Shipping Cost Estimation Using ML Item Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to accurately and efficiently estimate the shipping cost of items in online marketplaces, particularly due to the unknown exact dimensional weight of items, which can lead to losses for the online market if the estimation is incorrect, and lack a seamless integration of local carriers for delivery.
Innovation Solution
A machine learning model is employed to estimate shipping costs by using past transactions and ground truths, incorporating a bias towards heavier estimates to minimize errors, and integrating a Local Shipping feature that allows for programmatic creation of shipping orders with local carriers, ensuring accurate cost estimation and quick delivery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning model is used to estimate shipping costs, then estimation accuracy is improved, but computational complexity and model training requirements increase
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model using historical transaction data and ground truth shipping costs before actual use. The model is trained on past transactions to learn patterns in dimensional weight estimation, so that when the model is deployed, it can quickly estimate shipping costs without requiring complex real-time computation. This pre-computation approach resolves the contradiction by preparing the computational heavy lifting in advance.
Solution Approach 2:
The patent uses copying by creating a virtual representation of the machine learning model that can be deployed without the full training data. The trained model parameters and architecture are copied and deployed as a service, allowing accurate estimation without replicating the entire training process. This enables the system to maintain high accuracy while reducing the computational complexity during actual shipping cost estimation operations.
2Reliability
If bias towards heavier estimates is applied, then error minimization for heavier items is improved, but estimation accuracy for lighter items may deteriorate
Solution Approach 1:
The patent applies parameter changes by adjusting the loss function during model training to incorporate a bias towards heavier estimates. Specifically, the training process uses a weighted loss function that penalizes underestimation of heavier items more heavily, effectively changing the optimization parameters to favor accuracy in the high-value segment. This resolves the contradiction by modifying the training parameters to prioritize reliability for heavier items while maintaining acceptable accuracy across all weight ranges.
3Speed
If local carrier integration is implemented, then delivery speed is improved, but system integration complexity increases
Solution Approach 1:
The patent applies universality by creating a unified shipping order creation system that can handle multiple carrier types (local and national) through a single interface. The system uses a standardized API framework that works across different carriers, allowing the same codebase to support both local carriers (for speed) and national carriers (for broader coverage). This multi-functional design resolves the contradiction by enabling fast local delivery without requiring separate complex integration systems for each carrier type.
Solution Approach 2:
The patent uses an intermediary approach by introducing a standardized shipping order creation system as a mediator between the marketplace and various carriers. This intermediary layer abstracts the complexity of different carrier APIs and provides a uniform interface for order creation, making the system easier to integrate with local carriers while maintaining the ability to work with national carriers. The mediator handles the integration complexity centrally, allowing individual carriers to be added without increasing overall system complexity.
Data Source
AI summary
Disclosed herein are various embodiments for estimating shipping costs of items purchased on an online market. An embodiment operates by detecting a sales transaction between a seller and a buyer for an item via an online marketplace. An item vector representing the item as created by a deep learning model based on a plurality of characteristics associated with the item is received. A set of proximate vectors to the item vector, each of the proximate vectors corresponding to a past listings where a past shipping cost from shipping a product in the past listing is known are identified. A weighted average is calculated from the past shipping cost of each of the past listings. A dimensional weight of the item is estimated based on the weight average, and a shipping cost for the item is estimated based on the estimated dimensional weight prior to providing the item to a carrier.