Multi-Space Query Projections for Deep Learning Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional retrieval systems struggle to identify relevant items based on short queries, leading to increased user effort and resource consumption due to the need for lengthy queries and extensive processing of large item sets.
Innovation Solution
A deep machine learning model that generates predictions by projecting queries into multiple query embedding spaces, producing query projection embeddings which are then used to calculate relevancy scores or user-item interaction probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional retrieval systems process large item sets with short queries, then system resource consumption increases, but identification accuracy of relevant items deteriorates
Solution Approach 1:
The system segments the item set into multiple clusters based on query projection embeddings, processing only relevant clusters rather than the entire item set. This segmentation enables accurate identification of relevant items while reducing system resource consumption by avoiding unnecessary processing of unrelated items.
Solution Approach 2:
The system projects queries into multiple embedding spaces (different dimensions) to create query projection embeddings. This dimensional transformation allows the system to capture diverse query intents and improve identification accuracy without proportionally increasing resource consumption, as the additional dimensional processing is more efficient than exhaustive item set scanning.
2Measurement precision
If conventional retrieval systems require lengthy queries, then user effort increases, but identification accuracy of relevant items improves
Solution Approach 1:
The system changes the parameter of query representation by creating multiple projection embeddings from short queries. Instead of requiring longer queries to achieve sufficient accuracy, the system transforms short queries into enriched multi-dimensional embeddings, maintaining identification accuracy while reducing user effort and time investment.
Solution Approach 2:
The system creates multiple copies of the query in different embedding spaces through projection. These query projection embeddings serve as enriched representations that capture various aspects of the original short query, enabling accurate item identification without requiring users to invest more time in formulating lengthy queries.
3Measurement precision
If retrieval systems process extensive item sets, then system resource utilization increases, but relevancy score calculation accuracy improves
Solution Approach 1:
The system segments the extensive item set into manageable clusters and processes only those clusters that are relevant to the query. This approach maintains relevancy score calculation accuracy for relevant items while improving system efficiency by avoiding unnecessary processing of irrelevant items in the extensive item set.
Solution Approach 2:
The system performs preliminary projection of queries into multiple embedding spaces before processing item sets. This preliminary action creates query projection embeddings that guide subsequent item processing, enabling accurate relevancy scoring while improving overall system productivity by pre-filtering and organizing the processing sequence.
Data Source
AI summary
Disclosed herein are system, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for generating a prediction based on a query. An embodiment operates by providing a query to a deep machine learning (ML) model. The deep ML model generates a plurality of query projection embeddings by projecting the query into each of a plurality of different query embedding spaces and generates the prediction based at least on the plurality of query projection embeddings. Each of a plurality of query projection embedding layers of the deep ML may generate a corresponding one of the query projection embeddings by applying a hash function associated with the query projection layer to the query to generate a vector representation of the query, and applying a set of weights associated with the query projection layer to the vector representation to generate a query projection embedding in the plurality of query projection embeddings.