Unified Recall Model for Information Push Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information push methods are cumbersome and inefficient due to the need for constructing multiple recall models and complex analysis to compare item information across channels, leading to suboptimal retrieval of high-click-through-rate (CTR) item information.
Innovation Solution
An information push method that acquires user and item features with CTR information, determines their similarity, and uses this similarity to recall and push item information directly to user terminals, eliminating the need for multiple recall models and simplifying the process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple recall models are constructed to retrieve item information through different channels, then the comprehensiveness of item retrieval is improved, but the system complexity and processing time increase significantly
Solution Approach 1:
The patent merges multiple recall models into a single unified recall model that integrates features from different channels (user features, item features, interaction features). This unified model performs comprehensive retrieval in one pass, eliminating the need for separate recall models while maintaining the comprehensiveness of item retrieval across different channels.
Solution Approach 2:
The unified recall model serves multiple functions that previously required separate models: it can recall items based on user preferences, item characteristics, and interaction history simultaneously. This multi-functional model reduces system complexity while preserving the ability to retrieve diverse relevant items.
2Quantity of substance
If multiple recall models are used to search item information through different channels, then more item information can be retrieved, but the ranking and analysis process becomes cumbersome and inefficient
Solution Approach 1:
The patent combines the retrieval and ranking functions into a unified recall model that outputs directly ranked item information. By integrating multiple feature channels and performing comprehensive scoring within a single model, it eliminates the cumbersome multi-stage ranking process while maintaining the quantity and quality of retrieved items.
Solution Approach 2:
The unified recall model performs preliminary feature integration and scoring within its internal structure before outputting results. This preliminary processing of multiple features (user, item, interaction) within the recall model itself reduces the need for subsequent complex analysis and ranking operations, thereby improving efficiency.
3Measurement precision
If comprehensive scoring and ranking are performed on item information from multiple channels, then the quality of pushed items is improved, but the processing time and computational resources increase
Solution Approach 1:
The patent merges the scoring and ranking operations into the unified recall model's internal processing. The model integrates multiple feature channels and performs comprehensive scoring in a single computational pass, eliminating the need for separate ranking stages. This maintains item quality while significantly reducing processing time.
Data Source
AI summary
The present application discloses an information push method and apparatus, a device, and a storage medium, which relate to the field of intelligent search technology. A specific implementation of the method in the present application lies in: acquiring a user feature and an item feature, where both the user feature and the item feature include click through rate information; determining a degree of similarity between the user feature and the item feature; determining, according to the degree of similarity, item information that is recalled; and pushing the item information to a user terminal. According to the present application, item information having a high click through rate may be acquired through a one-channel recall approach, so that information push efficiency is improved and the pushed item information better conforms to user needs.