Two-Tower Retrieval Ranking for Partial Query Personalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional retrieval systems struggle to identify relevant items based on a small number of input characters, leading to increased user effort and resource consumption due to large candidate item inventories and inefficient ranking processes.
Innovation Solution
A personalized retrieval system utilizing a two-tower deep machine learning model processes user and item features to calculate relevancy scores, enabling efficient selection and ranking of candidate items, even with partial queries, by employing a context tower for user and query features and an item tower for item features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional retrieval systems process large candidate item inventories, then they can provide comprehensive search results, but they consume excessive computational resources and time
Solution Approach 1:
The system segments the candidate items into multiple clusters based on their features, and processes each cluster separately using appropriate retrieval strategies. This divides the large inventory into manageable segments that can be processed more efficiently while maintaining comprehensive coverage of relevant items.
Solution Approach 2:
The system dynamically adjusts retrieval parameters such as candidate selection thresholds and processing depth based on query characteristics and user context. This allows the system to optimize resource allocation for each specific retrieval task, processing fewer items when high confidence is achieved and more items when necessary for completeness.
2Measurement precision
If conventional systems require complete queries for accurate retrieval, then they achieve better precision, but they increase user effort and interaction time
Solution Approach 1:
The system performs preliminary retrieval operations on partial queries by leveraging user profile data, historical interaction patterns, and contextual information. This preliminary action provides initial results while the user is still typing, reducing the perceived interaction time while maintaining accuracy through subsequent refinement as the query completes.
Solution Approach 2:
The system provides feedback in the form of preliminary retrieval results during query input, allowing users to see potential matches early and potentially modify their query if the results are already satisfactory. This feedback loop reduces interaction time by allowing early termination when results meet user needs.
3Reliability
If the system processes more candidate items to ensure relevance, then retrieval quality improves, but computational complexity increases
Solution Approach 1:
The system applies different processing qualities to different candidate items based on their relevance indicators. High-priority items receive more thorough processing and analysis, while lower-priority items receive streamlined processing. This local differentiation maintains retrieval quality for important items while reducing overall computational complexity.
Solution Approach 2:
The system processes a partial set of candidate items in detail while using scoring and filtering to identify additional relevant items without exhaustive processing. This approach achieves sufficient retrieval quality by focusing computational effort on the most promising candidates rather than uniformly processing all items.
Data Source
AI summary
Disclosed are system, method and/or computer program product embodiments that retrieve items for a user based on a query using a two-tower deep machine learning model. An example embodiment provides input to a context tower, wherein the input includes the query and one or more of a query embedding corresponding to the query or a graph user embedding corresponding to the user. The context tower generates a context embedding in a vector space based on the input. The model determines a measure of similarity between the context embedding and each of a plurality of item embeddings in the vector space that are generated by an item tower and represent a plurality of candidate items. A relevancy score is calculated for each candidate item based on the measure of similarity between the context embedding and the corresponding item embedding. The relevancy scores are used for item retrieval and/or ranking.