Abstract: The purpose of this paper is to take advantage of "Internet plus" and big data resources for logistics optimization problem. The data value of e-commerce platform accumulated over the years needs to be further explored to promote e-business decision makers to provide meticulous service according to the different quality of the recipients. The degree of service quality preference of the receiving party is summarized:(1) "no memory" type of receiving party; (2) "memory" type of receiving party; (3) "uncertain" type of receiving party (4) the whole receiving party. In addition, four types of programming solution space are deduced, and then the cost linear function of "no memory" type has an accurate solution for dynamic planning, and the other three types are designed to approximate the particle swarm optimization algorithm. The detailed research idea of dissertation is as following:the perceived quality is embedded into model constraints and used to describe the memory attribute of the data; the problem of distribution resource planning with "no memory" is equivalent to the optimal solution of zero inventory strategy, which is indirectly proved existence of accurate solution by the transformation; "memory" type of receiving party for k times delivery service, there is tolerance for the quality of the delivery service for each purchase, which the total amount of weight accumulation per time does not exceed a limit ensuring this type consumer's loyalty; the overall model of the consignee is a special case of the model of the "uncertain" type of receiving party. The recipient's historical data are clustered by simulation of the results of the formation process to form data memory characteristics and analyzed to obtain the quality requirements of different receiving parties. The e-commercial enterprises are guided to provide the relative precision service according to the different quality sensitivities of the receiving party to enhance the last-kilometers distribution service quality, so as to optimize the quality of the delivery service quality of the decision. The sources of case study come from the fresh supermarkets in the city of Shenyang in China, for cooperating to construct a convenient fresh O2O platform. The experimental data is collected from January 31st, 2014 to January 31st, 2016, with the 10 stores of the company as the receiving party sets containing 35 groups of consignee attributes and 8200 historical data, which are used to verify the performance of the model. One the one hand, by virtue of data-driven research framework, it is easier for operators to provide a relatively accurate delivery service through big data resources from the consignee's quality preferences. Since the uncertain demands for service are satisfied effectively, the utilization rate of quality input costs is higher. One the other hand, an optimization idea is provided to deal with the NP-hard problem by extracting the parts that can be precisely optimized, reducing the NP range and improving the accuracy of the overall solution by the method of feature classification.

Key words: quality sensitive clustering, constraints of quality preference, distribution service quality optimization, data-driven