Abstract: With the rapid development of Internet technology, online P2P lending market investment recommendation has become an important research direction in the field of online finance. For potential investors in P2P lending market, there are two key issues which need to be solved. One is how to choose the right investment projects considering their investment needs and preferences, the other is how to reasonably allocate their investment amount to these projects. Previous studies on these two questions mainly focused on default risk prediction of lending projects, global optimal product recommendation and portfolio optimization. With further research, limitations of recommendation model which was designed based on historical transaction data and utility indifference of investors have become increasingly prominent. It does not meet the investment decision needs of investors with different risk preferences. In view of this, firstly, based on the historical data of Prosper platform, this paper establishes the P2P relationship network model and calculates the concept features of the lending projects and investors separately, and then obtains the corresponding conceptual model. Secondly, we build friend network model to capture the mutual influence of investment behavior among investors, excavate investment behavior influence factor and take it as indicator variables of investment interest prediction, then generate a candidate investment recommendation list. Thirdly, in order to optimize the shares of each recommended candidate, individualized portfolio recommended framework is constructed based on expected utility theory considering the influence of risk preference difference on investment demand of investors. Finally, the recommendation results of our model are compared with other benchmark models to comprehensively evaluate its recommend effect. We implement experiments on real datasets of Prosper platform, experimental results demonstrate that our method has better recommendation quality than traditional investment recommendation method.