Assessing Impact Of Explicit And Implicit Relationships On User’s Interests

A user’s interests are influenced by his relationships. A social network can be imagined to be made up of explicitly declared relationships as well as implicit relationships. Direct or explicit networks consist of explicitly declared friendship or follower relation. Hidden or implicit networks consist of similar users based on their similarity as decided by their footprints in the social network. In this work, the impact of user’s implicit network as compared to explicit network is investigated with respect to user’s behavior. Understanding this impact is of paramount significance for product driven websites such as Amazon, Yelp etc. These websites offer social networking features such as option to follow another user to enable product recommendation. In spite of such features, explicit networks on such websites deliver limited benefit for reasons such as very small fraction of users forming such explicit relationships and rapid staling of such explicit social links. Hence, these websites have to depend mostly on classical collaborative filtering techniques. In such scenario, an implicit network based on user activity footprints can be a viable alternative. In this work, an appropriate data set is chosen for this investigation based on the richness of the data set. The user’s interest is interpreted as the business that he frequents and the task of finding user’s interest is modelled as a link prediction problem between user and business. After detection of the networks, the network information (users’ relations) is captured using node embedding and then link prediction is performed. The baseline is provided by the explicit network. This investigation has looked at various combination of implicit and explicit networks to retain maximum information about users and their implicit/explicit relationships. It is observed that a certain combination that includes implicit network outperforms the baseline.