Search Autocompletion Framework Using Knowledge Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional search applications provide irrelevant autocompletion suggestions due to a lack of customization based on user history and contextual data, and fail to account for available content in the target search space, leading to inefficient search processes.
Innovation Solution
An enhanced search autocompletion framework that uses a knowledge graph to match user input with exact and semantic matches, incorporates historical interaction data, and personalization to generate relevant autocompletion candidates, scored and ranked using machine-learned models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If common or popular words and phrases are used for autocompletion, then the search application can provide suggestions quickly, but the suggestions become irrelevant to the user's current needs and available content
Solution Approach 1:
The system pre-processes and stores user interaction history, search patterns, and content metadata in advance to enable rapid retrieval and matching during autocompletion operations
Solution Approach 2:
The system dynamically adjusts autocompletion parameters based on user profile, search context, and content availability, transitioning from static popular words to dynamic personalized suggestions
2Device complexity
If a simple matching of input text with stored words is used, then the system complexity is reduced, but the ability to provide relevant suggestions based on user history and context is lost
Solution Approach 1:
The system divides the autocompletion task into separate modules: text input processing, user history retrieval, content matching, and suggestion ranking, allowing each component to be optimized independently
Solution Approach 2:
The system introduces intermediate data structures and processing layers between user input and final suggestions, including user profiles, search contexts, and content metadata as intermediary elements
3Ease of manufacture
If autocompletion suggestions are generated without considering available content in the search space, then the generation process is simplified, but the suggestions fail to associate with available content in systems with limited content universes
Solution Approach 1:
The system continuously monitors user interactions and search results to feedback into the autocompletion model, improving the association between suggestions and actually available content over time
Solution Approach 2:
The system creates a universal content index and metadata structure that serves multiple functions: content retrieval, relevance ranking, and autocompletion suggestion generation
Data Source
AI summary
Systems and techniques are disclosed for determining relevant search query autocompletions for presentation to a user that has entered a prefix into a search application interface. A target search space represented as a knowledge graph may be searched to generate a subgraph of nodes representing autocompletion candidates that correspond to the prefix. Further historical and/or interaction data associated with the user and/or other users of the search application may be used to detemir additional autocompletion candidates. A machine-learned model may be trained to score and rank the candidates based on embeddings extracted by the model for the autocompletion candidates. A listing of the autocompletion candidate ordered based on the scores may be presented as autocompletion suggestions to the user.