Virtual Assistant Command Clustering for Onboarding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The process of on-boarding actions for virtual assistants is labor-intensive due to the ambiguity and imprecision of natural languages, requiring users to anticipate various commands and terms, leading to inconsistent and incomplete mappings of commands to actions and applications.
Innovation Solution
A method that recommends additional commands, actions, and applications relevant to an initial command group by employing preprocessed document clusters, using search queries to retrieve documents with contextual relevance, and determining command and action scores based on discounted cumulative gain and Jensen-Shannon divergence, thereby assisting users in creating accurate and comprehensive command groups.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual on-boarding of commands is performed by users, then command-to-action mappings can be created, but the process becomes labor-intensive and results in inconsistent mappings
Solution Approach 1:
The system pre-processes a large corpus of documents and clusters them by semantic similarity before the on-boarding process. When a user provides an initial command, the system has already prepared document clusters that can be quickly searched and matched, eliminating the need for real-time analysis and enabling rapid recommendation of relevant commands and actions.
Solution Approach 2:
The patent introduces document clusters as an intermediary layer between user commands and actions. Instead of directly mapping commands to actions, the system uses document clusters that semantically relate to both, serving as a mediator that enables automated recommendation and reduces manual mapping effort while improving consistency.
2Adaptability or versatility
If comprehensive command coverage is achieved through manual on-boarding, then more natural language variations are captured, but the effort and complexity increase significantly
Solution Approach 1:
The system enables automated self-service on-boarding where the system itself generates recommendations for additional commands and actions based on the initial command provided by the user. The system searches document clusters, identifies relevant commands, and presents them to the user for confirmation, eliminating the need for users to manually anticipate all possible command variations.
Solution Approach 2:
The system changes the parameter of command discovery from manual enumeration to automated semantic search. By transforming the on-boarding process from a manual listing task to an automated search and recommendation task based on semantic similarity in document clusters, the system achieves comprehensive coverage with reduced complexity.
3Productivity
If automated command recommendation is implemented, then on-boarding efficiency improves, but computational resources are consumed for document processing
Solution Approach 1:
The system performs computationally intensive document clustering and semantic analysis in advance, before the actual on-boarding interaction. This pre-processing creates ready-to-search document clusters that can be quickly queried during user interactions, shifting computational energy consumption from real-time to offline batch processing.
Solution Approach 2:
The patent segments the large document corpus into smaller thematic clusters based on semantic similarity. This segmentation allows the system to search within relevant clusters rather than processing the entire corpus for each query, reducing computational energy consumption during recommendation while maintaining comprehensive coverage.
Data Source
AI summary
The present invention provides assistance to a user that is on-boarding an action for a virtual assistant. Methods and systems provide, based on an initial command group, recommendations for additional commands, actions, and applications that are relevant to commands included in the initial command group. The initial command group includes commands that include a sequence of natural language words or tokens. Additional commands are determined. The additional commands are relevant to the initial commands and provided to the user as recommended commands to include in the command group. The additional commands are determined by employing the initial commands as search queries to search preprocessed document clusters. Each document included in the document clusters relates to a potential command to include in the command group. In response to receiving a selection of a recommended command, the command group is updated to include the selected recommended command.