In recent years, recommender systems have become necessary in overcoming the challenges related to the incredible growth of information. They are used in a wide range of contexts and applications, mainly as prediction tools for customer interest, designed to help customers decide, compare, discover and explore products (Meyer in Recommender systems in industrial contexts, Sciences et Technologies de l’Information, Grenoble, 2012). Therefore, research in the field has focused on improving the efficiency of data processing for instant and accurate recommendations. Recommendation of products, accordingly, does not take into consideration supply chain constraints for deliveries. This can lead to recommendations for products that can be costly or too long to ship to the customer, resulting in an avoidable increase in the stress on the supply chain. This paper addresses the problem of considering delivery constraints in product recommendations. The objective is to shift demand toward products that can be delivered using the current network state without additional resources in a given time window, perimeter and with a minimum acceptable profit, in the context of e-commerce. To achieve this goal, we propose a methodology to adjust product recommendations in order to shift customers’ interests towards particular products with consideration for remaining unit loads of scheduled deliveries. For this, quasireal-time information about the supply chain is taken into consideration to improve the number of shippable products in the recommendation list, resulting in a possible improvement in truck-load utilization, lower operation costs and reduced lead-times for delivery. This method works in two stages: the first stage is the computation of the recommendation with traditional recommendation systems, and the second stage is recommendation adjustments in four phases that consider the evaluation of active trucks, evaluation of physical constraints for transportation, evaluation of the profits associated with adding a pickup/delivery to a scheduled tour for each recommended item and adjustment of recommendation scores. A sensitivity analysis of the impact of the recommendation adjustment on the recommendation list has been conducted for each of the parameters considered in the proposed method: time window, perimeter radius and minimum acceptable profit. Various experimental results prove that the method permits increasing the number of recommended products that can be shipped using the available resources within a given perimeter radius, time window and minimum profit.

近年来，推荐系统已成为克服与信息惊人增长有关的挑战所必需的系统。它们被广泛用于各种环境和应用中，主要用作满足客户兴趣的预测工具，旨在帮助客户决定，比较，发现和探索产品（工业环境中推荐系统中的Meyer，Sciences et Technologies de l'Information，格勒诺布尔，2012）。因此，该领域的研究集中于提高即时和准确建议的数据处理效率。因此，产品推荐未考虑供应的供应链约束。这会导致对产品的建议，这些产品的价格昂贵或太长而无法交付给客户，从而导致可避免的供应链压力增加。本文解决了在产品推荐中考虑交货限制的问题。目的是将需求转向可以使用当前网络状态交付的产品，而在电子商务环境下，在给定的时间范围，周长和可接受的最低利润范围内，无需额外的资源。为了实现这一目标，我们提出了一种方法来调整产品推荐，以便将客户的兴趣转向特定产品，同时考虑计划交付的剩余单位负载。为此，考虑了有关供应链的准实时信息，以提高推荐清单中的可运输产品数量，从而有可能提高卡车装载利用率，降低运营成本并缩短交货时间。此方法分为两个阶段：第一阶段是使用传统推荐系统计算推荐，第二阶段是四个阶段的建议调整，其中包括对活动卡车的评估，对运输的物理约束的评估，对利润的评估与为每个推荐项目的预定游览添加接送服务和调整建议分数相关。针对建议方法中考虑的每个参数，对建议调整对建议列表的影响进行了敏感性分析：时间窗，周长半径和最小可接受利润。