METHOD OF RESPONSE GENERATION WITH RETRIEVAL-AUGMENTED INFORMATION SEARCH FOR ASSISTIVE COMMUNICATION

Abstract

The issue of improving communication for people with speech impairments is critically important in modern society, where technology can greatly facilitate interaction and integration of such individuals into social processes. Traditional communication methods, such as sign language or written messages, often fail to provide the necessary accuracy and speed of communication, creating barriers in everyday life. Modern technologies, such as large language models (LLMs) and retrieval-augmented generation (RAG) technology, can significantly enhance these processes. They enable the automatic generation of individualized text responses, considering not only the user's query but also the context, making communication more accurate and faster.

The article proposes a method for response generation with retrieval-augmented information search for assistive communication, which integrates relevant information from various sources, including chat history and user profiles. This allows for generating responses that better address the individual needs of each user. The method consists of three stages: context loading, retrieval of relevant information, and generation of multiple response options. A distinctive feature of the proposed method is its ability to consider extensive context and user needs, ensuring high accuracy and personalization of the response.

Experiments conducted demonstrated the method's high efficiency, particularly in the accuracy of responses, which ranged between 85% and 95%. This significantly improves communication for people with speech impairments, reducing social barriers and enhancing the quality of their daily lives.