Notification Text Refinement Using Generation and Response Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for generating advertising texts lack effectiveness in predicting user response, necessitating improved approaches for generating and optimizing notification texts based on user engagement metrics.
Innovation Solution
An information processing system that integrates a generation model for text modification and a prediction model to generate revised drafts of notification texts, utilizing a prediction model to assess user response metrics like click-through rate, enabling iterative refinement of the text based on user engagement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a machine learning model is used to generate advertising texts, then text generation capability is improved, but prediction accuracy of user response deteriorates
Solution Approach 1:
The system divides the single model into two specialized models: a generation model (LLM) for creating advertising texts and a prediction model for accurately predicting user response metrics. This segmentation allows each model to excel at its specific function, resolving the contradiction between generation capability and prediction accuracy.
Solution Approach 2:
The prediction model acts as an intermediary between the generation model and the final advertising text output. It evaluates generated texts and provides feedback for optimization, enabling accurate prediction of user response while maintaining the generation model's creative capabilities.
2Reliability
If iterative refinement of notification texts is performed, then user engagement metrics are improved, but processing time increases
Solution Approach 1:
The system implements a feedback loop where the prediction model evaluates generated texts and provides guidance for improvement. This automated feedback enables iterative refinement that improves user engagement metrics while reducing the need for manual trial-and-error, balancing processing time with result quality.
Solution Approach 2:
The prediction model performs preliminary evaluation of generated texts before final deployment. By predicting user response metrics in advance, the system can identify and refine promising texts early in the process, reducing overall processing time while improving final engagement metrics.
Data Source
AI summary
An information processing method executed by one or more processors includes inputting a prompt into a machine learning generation model configured to output, when receiving an instruction for generating data. The prompt includes one initial draft of a notification text to be sent to one or more user terminals and includes a modification instruction for modifying the one initial draft. The information processing method includes acquiring a revised draft obtained by the generation model by modifying the one initial draft in accordance with the modification instruction. The information processing method includes inputting the revised draft into a machine learning prediction model configured to output, when receiving the revised draft, a prediction result for an index indicating response from a user in a case in which the revised draft has been received. The information processing method includes acquiring the prediction result that has been output by the prediction model.