Abstract: Present social networking services suggest friends to userís based on their social activities, which may not be the most suitable to re?ect a userís taste on friend choice in real life. In this project, we present Friends wall, a semantic-based friend recommendation system for social networks, which present friends to users based on their life styles instead of social activities. By taking benefit of sensor-rich Smartphoneís, Friends wall discovers life styles of users, measures the resemblance of life styles between users, and suggest friends to users if their life styles have high resemblance. Inspired by text mining, in this project we design a userís daily life as life documents, from which his/her life styles are taken out by using the Latent Dirichlet Allocation [LDA] algorithm. We further aim a similarity metric to measure the resemblance of life styles between users, and estimate userís impact in terms of life styles with a friend-matching graph. Upon acquiring a request, Friends Wall returns a list of people with maximum resemblance scores to the query user. At last, Friends Wall incorporates a feedback mechanism to further amend the recommendation accuracy. This project will build Friends Wall and evaluate its performance on both small-scale experiments and large-scale model.

Keywords: Friend recommendation, Mobile sensing, Social networks, Life style, Data Mining, Machine Learning.