[databeers] 06/05/2014 - dani villatoro: “cicerone: your venue recommendations through twitter”

22
@BCNPlaces4 @jordi_aranda @djkram @techisthenewpop @dani_agent

Upload: data-beers

Post on 01-Jul-2015

136 views

Category:

Data & Analytics


0 download

TRANSCRIPT

Page 1: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

@BCNPlaces4

@jordi_aranda @djkram

@techisthenewpop @dani_agent

Page 2: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

@dani_agent

Page 3: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

@dani_agent

Page 4: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

What are we going to tell you?

location based

social

in-situ

recommender system

(e.g. Cicerone! Tell me a place for coffee now! // Go to Café Las Maravillas)@dani_agent

Page 5: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Location based

@dani_agent

Page 6: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

If I click “Coffee”

1 Km.

Seriously

Foursquare?

@dani_agent

Page 7: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Social

@dani_agent

Page 8: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

In Situ

@dani_agent

Page 9: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

• My friends’ opinions matter.

• Experts’s opinions matter even more.

• Easy to query.– Not another app!

Requirements

@dani_agent

Page 10: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Twitter as a Communication

Channel and Foursquare as a

Location Provider

41.371141, 2.144209

41.378676, 2.153479

@dani_agent

Page 11: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

• Streaming API offers 1% of the total Twitter

Traffic.

– We never get the Over-Exceed alarm.

• Is the sample good enough?

– YES!

Is the Twitter Streaming Sample

Good Enough?

@dani_agent

Page 12: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Area Knowledge@dani_agent

Page 13: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Social Importance

ME

Stranger

@dani_agent

Page 14: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

How does it work?

@dani_agent

Page 15: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Scalability Issues

@dani_agent

Page 16: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

MongoDB

• Why?

– Raw tweets from Streaming API: JSON

– Geospatial Index

– Scalability

• When?

– Crawlers Writting

– Tweets and venues access by the algorithm.

@dani_agent

OVER

3.000.000

geo-tweets

Page 17: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Neo4J

• Why?

– Social Graph Representation.

– Java Native

• When?

– Insert relationships

– Querying on social graph

@dani_agent

OVER 200.000

nodes

and 1.5M

relationships

Page 18: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

So… how good is Cicerone?

@dani_agent

Page 19: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Pilot Experiment

One of the options is calculated by

Cicerone and the other one by

Foursquare. But only Cicerone knows

which is which. TELL US WHICH ONE

YOU LIKE BETTER.@dani_agent

Page 20: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Pilot Results

• 20 users – 150 requests

@dani_agent

Page 21: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

Conclusions

• Pre-production Geo-Social Recommender

System.

• Spatial Capabilities and JSON by MongoDB

• Social Relationships by Neo4j.

• Good experimental results.

@dani_agent

Page 22: [Databeers] 06/05/2014 - Dani Villatoro: “Cicerone: Your venue recommendations through Twitter”

@dani_agent

@dani_agent