Recommendation method based on social topology for cold-start users

It is very difficult to give recommendations for cold-start user who usually has very sparse historical behavior records. A cold-start recommendation method was proposed based on comparison of topology of social relationships in social networks to improve recommendation effectiveness for cold-start user. Social network contains many social relationships which could reflect user's preference. However, most of existing social network based recommendation methods use only one or a few social relationships of social network, which do not make full use of multiple social relationships; rarely consider how to merge dissimilar social relationships, and could not give satisfactory recommendation in actual environment. In social network the higher weight a kind of social relationship takes, the greater right of recommendations it will have. In order to give accurate recommendations for cold-start user, a social topology based similar user matching method (STSUM) was proposed, Maximum entropy principle was introduced to merge multiple social relationships, and graph pattern matching was used to find similar users for cold-start user. Then recommendations were given according to similar users' records. Social relationship and user data from a real website to show the recommendation effectiveness of STSUM. The experimental results show that STSUM con give accurate recommendations for cold-start user and needs a few training set.