On-device query rewriting for mobile search context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search engines struggle to provide relevant results for mobile web searches as they lack understanding of the user's context and preferences, which are often available on the user's mobile device.
Innovation Solution
A method where a mobile device processes search queries by identifying entities and categories, accessing a local knowledge model to select relevant facts, annotating the query with these facts, and transmitting the annotated query to a search engine for improved processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If query rewriting is performed server-side without user context, then processing simplicity is maintained, but search result relevance deteriorates
Solution Approach 1:
The query rewriting process is segmented between client and server: the mobile device performs local query rewriting using user context from the knowledge model, while the server focuses on search execution. This division allows relevant personalization without overburdening either system.
Solution Approach 2:
User context and entity facts are pre-loaded into the mobile device's knowledge model before search queries are generated. This preliminary action enables the device to perform informed query rewriting without requiring complex real-time processing during the search operation.
2Adaptability or versatility
If user-specific knowledge is transmitted to the search engine, then search personalization is improved, but information privacy deteriorates
Solution Approach 1:
Only the specific facts and context relevant to the current search query are extracted from the user's knowledge model and transmitted to the search engine. Complete user profiles or unrelated personal information remain on the device, minimizing privacy exposure while enabling personalization.
Solution Approach 2:
The mobile device acts as an intermediary that processes and filters user information before transmission. It selects and annotates only the necessary facts for the search query, serving as a privacy-protecting mediator between the user's personal data and the external search engine.
3Measurement precision
If extensive user context is stored on the mobile device, then query understanding is improved, but device storage requirements deteriorates
Solution Approach 1:
The knowledge model on the mobile device uses a graph structure where entities and facts are organized locally with relationships defined by types. This structure allows efficient storage and retrieval of user context without requiring redundant data, optimizing the balance between query understanding and storage usage.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, relating to on-device query annotating. In some implementations, a search query is received, and a mobile device identifies a reference to a particular entity and a reference to a category based on the query. A model that is stored on the mobile device and stores one or more facts that are associated with one or more entities is accessed. A subset of facts from among the facts that are stored in the model for the particular entity is selected. The search query is annotated based at least on one or more facts of the subset of facts that are stored in the model for the particular entity. The annotated search query is transmitted, from the mobile device to a search engine, for processing. A result of processing the annotated search query is received by the mobile device.