A Three-Level Hidden Bayesian Link Prediction Model in Social Networks

INTRODUCTION OF THE PROJECT

In social networks, link establishment among the users is suffering from complex factors. during this paper, we attempt to investigate the interior and external factors that affect the formation of links and propose a three-level hidden Bayesian link prediction model by integrating the user behavior also as user relationships to link prediction. First, supported the user multiple interest characteristics, a latent Dirichlet allocation (LDA) traditional text modeling method is applied into user behavior modeling. Taking the advantage of LDA topic model in handling the matter of polysemy and synonym, we will mine user latent interest distribution and analyze the consequences of internal driving factors. Second, due to the power-law characteristics of user behavior, LDA is improved by Gaussian weighting. during this way, the negative impact of the interest distribution to the high-frequency users are often reduced and therefore the expression ability of interests are often enhanced. Furthermore, taking the impact of common neighbor dependencies in link establishment, the model are often extended with hidden naive Bayesian algorithm. By quantifying the dependencies between common neighbors, we will analyze the consequences of external driving factors and mix internal driving factors to link prediction. Experimental results indicate that the model can't only mine user latent interest distribution but can also improve the performance of link prediction effectively.

EXISTING SYSTEM

A diversity range of link prediction methods are developed. for instance , a user relationship recommendation system, a hierarchical model, and an SAN model are often viewed as effective link prediction methods. this type of methods held the view that folks were more likely to become friends who have similar hobbies, language, culture, geographical information, or frequent interaction. Chang and Blei presented a relational topic models with the text data and analyzed the subject distribution of texts to predict links among the texts.

Presented an easy but effective similarity-based prediction strategy supported the label propagation, which imitated the communication between people naturally. However, the abovementioned methods focused on analyzing the user direct interests generated by key words and labels, none of those approaches leverage user latent interests generated by user behavior.

Link prediction methods leverage the behavior analysis that gently becomes a replacement research perspective. Relational learning focused on web link prediction, by clustering user behavior and representing the resulting clusters as a click-stream tree, using the click-stream tree to get the advice set to prediction. Subsequently, user behavior is applied to mobile web systems and recommendation systems. Although a couple of works are wiped out the sector of prediction supported the user behavior, most of them have addressed only single activity, e.g., click activity and move activity. Considering multiple activities as a user behavior is that the solution.

Disadvantages

• there's no more data influences thanks to a diversity range of link prediction methods.
• Data security is extremely less lack of Hidden data techniques.

PROPOSED SYSTEM

Within the proposed system, the system uses user attributes and relationships to enhance performance of both the user latent interest and link prediction. the idea of random walks suggested a random walker is more likely to go to nodes to which new links are going to be created within the future.

Namely, user relationships are the important characteristics to link prediction. Other evidence of interaction shows that users with similar key words, attribute information, or text content are likely to link to at least one another, motivating the utilization of attribute information for link prediction. Additionally proved that the fusion of the user relationships’ characteristics and user attributes information could improve the performance of link prediction.

Advantages

• More security on network data thanks to Hidden naive Bayes.
• The top users can identify an affective data by Influencing Factors Quantification.