Method and System for Learning Topic based Social Influence in Social Network
Publication Date: 2013-Jul-23
The IP.com Prior Art Database
Francesco Bonchi: INVENTOR [+3]
AbstractA method and system for learning topic based social influence in social network is disclosed.
A method and system for learning topic based social influence in social network is disclosed.
In general, the standard social influence based propagation models assume that influence of a user does not depend on a considered topic. The assumption fails in recognizing real world cases, in which a user can exert high influence on a topic and a lower influence on other topics.
is a method and system for learning topic based social influence in the social
network. The method models and estimates
trust and influence of a node on different topics and models influence-driven social propagations that are used in viral marketing campaigns. The method selects the set of nodes to target in order to obtain the largest diffusion for a given topic.
In an exemplary embodiment, the method monitors interactions of the user in the social network exhibiting different degrees of authoritativeness and interest on various topics.
In an exemplary embodiment, topics can be considered as categories of one or more of products and information, where each category gathers together items that tend to exhibit a similar diffusion pattern. The topics include, but are not limited to, consumer goods, gadgets, books and movies.
Thereafter, the method generates a propagation model that relies on the authoritativeness and influence of the user regarding a topic of relevance. For each user u in the social network authoritativeness for each topic z is defined as a weight wuz which measures the strength of the influence of the user u on the topic z. For each user u of the social network a distribution of probability (P) over the topic z is provided, which denote the degree of the interest of the node (u) associated with the topic z. Relevance of an item i under the topic z is defined by a multinomial distribution of the topic z over the item i.
Initially, user u chooses a threshold θu at random from [0; 1]. At time t, the decision of the user u to activate the item i depends on the influence exerted by neighbors of the user u who have already activated item i (their authoritativeness) on topic-wise interests of the user u and on the relevance of item i. Therefore, at time t user u actives item i, if and only if, the condition in Eq. 1 is satisfied.
where P(z|u) = νzu, while P(i|u; z; t) is the following logistic selection function:
selection scaling factors fv(i; u; t) and f(i; u; t) are used to
distinguish potential influencers from non influencers (fv(i; u; t)
= 0 if v
Є Fi(u; t)) and to potentially relate influence to
time, where Fi;u of u’s neighbors who failed in influencing user u
over item i is defined as shown below in Eq. 3:
The method includes a machine learning module which learns one or more parameters of the social network by mining a log of past interactions of the us...