Study On Link Prediction Of The Weighted Networks With Time-aware

Link prediction is an important branch of social network sub-ject which mainly focuses on predicting connections which are most likely to appear in future considering current existing links in social net-works. In recent years, link prediction in social networks has attracted increasing attention from researchers because it is of both theoretical in-terest and practical significance in modern science.However, the existing link prediction algorithms mainly used in simple unweighted static networks without considering the impact of weights and time series, so it is not well adapted to the case of complex networks. To solve this problem, this thesis proposes several new me-thods which synthesize and improve existent link prediction algorithms with the concepts of vertex weights and link weights based on weighted networks. Besides, this thesis introduces a new index which is called time factor to help computing the similarity degree between nodes during a time series based on time-aware networks. Finally, this thesis proposes a hybrid method which combines weights information and time factor to get a better result for time-aware weighted networks.In this thesis, several real world datasets are chosen to analyze and compare proposed link prediction algorithms according to weighted net-works and time-aware networks separately. Experimental results show that the proposed weighted link prediction algorithms outperform un-weighted prediction methods, and time-aware based link prediction algo-rithm is better than traditional methods without time series; the hybrid time-aware weighted link prediction algorithm can achieve high accuracy as well.