| Interdisciplinary research has become an important research model in academia. The cross study between computer science and other disciplines are particularly important. In order to further promote the development of interdisciplinary research, this paper looks at the application of Learning to Rank in management science, and developing a new type of multivendor selection model. Learning to Rank(LETR) is a novel learning method developed in recent years, whose task is to rank objects. The application areas include information retrieval, recommendation systems, and e-commerce systems etc. However, to our best knowledge, there is no related work which applied Learning to Rank theory in the problem of supplier selection. Inspired by the mature theory of Learning to Rank, we apply them to solve the multiple vendor selection problem. The main research contents are as follows:Firstly, In order to support the decision of supplier selection, we also formalized a number of features including the supplier features, order features, and the relationship features between orders and supplier. These features can be naturally integrated into the proposed selection algorithm.Secondly,we proposed a unified framework for multi-vendor selection driven by specific orders. Based on this framework, we further propose three selection algorithms utilizing the theory of Learning to Rank.Firstly we propose selection algorithm based on basic of the neural network; Secondly sort neural network an upgrade method is proposed namely Lambda Rank. Finally combining the thought of multiple additive regression tree, Lambda MART sort algorithm is proposed in order to further optimizing the Lambda Rank sort algorithem.Finally, large scale simulations are conducted to verify the effectiveness of the framework and the validity of the models presented in this paper. In order to carry out the simulation, we designed a large number of orders-supplier data and employ a large number of volunteers to do data annotations. |
PDF Full Text Request