Query Auto-Completion Ranking Model Using User Habit Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing query auto-completion systems do not effectively consider user habits when ranking search suggestions, leading to increased input cost and reduced efficiency for users.
Innovation Solution
A ranking model is developed that uses positive and negative example pairs from user query logs to learn user preferences, incorporating a prefix embedded network, POI embedded network, and ranking network to maximize similarity between query prefixes and preferred POIs, thereby improving suggestion ranking based on user habits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If POIs are ranked in query completion suggestions based on search popularity, then popular POIs are always ranked higher, but user input cost increases and search efficiency decreases when the user's intended POI is ranked lower
Solution Approach 1:
The system pre-processes user query logs to extract user habits and preferences before the actual query completion task. By analyzing historical data in advance and building a trained model that captures user behavior patterns, the system prepares the ranking criteria beforehand, enabling it to quickly and accurately rank POIs according to individual user preferences rather than generic popularity metrics.
2Ease of operation
If the same query completion suggestions are provided for all users, then the system is simple to operate, but it does not conform to individual user habits
Solution Approach 1:
The system transitions from providing uniform query completion suggestions to delivering personalized suggestions tailored to each user's specific habits and preferences. By analyzing individual user query logs and training separate models for different users, the system adapts the ranking criteria to local user characteristics, ensuring that each user receives suggestions that conform to their personal search behavior patterns.
Solution Approach 2:
The system dynamically adjusts the ranking parameters of query completion suggestions based on user behavior patterns extracted from query logs. Instead of using fixed popularity-based ranking, the system changes the ranking parameters to reflect individual user preferences, such as frequently searched POIs, typical search patterns, and historical selection behavior, thereby making the suggestions adaptable to different users.
3Productivity
If a ranking model considering user habits is implemented, then query completion suggestions better conform to user habits, but the device complexity increases
Solution Approach 1:
The system divides the complex task of query completion ranking into separate functional modules: a query log analysis module that extracts user habits, a model training module that builds ranking models, and a suggestion generation module that applies the models. By segmenting the system into these independent components, each with a specific function, the overall complexity is managed more effectively while maintaining high search efficiency.
Solution Approach 2:
The system introduces a trained ranking model as an intermediary between the raw query logs and the query completion suggestions. This intermediary model processes and transforms user behavior data into personalized ranking criteria, enabling the system to deliver habit-conforming suggestions without requiring complex real-time analysis during the query completion process itself.
Data Source
AI summary
The present application discloses a method for building a ranking model, a query auto-completion method and corresponding apparatuses, which relates to the technical field of intelligent search. An implementation includes: acquiring from a POI query log a query prefix input when a user selects a POI from query completion suggestions, POIs in the query completion suggestions corresponding to the query prefix and the POI selected by the user in the query completion suggestions; constructing positive and negative example pairs using the POI selected by the user and the POIs not selected by the user in the query completion suggestions corresponding to the same query prefix; and performing a training operation using the query prefix and the positive and negative example pairs corresponding to the query prefix to obtain the ranking model.