Encouraging Inactive Users towards Effective Recommendation




Journal Title

Journal ISSN



Volume Title


De Montfort University


Thesis or dissertation

Peer reviewed


Disagreement amongst users in a social network might occur when some of them have different opinion or preferences towards certain items (e.g. topics). Some of the users in the social network might have dynamic preferences due to certain situations. With these differences in opinion amongst the users, some of the users might decide to become either less-active or inactive in providing their opinions on items for recommendation processes to be possible or effective. The current state of the users will lead to a cold-start problem where the recommender system will be unable to find accurate preference information of the users for a recommendation of new items to be provided to them. It will also be difficult to identify these inactive or less-active users within a group for the recommendation of items to be done effectively.

Attempts have been made by several researchers to reduce the cold-start problem using singular value decomposition (SVD) algorithm, but the disagreement problem amongst users will still occur due to the dynamic preferences of the users towards items. It was hypothesized in this thesis that an influence based preference modelling could resolve the disagreement problem. It is possible to encourage less-active or inactive users to become active only if they have been identified with a group of their trustworthy neighbours. A suitable clustering technique that does not require pre-specified parameters (e.g. the number of clusters or the number of cluster members) was needed to accurately identify trustworthy users with groups (i.e. clusters) and also identify exemplars (i.e. Cluster representatives) from each group. Several existing clustering techniques such as Highly connected subgraphs (HCS), Markov clustering and Affinity Propagation (AP) clustering were explored in this thesis to check if they have the capabilities to achieve these required outputs. The suitable clustering technique amongst these techniques that is able to identify exemplars in each cluster could be validated using pattern information of past social activities, estimated trust values or familiarity values. The proposed method for estimating these values was based on psychological theories such as the theory of interpersonal behaviour (TIB) and rational choice theory as it was necessary to predict the trustworthiness behaviour of social users. It will also be revealed that users with high trust values (i.e. Trustworthy users) are not necessarily exemplars of various clusters, but they are more likely to encourage less active users in accepting recommended items preferred by the exemplar of their respective cluster.





Research Institute