Two-Stage Friend Recommendation Based on Network Alignment and Series-Expansion of Probabilistic Topic Model

Two-Stage Friend Recommendation Based on Network Alignment and Series-Expansion of Probabilistic Topic Model


Precise friend recommendation is an important problem in social media. Although most social websites provide some kinds of auto friend searching functions, their accuracies are not satisfactory. In this paper, we propose a more precise auto friend recommendation method with two stages. In the first stage, by utilizing the information of the relationship between texts and users, as well as the friendship information between users, we align different social networks and choose some “possible friends”. In the second stage, with the relationship between image features and users we build a topic model to further refine the recommendation results. Because some traditional methods such as variational inference and Gibbs sampling have their limitations in dealing with our problem, we develop a novel method to find out the solution of the topic model based on series expansion. We conduct experiments on the Flickr dataset to show that the proposed algorithm recommends friends more precisely and faster than traditional methods.

PROJECT OUTPUT VIDEO: (Click the below link to see the project output video):


  • Existing multi-stage recommendations are usually applied to find some patterns of users or items.
  • In an existing system, a two-stage mobile recommendation is proposed to help users find the correct events. The first stage clusters people according to their profile similarity and the second stage discovers the event participating pattern.
  • The another existing system designs the first stage to find some related resources that one user requires, and the second stage is used to find some patterns that the user might prefer from the previous stage for further recommendation. Both the systems can handle the cold-start problem well but do not consider much about the cross-domain problem.


  • Traditional friend recommendations widely applied by Facebook and Twitter are often based on common friends and similar profiles such as having the same hobbies or studying in the same fields. These methods usually provide a long ranked possible friend list, but the recommendation precision is usually not satisfactory due to its complexity.
  • Co-clustering method lacks the ability to tell the intimacy distance between two individuals exactly but only to group people roughly with similar properties, and thus cannot make precise recommendation.
  • The presence of so many unknown variables not only greatly increases the complexity of the algorithm, but also leads to other problems such as over-fitting or redundancies.


  • In this paper, we approach this recommendation problem in a different way by utilizing the multi-domain information in different stages for a more precise recommendation.
  • In the first stage, based on the correlation of different networks, we align the tag-similarity network to friend network to obtain a possible friend list. Specifically, we consider each user as one node in a graph, and we crawl the uploaded tags from each user and calculate the tag similarity between any two users as the edges to form a tag-similarity network.
  • In the second stage, to overcome the problem that the mass election considering only the tag information might not be precise, we build a topic model to illustrate the relationship between user’s friend making behaviour and the image features they have uploaded. This stage refines the list obtained in the first stage. The main reason for applying a topic model in our second stage lies in the fact that the topic model has the ability to tell on what probability a user would prefer a photo/item/friends.


  • Compared with some previously cross-domain topic models, our model is more compact with less parameters, which leads to some computational convenience.
  • Our proposed method provides a way to describe the whole distribution of the social network, to perform a better recommendation.
  • As far as we know,this is the first time to solve a topic model from the aspect of integral series expansion. We also make comprehensive experiments to show the effectiveness of our method.




  • System : Pentium Dual Core.
  • Hard Disk : 120 GB.
  • Monitor : 15’’ LED
  • Input Devices : Keyboard, Mouse
  • Ram : 1 GB


  • Operating system : Windows 7.
  • Coding Language : JAVA/J2EE
  • Tool : Netbeans 7.2.1
  • Database : MYSQL


Shangrong Huang, Jian Zhang, Dan Schonfeldy, Lei Wangz, and Xian-Sheng Hua, “Two-Stage Friend Recommendation Based on Network Alignment and Series-Expansion of Probabilistic Topic Model”, IEEE Transactions on Multimedia, 2017.

About the Author