommend the Apps that will be used the most.

Recommend the Apps that will be used the most.

Predict usage of all Apps for a User, based on how they use the Apps they already

have

• Launch Count as Proxy for User Ratings• Item-based Collaborative Filtering• Cosine-Similarity Matrix

• Sparsity of Dataset: 0.6%• Sparsity of Recommendation Matrix: 2.5%

Recommend the Apps that will be used the most.

Predict usage of all Apps for a User, based on how they use the Apps they already

have

Pebble Database

PebbleAppTailor

Leave-One-Out ValidationUser’s Apps = [PebbMine, Cards, YachtTimer, …….]

Leave-One-Out ValidationUser’s Apps = [PebbMine, Cards, YachtTimer, …….]

Can I recommend the App I removed?

Leave-One-Out ValidationUser’s Apps = [PebbMine, Cards, YachtTimer, …….]

Can I recommend the App I removed?

Next Steps

• Categorize and subcategorize Apps (NLP)• Select/Exclude App categories• Highlight App categories that emphasize

uniqueness of Pebble Hardware

Built a recommender:

• emphasizes Apps with high user engagement• works really well• will be a tool that Pebble uses going forward

Maureen Teyssier

Maureen Teyssier

Creating PebbleAppTailor

for

Maureen Teyssier, Insight Fellow

This user loves games.…

This user loves games, and is recommended games.

This user cares about weather, directions, GPS. Lives in Netherlands?

…and is recommended 2 Apps for the Netherlands, 3 different types of GPS/Map Apps, a functionality App, a weather App.