Inferring User Interest Familiarity and Topic Similarity with Social Neighbors in Facebook INSTRUCTOR: DONGCHUL KIM ANUSHA BOOTHPUR

INTRODUCTION  Active users converse with their social neighbors via social activities such as posting comments one after another.  Social correlation, researchers have proposed solutions to inferring not only user attributes like geographic location and schools attended but also a user’s interests in social networks.  we explore how we can formulate a method of inferring user interests by combining both familiarity and topic similarity with social neighbors.

System Workflow

Inferring User Interest Using Topic Structure  We formally define Interest-Score for interest i k of user u i as:  Correlation i,j,k ， the strength of correlation between u i and u j for i k, is defined as:

Correlation-Weight  We compute Correlation-Weight w i,j,k by estimating similarity between the two topic distribution vectors and averaging them  h = 1 to H(the number of total social activities between u i and u j )  = a topic distribution vector of each social content  = a topic distribution vector of each interest content

Latent Dirichlet allocation  Latent Dirichlet allocation ( LDA ) is a generative model that allows sets of observations to be explained by unobserved groups that explain why some parts of the data are similar.  The basic idea is that the documents are represented as random mixtures over latent topics, where a topic is characterized by a distribution over words.  we want to put a distribution on multinomials. That is, k-tuples of non-negative numbers that sum to one.  The space is of all of these multinomials has a nice geometric interpretation as a (k-1)-simplex, which is just a generalization of a triangle to (k-1) dimensions

Latent Dirichlet Allocation(LDA) The parameters and are corpus-level parameters. The variables are document-level variables The variables z dn and w dn are word-level variables and are sampled once for each word in each document

Dirichlet Distributions Useful Facts: This distribution is defined over a (k-1)-simplex. That is, it takes k non-negative arguments which sum to one. Consequently it is a natural distribution to use over multinomial distributions. In fact, the Dirichlet distribution is the conjugate prior to the multinomial distribution. (This means that if our likelihood is multinomial with a Dirichlet prior, then the posterior is also Dirichlet!) The Dirichlet parameter  i can be thought of as a prior count of the i th class.

Online Familiarity  We formulate Online Familiarity f i,j as:  Freq(i,j) is defined as:  P i,j = u i writes a posting into u j ’s wall  C i,j = u i writes comment(s) in u j ’s posting  L i,j = u i likes u j ’s posting.

Dataset  A user writes a posting on his social neighbor’s wall or a posting is written in his wall by the social neighbor.  A user writes comment(s) in his social neighbor’s posting or comment(s) is written in his posting by the social neighbor.  A user likes his social neighbor’s posting or his posting is liked by the social neighbor.(In Facebook, a user expresses his/her preference about a post by pressing the “Like” button.)

Based on Questionnaire

Online Familiarity

Evaluation  Based on User Explicit Interest  EXP is set of the user’s explicit interests  INF N is a set of top-N inferred interests ordered by Interest-Score  Based on Questionnaire

Result

Conclusion  We consider topic similarity between communication contents and interest descriptions as well as the degree of familiarity  We plan to extend the proposed scheme by using not only spatial aspects such as a user’s location, trace history or characteristics of a place, but also temporal context like time slots

THANK YOU