Search Query Recommendation Using Intent-Shift Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing query recommendation systems fail to accurately account for changing user intent during a series of search queries, leading to irrelevant recommendations and increased computational resources due to unnecessary interactions.
Innovation Solution
A system that utilizes two machine learning models to predict user intent based on a sequence of search queries, determining intent types (specification, generalization, substitution, complement, and irrelevant) and using these to recommend relevant subsequent queries, reducing manual input and conserving computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If query recommendation systems use previous queries to output recommended queries, then the user becomes aware of additional features and manual search steps are reduced, but the recommendations become irrelevant when user intent changes during the search process
Solution Approach 1:
The system dynamically adapts recommendation strategies based on detected user intent changes. When a user's search behavior indicates a shift in intent (e.g., from product specification to brand comparison), the system transitions from recommending queries based on historical patterns to recommending queries that align with the newly detected intent, ensuring continuous relevance throughout the search session
Solution Approach 2:
The system monitors user interactions with recommended queries and search results to detect intent changes. This feedback loop allows the system to identify when recommendations are no longer aligned with user needs and adjust subsequent recommendations accordingly, maintaining reliability even as the search progresses
2Loss of energy
If the system presents inaccurate or irrelevant recommended queries, then computational resources are wasted on additional interactions, but presenting no recommendations at all eliminates this waste
Solution Approach 1:
The system presents a limited number of highly relevant recommended queries rather than numerous potentially irrelevant ones. By focusing on quality over quantity and using intent detection to filter recommendations, the system minimizes computational waste from processing irrelevant queries while still providing enough options to maintain search efficiency
Data Source
AI summary
A sequence of search queries received from a user and the manner in which each successive query is changed is used to determine the intent associated with each query. The resulting sequence of intents is used to predict the next intent of the user, in combination with user interactions with search results and the number of queries of different intents that have occurred in the sequence. The predicted next intent is then used in combination with the sequence of queries to predict and recommend one or more subsequent search queries that correspond with the predicted intent of the user. Recommended queries that correspond to a changed intent of the user avoid presentation of irrelevant information, facilitate performance of searches by the user with less manual input, and in some cases may enable precomputation of search results for recommended queries to enable results to be presented more efficiently.