Vous êtes sur la page 1sur 25

Online Location Based Social Networks (LBSNs)

- Combine location information with social network features - For Example: BrightKite, Gowalla, and Foursquare

Foursquare is one of the largest and mostpopular LBSN

To analyze different behavioral profiles on Foursquare by doing an in-depth study of how users exploit various Foursquare features like Tips, dones and To-dos

Characterization of Users and Venues wrt number of tips, dones , to-dos and percentage of tips containing links

Clustering of users into profiles based on Total number of dones and to-dos received Number of venues tipped % of tips containing links

Venues

- Named locations like restaurants, schools, museums etc. - Points on every check-in - Mayor-ship

Tips

To-dos Dones

- Public user Feedback - Also useful for business owners- provide public opinion
- User may choose to follow another users tip
- A sign of agreement

FourSquare API was used to gather data about venues and about user who interacted with those venues. Each venue is assigned a unique and sequential numerical Identifier (ID). Largest ID is denoted by M. Crawling strategy randomly selects ID between 0 and M. This ID is used in HTTP GET request to FourSquare API.

Information obtained for each venue includes:


All tips posted for the venue. Identification of users who posted these tips. Number of Dones, ToDos received by each tip. Number of users who checked-in the venue. Venue Category. Geographic coordinates of venue.

Largest ID for which response was obtained was 20 million. 1.6 million venues gathered during the time period May 23rd to July 19th . 1 million tips were gathered. 3.8% venues were verified.

More than half millions unique users were identified. Out of these 1248 were Brand Users. 18.5% received atleast 1 tip.

Venues

characterized in terms of -Total number of tips - Total number of dones and to-dos associated with all the tips Tips represent amount of reviews Dones/ To-dos represent amount of user feedback per review

57% venues have only one tip Some venues (approx. 500) have more than 100 tips

Few venues (approx. 200) received more than 1000 dones/to-dos on tips

Used spearmans rank correlation coefficient


Number of done/to-dos is strongly correlated to Number of tips (Spearmans coeff. = 0.6) Number of dones/ to-dos and number of distinct users visiting the venue also show a strong correlation ( Spearmans coeff. = 0.56)

Considered Venues with at least 2 tips and one done/to-do 70% of venues have CVs larger than 1 9% of venues have CVs larger than 2

Users are characterized on the basis of:


- Number of tips posted by each user - Total number of dones and to-dos received by all tips - Percentage of tips containing links

- Number of venues tipped

It was observed most users posted few tips at few venues and few active users posted many tips at many different venues. 66% users posted only one tip, 70% users posted tips at only one venue. 67 users posted more than 100 tips and 39 users posted tips at more than 100 venues.

Figure c:
103,348 users received only one Done/ToDo in their post. 90 users received 1000 Dones and ToDos.

Figure d:
Almost 98.7% of user who posted at least one tip did not include any link in tips. Around 200 users were identified with more than 60% of their tip containing links.

Number of tips and number of venues tipped by a user are strongly correlated (0.93). Number of tips and number of to-dos/dones

are only moderately correlated (0.4).

This implies that users influence is only

moderately correlated with tipping activity, but


more related with tipped venue

Determined user influence by plotting number of tipped venues Vs. total number of dones and to-dos received per user Diameter of venues: Maximum distance between any pair of venues tipped Divided users in to two categories using diameter of 40 Km as threshold Only users with atleast 10 tips were considered.

There are highly active users who do not receive much feedback (User 1) Users at right top corner are highly influential brand users User 2& 3 are focused users with high influence Both graphs show a trend for users with larger number of tips tipped venues having bigger diameter

To detect spamming users with atleast 10 tips and atleast 60% of tips containing links were analyzed No clear correlation between % of tips containing links and number of tipped venues or number of dones/to-dos was observed No clear correlation between diameter of tipped venues and suspicious behavior

User5, 6 and 7 show behavior pattern similar to spammers User 8 is a brand user who got high feedback despite posting tips more than 90% of which contained links Some spammers succeed in triggering interests of users

Vous aimerez peut-être aussi