Generative Query Variant Expansion for Low-Frequency Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing query processing systems struggle with efficiently generating query variants, particularly for low-frequency queries, leading to suboptimal search result retrieval and resource inefficiency.
Innovation Solution
Utilizing a trained generative model, such as a neural network with memory layers, to dynamically generate query variants based on tokens and additional user attributes, enabling efficient and targeted query expansion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If rules-based rewrites and predefined similar queries are used, then query processing is simple and fast, but query variant coverage is limited especially for low-frequency queries
Solution Approach 1:
A trained generative model is introduced as an intermediary component between the query input and search processing. This model generates query variants by applying tokens and user attributes through learned parameters, enabling the system to handle low-frequency and novel queries effectively without requiring complex manual rule sets for each query type
Solution Approach 2:
The generative model is trained offline on query variant specific training data to autonomously learn query transformation patterns. During runtime, it self-generates query variants without requiring real-time human intervention or complex predefined rules, making the system adaptable to diverse query types including tail queries
2Reliability
If multiple query variants are generated and processed simultaneously, then search result relevance improves, but processing power and power consumption increase
Solution Approach 1:
Query variants are generated in advance by the trained generative model before actual search processing occurs. The model uses pre-trained learned parameters to efficiently produce multiple query variants from the input query and user attributes, allowing the search system to process these pre-generated variants rather than generating them on-the-fly during query execution
Solution Approach 2:
The system changes the parameters of the query by generating multiple variants with different token combinations and transformations. The generative model applies learned transformations to create variants that explore different search spaces, improving the likelihood of finding relevant results while managing computational resources through efficient model-based generation
3Productivity
If query variants are generated using a trained generative model, then efficiency and relevance improve, but model training and inference complexity increase
Solution Approach 1:
The generative model is trained offline on query variant specific training data before deployment. This preliminary training phase allows the model to learn effective query transformation patterns, user attribute correlations, and token relationships. During runtime, the pre-trained model efficiently generates query variants using its learned parameters without requiring complex real-time computations
Solution Approach 2:
The generative model serves as an intermediary layer that translates input queries and user attributes into optimized query variants. It uses its trained parameters to mediate between the simple query input and the complex search processing requirements, improving overall system efficiency while managing complexity within the model layer
Data Source
AI summary
Systems, methods, and computer readable media related to generating query variants for a submitted query. In many implementations, the query variants are generated utilizing a generative model. A generative model is productive, in that it can be utilized to actively generate a variant of a query based on application of tokens of the query to the generative model, and optionally based on application of additional input features to the generative model.