Target User Selection Using Similarity and Conversion Probability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing user directing methods for delivering advertisement information are ineffective in achieving accurate user conversion, as they rely solely on similarity models without considering the probability of conversion operations, leading to suboptimal targeting and reduced delivery effectiveness.
Innovation Solution
A method and apparatus that determine the similarity between candidate users and a seed user using a similarity model, predict the probability of conversion operations using a conversion prediction model, and select target users based on both similarity and probability, enhancing the accuracy and effectiveness of user directing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If user directing is performed by calculating similarity between candidate users and seed user using only a classification model, then the delivery process is simple, but the accuracy of target user selection is insufficient and cannot effectively implement user conversion
Solution Approach 1:
The patent combines a classification model with a conversion prediction model into an integrated user directing system. The classification model identifies candidate users based on similarity to seed users, while the conversion prediction model estimates the probability of conversion operations. By merging these two models, the system achieves both user similarity matching and conversion probability prediction, thereby improving target user selection accuracy without excessive complexity increase.
Solution Approach 2:
The patent introduces a new dimension of conversion probability prediction alongside the traditional similarity-based user directing approach. Instead of relying solely on the classification model's similarity metrics, the system adds a probabilistic dimension by predicting the likelihood of conversion operations. This dimensional expansion transforms the single-criterion directing method into a multi-dimensional evaluation system, significantly enhancing selection accuracy.
2Reliability
If only similar users to seed user are selected as target users, then the directing process is straightforward, but the probability of conversion operation after delivery is low
Solution Approach 1:
The system incorporates feedback mechanisms by using historical conversion data to train the conversion prediction model. The model learns from past user behaviors and conversion outcomes, continuously improving its ability to predict which users are most likely to perform conversion operations. This feedback-driven approach enhances the reliability of conversion probability predictions while maintaining operational simplicity through automated model training and updating.
Solution Approach 2:
The patent performs preliminary conversion probability prediction for all candidate users before the actual information delivery. By pre-calculating conversion probabilities using the conversion prediction model, the system identifies high-potential target users in advance. This preliminary action allows the system to prioritize users with higher conversion likelihood, improving the overall probability of successful conversion operations while streamlining the directing process.
3Productivity
If a single classification model is used for user directing, then the system structure is simple, but the delivery effectiveness is suboptimal
Solution Approach 1:
The patent segments the user directing task into two distinct functional components: user similarity assessment (classification model) and conversion probability prediction (conversion prediction model). This segmentation allows each model to specialize in its specific function, with the classification model handling user categorization and the conversion prediction model focusing on predicting conversion likelihood. The segmented approach improves delivery effectiveness by addressing both user matching and conversion potential, while the modular structure manages complexity through clear functional separation.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
Embodiments of the present invention provide a target user directing method and apparatus, including: determining a similarity between candidate users and a seed user by using a similarity model; predicting, by using a conversion prediction model, a probability that the candidate users perform a predetermined conversion operation on to-be-delivered information; and selecting a target user of the to-be-delivered information from the candidate users according to the similarity and the probability. The embodiments of the present invention further disclose a computer storage medium.