Fortune Analytics Language Model with Query Regeneration and Answer Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
General-purpose large language models struggle to maintain high accuracy and contextual fidelity in domain-specific industry scenarios, particularly in business-related contexts requiring financial literacy and organizational decision-making processes, lacking robustness and domain-specific reasoning capabilities.
Innovation Solution
The implementation of a Fortune Analytics Language Model (FALM) that leverages a curated knowledge base from professional journalism and performs fine-tuning for domain-specific question and answering, incorporating article-based, topic-based, metric-based, and persona-based question-answering capabilities, with guardrails for ethical and accurate responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If general-purpose large language models are used, then broad language understanding capability is achieved, but domain-specific accuracy and contextual fidelity deteriorate
Solution Approach 1:
The patent segments the language model into two distinct components: a general-purpose LLM for broad language understanding and a domain-specific fine-tuned model for industry-specific accuracy. This segmentation allows each component to excel at its designated function, resolving the contradiction between versatility and domain precision.
Solution Approach 2:
The patent applies local quality by enhancing specific parts of the system with domain-specific knowledge through fine-tuning on industry datasets. The general LLM maintains broad capabilities while the fine-tuned portions provide localized domain expertise, achieving both versatility and accuracy in different contexts.
2Adaptability or versatility
If general-purpose large language models are used, then open-domain processing capability is achieved, but contextual fidelity in industry scenarios deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training the model on general corpora to establish broad language capabilities, then subsequently fine-tuning it on industry-specific datasets. This sequential preparation ensures the model has both open-domain processing ability and industry-specific contextual fidelity before deployment.
Solution Approach 2:
The patent introduces an intermediary fine-tuning layer between the general LLM and industry applications. This intermediary component acts as a bridge, translating general language capabilities into domain-specific contextual understanding while maintaining reliability in industry scenarios.
3Measurement precision
If domain-specific fine-tuning is performed, then industry accuracy is improved, but model complexity increases
Solution Approach 1:
The patent segments the training process into distinct phases (pre-training and fine-tuning) and maintains separate model components. This segmentation manages complexity by organizing the complex fine-tuning process into manageable stages while preserving industry accuracy.
4Reliability
If multiple fine-tuning aspects are implemented, then domain-specific reasoning capability is improved, but training time and computational resources increase
Solution Approach 1:
The patent performs preliminary pre-training on general corpora to establish foundational language skills before conducting more targeted fine-tuning. This preliminary action reduces the overall training time by avoiding redundant learning, allowing the model to acquire domain-specific reasoning capabilities more efficiently.
Data Source
AI summary
Method, application server, and non-transitory computer-readable medium for question-and-answer generation using a fortune analytics language model (FALM) are disclosed. In an aspect, a pre-trained large language model (LLM) is generated using information associated with a particular practice area. Further, fine tuning of the pre-trained LLM is performed for a plurality of different aspects to generate the FALM. A user query is then received from a client device. A plurality of new queries are then regenerated based upon the user query. Furthermore, the new queries are executed using the FALM to receive a plurality of answers, each answer of the plurality of answers corresponds with a new query of the new queries. Each answer is then ranked. Also, one or more answers are presented on a display of the client device, the one or more answers are displayed according to a predefined criterion and based upon the ranking.