Zero-Shot API Selection via LLM Translation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches restrict direct interaction between non-specialist users and applications due to the need for technical expertise in understanding structured syntax and domain-specific knowledge of APIs, and struggle with translating complex natural language requests involving multiple operations into API calls.
Innovation Solution
A framework using zero-shot classification and generative LLMs to interpret human instructions, select relevant tools, plan actions, and execute API calls to fulfill user requests, enabling non-specialist users to interact with computing systems through natural language.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If structured syntax and domain-specific API knowledge are required for application interaction, then system reliability and precision are improved, but ease of operation deteriorates as non-specialist users cannot directly interact with applications
Solution Approach 1:
The patent introduces a translation layer that acts as an intermediary between natural language queries and structured API calls. This mediator component automatically translates user-friendly natural language into precise API syntax, eliminating the need for users to directly learn and use complex structured syntax while maintaining accurate API invocation
Solution Approach 2:
The patent replaces the mechanical requirement of manual API syntax construction with an automated translation system. Instead of requiring users to mechanically assemble structured API calls, the system uses natural language processing to automatically generate the correct API invocations, substituting human mechanical effort with automated intelligent translation
2Ease of operation
If conventional approaches are used for natural language to API translation, then simple single-API requests can be handled, but complexity increases when multiple operations and computations are involved
Solution Approach 1:
The patent segments complex natural language requests into multiple simpler sub-requests, each corresponding to a single API call. The translation system processes these segmented requests sequentially or in parallel, allowing complex computations to be broken down into manageable API operations while maintaining the simplicity of the natural language interface
Solution Approach 2:
The patent implements a dynamic translation approach where the system adapts its translation strategy based on the complexity of the input request. For simple requests, direct translation is used; for complex requests involving multiple operations, the system dynamically generates multi-step execution plans, adjusting the translation process to the specific needs of each request
Data Source
AI summary
One example method includes receiving a plain language request from a user, with a zero shot classification model, determining, and selecting, application program interfaces (APIs) needed to fulfill the plain language request, with a large language model (LLM), processing the APIs to generate a preliminary plan which comprises an order in which the APIs must be called in order to fulfill the plain language request, for each of the APIs in the preliminary plan, determining respective parameters, and combining the parameters with the APIs to generate a final plan which, when executed, fulfills the plain language request.