Friends and Locations Recommendation with the use of LBSN By EKUNDAYO OLUFEMI ADEOLA 16.02.2014.

Friends and Locations Recommendation with the use of LBSN

By

EKUNDAYO OLUFEMI ADEOLA

16.02.2014

TERMINOLOGIES

• LBSN (Location Base Social Networks)• POI (Points of Interest)Specific point on earth • User Trajectory P1->P2->P3->…Pn

• Stay Point Time threshold (tout – tin >∂time)

Are we really just scattered around the world with random behaviors?

Law of Geography

Everything is related to everything else, but near things are more related than distant things.

Users and Locations are the two things to be considered here.

Here we have different kinds of graphs;

User - Location graph

User - User graph

Location - Location graph

Understanding Users

• How best can we understand users behavior, so that we can recommend for them some friend or locations. Here are some tips;

• One day Activity; Location history in one day

• Routine Activity; People with rich knowledge about a region

• Visit Point/Reference Point; Visit point(P) (tout – tin > time threshold(∂time)). Reference point eg House, School

• One day Activities;

Routine activities mining

• Discovering activities patterns from multiple 1-day activities.– Find groups of similar 1-day activities – Represent 1-day activities within the same group

with a routine activity model.

Time based clustering algorithms for visit point extractions

One-day activity clustering algorithm for routine activity mining.

Modeling human location history in semantic spaces

• Stay Point Representation

Stay region

Using IT-IDF (Term Frequency Inverse Document Frequency) the semantic meaning of a region could be gotten. Regards POIs as word and treat stay region as documents.

• Feature Vector; The feature of a stay region r in a collection of regions R is fr = <w1,w2 ...,wK >, where K is the number of unique POI categories in a POI database and wi is the weight of POI category i in the region r. The value of wi is calculated as ;

Example; Suppose thats1 contains two restaurants and one museum, and s2 only has four restaurants. The total number of stay regions created by all the users is 100, in which 50 have restaurants and two contain museums. So, the feature vectors of s1 and s2 are f1 and f2 respectively:

PS: Although we still cannot identify the exact POI category visited by an individual, this feature vector determines the interests of a user to some extent by extracting the semantic meaning of a region accessed by the individual

Using IT-IDF the semantic meaning of a region could be gotten. Regards POIs as word and treat stay region as documents.

User Similarities Calculations

• Popularity of different locations

• Hierarchical property of geographic spaces

• Sequential property of users’ movements; Loc A -> LocB -> LocC ->…-> LocN. – Here we consider Travel Match

• Travel Match

• The locations in a travel match do not have to be consecutive in the user’s original location history• What we need to detect for the calculating of user similarity are the maximum travelmatches

• Similarity Score

N = Total number of users in the database, n = number of users visiting location c.

fw(l) is employed to assign a bigger weight to the similarity of sequences occurring at a lower layer, where l = depth of a layer in the hierarchy.

Hierarachical User clustering

Generic Travel Recommendations

• Interesting Places Popular Travel Sequence Itinerary Planning Activities Recommendation.

Mining Interesting Locations

1) Formulate a shared hierarchical framework F, using the stay points detected from user's GPS logs, which are then clustered into hierarchy geospatial region. 2) Build location graphs on each layer based on shared framework F and user’s location histories

Factor considered

• Location Interest Uses HIT (Hypertext Induced Topic Search) to give Authority score and Hub score to places.

Location interest (cij) is represented by; where Iij = authority score of cluster cij conditioned by its ascendant nodes on level l, where 1 ≤ l < i.

• Travel Experience; In our system, a user’s (e.g., uk) travel experience is represented by a set of hub scores:

Where denoted uk’s hub score conditioned by region cij.

Detecting travel sequences

Example: Demonstrates the calculation of the popularity score for a 2-length sequence (i.e., a sequence containing two locations),A→C.

First we consider the authority score of location A(IA), weighted by the Probability of people leaving by the sequence (Out AC) = 5/7.Secondly authority score of location C(Ic) Then the hub score of users (UAC) who have taken this sequence.

Following this method, the popularity score of sequence C → D is calculated as follows:

Thus, the popularity score of sequence A→C→D equals:

PS: This calculation will continue until we get the K-length most travel sequence.

Itinerary Recommendation

Factors considerd

• Query Verification; Check the feasibility of query according to spatial and temporal constraints.

• Trip Candidate Selection; Searches a location-interest graph for candidate itineraries satisfying a user’s query

• Trip Candidate Ranking; It ranks candidate itineraries according to three factor namely;

• Elapsed Time ratio (ETR): Ratio between the time length of a recommended itinerary and that given by a user.

• Stay Time Ratio (STR): Ratio between time a user could spend in a location and that for traveling between locations.

• Interest Density Ratio (IDR): Sum of the interest values of the locations contained in an itinerary.

Itineraries is ranked by the euclidean distance in these three dimensions:

NB: Popular travel sequence could also be considered here.

Location-Activity Recommendation

Location – feature Matrix; It uses IT-IDF to assign a feature to a location (e.g. Restaurants, movies theatre).

Activities – Activities: It models the correlation between two diifferents activities that can be performed at a location.

• This is my first draft Sir, your comment will be well appreciated. Thanks you.