Cold-Item Ad Bidding Using Cluster-Level Click Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing advertisement technologies focus on 'warm' items with historical traffic data, neglecting 'cold' items that are new or have little data, leading to inefficient online traffic attraction and suboptimal return on ad spend (ROAS).
Innovation Solution
A system and method for automatically computing and updating advertisement biddings for cold items using a machine learning model trained on BERT embeddings, clustering based on performance metrics and content features, and adjusting bids at the cluster level to balance click distribution and reduce sparsity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If advertisement bidding focuses on warm items with historical traffic data, then bidding accuracy is improved, but coverage of cold items is reduced
Solution Approach 1:
The patent segments items into warm items (with historical traffic data) and cold items (without sufficient historical data). Different bidding strategies are applied to each segment: traditional data-driven bidding for warm items and cluster-based predictive bidding for cold items. This segmentation allows the system to maintain high bidding accuracy for warm items while extending coverage to cold items that would otherwise be neglected.
Solution Approach 2:
The patent introduces cluster-level metrics as an intermediary between individual cold items and bidding decisions. Since cold items lack sufficient individual historical data, the system aggregates them into clusters based on content features and metrics, then uses cluster-level performance data to inform bidding for individual items within those clusters. This intermediary approach enables bidding for cold items by leveraging shared characteristics and aggregated performance data.
2Quantity of substance
If advertisement bidding includes cold items with little data, then item coverage is improved, but bidding accuracy deteriorates
Solution Approach 1:
The patent merges multiple cold items into clusters based on their content features and performance metrics. By combining items with similar characteristics into the same cluster, the system aggregates their performance data to create sufficient statistical basis for bidding decisions. This merging allows the system to extend coverage to cold items while maintaining bidding accuracy through the use of aggregated cluster-level data.
Solution Approach 2:
The patent introduces cluster-level metrics as an intermediary between individual cold items and bidding decisions. Since cold items lack sufficient individual historical data, the system aggregates them into clusters based on content features and metrics, then uses cluster-level performance data to inform bidding for individual items within those clusters. This intermediary approach enables bidding for cold items by leveraging shared characteristics and aggregated performance data.
3Measurement precision
If cluster-level bidding is used for cold items, then bidding accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the bidding process into two distinct phases: cluster formation phase and bidding phase. During cluster formation, items are grouped based on content features and metrics. During the bidding phase, bids are computed at the cluster level using aggregated performance data. This segmentation reduces computational complexity by avoiding individual item-level computations for all cold items while maintaining bidding accuracy through cluster-level aggregations.
Solution Approach 2:
The patent applies partial action by computing bidding metrics at the cluster level rather than for every individual item. Instead of performing full bidding computations for each cold item (which would be computationally expensive), the system computes cluster-level metrics once and applies them to multiple items within the cluster. This partial computation approach significantly reduces computational complexity while maintaining sufficient bidding accuracy.
4Productivity
If cluster-based approach is applied to cold items, then traffic attraction is improved, but system complexity increases
Solution Approach 1:
The patent segments items into warm and cold categories, and further segments cold items into clusters based on content features and performance metrics. This segmentation allows the system to apply specialized traffic attraction strategies: traditional individual-item optimization for warm items and cluster-based optimization for cold items. The clustering approach improves traffic attraction for cold items by leveraging shared characteristics and aggregated performance data, while the segmentation keeps system complexity manageable by treating different item types differently.
Data Source
AI summary
Systems and methods for automatically computing biddings for cold items are disclosed. A first bid value is computed and submitted for a first bidding for a first set of items each having a predicted sales metric larger than a first threshold in a first time period. From the first set of items, a second set of items are selected that have a traffic metric smaller than a second threshold in a second time period. Each item in the second set is assigned to one of a plurality of clusters. A second bid value is computed for each cluster based on cluster-level metrics and a click distribution among the plurality of clusters. An updated bid value is computed, for each item in the second set, to be submitted for a second bidding, based on the second bid value for a cluster to which the item is assigned.