Viewport Context Integration for AI Chatbot Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional AI models, such as ChatGPT, lack specialized knowledge and are limited in processing context, making it difficult to provide accurate answers in specialized domains like software development, and they can only handle up to 4000 tokens of information.
Innovation Solution
A context-based interactive service system that uses information from a user's terminal viewport, including text, images, and videos, to generate context information, which is then used to improve the accuracy of AI model responses by integrating relevant data from the current screen, even when scrolling, and utilizing web addresses and links.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a general-purpose AI model like ChatGPT is used to answer user questions, then the model has excellent natural language processing capability, but it lacks specialized domain knowledge and cannot provide accurate answers in specialized areas
Solution Approach 1:
The patent combines a general-purpose AI model with specialized domain knowledge by integrating technical document data into the system. The AI model processes user questions while the system simultaneously retrieves and utilizes relevant information from technical documents, merging general language processing with domain-specific accuracy.
Solution Approach 2:
The patent introduces an intermediary mechanism that acts as a bridge between the general AI model and specialized knowledge. This intermediary retrieves relevant context from technical documents based on the user's question and the current viewport, then provides this contextualized information to the AI model to enhance its answer accuracy.
2Measurement precision
If more context information is provided to the AI model, then the accuracy of answers improves, but the AI model can only process up to 4000 tokens which limits the amount of context that can be added
Solution Approach 1:
The patent extracts only the most relevant context information from technical documents based on the user's question and the current viewport. Instead of providing all available context, the system selectively extracts and transmits only the necessary portions that directly relate to the user's query, staying within the 4000 token limit while maintaining answer accuracy.
Solution Approach 2:
The patent applies local quality by providing different amounts of context information based on specific needs. The system analyzes the user's question and the current viewport to determine which specific portions of technical documents are relevant, then provides customized context information tailored to that specific interaction rather than a uniform amount of context for all queries.
3Measurement precision
If the user reads and understands technical documents manually to find solutions, then the user can find accurate information, but the process requires significant time and effort
Solution Approach 1:
The patent implements self-service by enabling the system to automatically retrieve, analyze, and provide relevant information from technical documents without requiring the user to manually read and understand them. The system autonomously processes the user's question, identifies relevant context from the current viewport and document, and generates accurate answers, freeing the user from time-consuming manual document analysis.
4Ease of operation
If the chatbot uses only the currently visible viewport information, then the context is limited to what is immediately visible, but important information that scrolled out of view may be missed
Solution Approach 1:
The patent applies preliminary action by proactively detecting and capturing information that has scrolled out of the current viewport before the user even asks a question. The system monitors scrolling behavior and pre-retrieves relevant context information from areas that are no longer visible, ensuring that this information is available when the user formulates their query, thus preventing information loss.
Data Source
AI summary
Context-based interactive service providing system and method disclosed. Method of providing context-based interactive service, running on a user terminal, includes executing a chatbot by a chatbot execution unit to display a chatting window in accordance with a user input during execution of an application on a user terminal, generating a context information based on a viewport information of the user terminal by a context information generation unit, receiving a chatting message through the chatting window by a message input unit, transmitting the chatting message and the context information as an input message to a chatbot server by a terminal message transmission unit, and receiving an answer message corresponding to the input message from the chatbot server to display as an answer in the chatting window.