With the rapid development of various Internet services, it has become easier to obtain auxiliary information. The application of such auxiliary information to the recommendation algorithm will improve the recommendation performance, but it also brings new ideas to the modeling ability of the recommendation algorithm. Based on the DeepWalk algorithm, this paper proposes a network representation learning recommendation algorithm based on a random walk. The number of wandering sequences is determined according to the importance of the nodes, and the termination probability is set so that the lengths of the wandering sequences are not the same, which is closer to the real situation. At the same time, in the process of node representation learning, the attribute information of the node is merged, the weight of the attribute information of the node is adaptively adjusted, and the distance between the context node and the central node is considered to obtain more accurate recommendation results. Experimental results on 3 data sets show that the algorithm has good recommendation performance and effectively solves the cold start problem.

Customer segmentation helps the company's strategy formulation and competitiveness enhancement. To better meet customer needs and preferences, companies must recognize the differences of customers and formulate precise marketing strategies. This article focuses on the current customer segmentation background and combines Data mining tools, proposed a multiindex customer segmentation model. Considering the micro and macro perspectives, the traditional indicators are refined, and new segmentation indicators are added. The indicators are weighted by the entropy method. To reduce the time complexity of clustering, factor analysis is used to reduce the data dimension. Finally, the improved K-means clustering algorithm is used to optimize the determination of the K value and the selection of the initial center point to determine the customer segmentation results. The empirical research results on the segmentation of a retailer's membership data show that the improved algorithm is superior to the classic customer segmentation method in terms of clustering compactness and feature division capabilities. With this, it can help companies to improve the level of customer relationship management and the quality of decision-making.