An AI-based system for answering educational questions

Abstract

An AI-based system for answering educational questions (100), comprising: a user interface module configured to receive a natural language query from a user; a module for generating multiple queries, configured to process the received query using a large language model to generate a variety of semantically reworded queries; a semantic retrieval module that includes a vector database retriever configured to retrieve relevant documents based on semantic similarity to the reformulated queries; a keyword-based retrieval module that includes a BM25 retriever configured to retrieve documents based on keyword matches with the original query; a ranking and aggregation module configured to combine and evaluate the outputs of both the semantic retrieval module and the keyword-based retrieval module using a weighted scoring mechanism; a response generation module configured to process the evaluated documents and the original request using a large language model to generate a contextually relevant response for the user.

Description

[0001] The present invention relates to the field of artificial intelligence (AI) and information retrieval systems, in particular an AI-based system for answering educational questions.

[0002] Given the exponential growth of digital academic content, students and instructors often struggle to find accurate and contextual information. Existing technologies in the field of question-answering systems and academic search tools largely rely on traditional keyword-based search methods such as BM25 or simple implementations of Retrieval-Augmented Generation (RAG). While conventional search systems are effective at identifying specific keywords, they often fail to capture the semantic intent behind users' search queries. Furthermore, current RAG systems are often unable to generate various rewordings of user input, resulting in limited search coverage.A significant limitation of many current solutions is the inability to process multiple vector embeddings per document, as well as the lack of intelligent document evaluation based on keyword distribution within the content. These shortcomings impair the performance of academic information retrieval tools, particularly in domain-specific contexts such as educational institutions, where terminology and document formats can be complex and diverse. There is a need for a more intelligent, hybrid solution that combines the strengths of semantic search and keyword-based indexing while simultaneously improving the understanding of search queries through advanced language models. To address this problem, the present invention provides an AI-based query system for educational topics.

[0003] The system is designed to efficiently retrieve academic information using a hybrid Retrieval-Augmented Generation (RAG) approach.

[0004] The system integrates both semantic and keyword-based search mechanisms by combining a multi-query retriever with a BM25 retriever.

[0005] The system improves the relevance and accuracy of academic answers by using a large language model to reformulate search queries.

[0006] The system minimizes the time and effort that users have to spend manually searching through extensive scientific documents and resources.

[0007] The system enables scalable, AI-based educational support suitable for use in colleges and universities.

[0008] The system improves the user experience through a domain-specific chatbot that can process and answer complex academic queries using a weighted combination of multiple search techniques.

[0009] In one embodiment, the present invention provides an AI-based system for answering educational questions. The present invention provides an AI-based educational chatbot system that utilizes a hybrid RAG (Retrieval-Augmented Generation) framework for the accurate and efficient answering of academic questions. The system integrates a multi-query retriever and a BM25 retriever to leverage the strengths of both semantic and keyword-based search methods. In the disclosed system, a user's initial query is first processed by a large language model to generate several reformulated versions. These reformulated queries are then forwarded to a vector database for semantic search, while the original query is simultaneously processed by a BM25-based keyword retriever.The results from both search mechanisms are evaluated and combined with equal weighting to identify the most relevant academic documents. The aggregated search results are then passed to a large language model to generate a final answer, ensuring that both the intent and contextual meaning of the user's question are accurately captured. This hybrid approach improves search coverage and relevance, significantly reduces manual search effort, and delivers precise answers from a large academic corpus. The system is particularly well-suited for use in educational institutions to facilitate students' and faculty's access to institutional knowledge, manuals, FAQs, and course-related content, thus demonstrating both its economic viability and industrial applicability.

[0010] The invention is explained again below with reference to the figure. This shows: Fig. : a representation of the complete architecture of an AI-based system for answering educational questions. Fig. : a representation of the workflow architecture of an AI-based system for answering educational questions

[0011] Fig. This document presents a representation of the complete architecture of an AI-based system for answering educational queries. The disclosed system (100) significantly reduces the time and cognitive effort required to search large volumes of academic content. It intelligently combines the strengths of traditional keyword search and modern semantic search, while the process of reframing search queries ensures broader coverage and a deeper understanding of the user's intent. The invention is scalable and can be extended to other fields such as healthcare, law, e-commerce, or customer service, where domain-specific document retrieval is crucial. Furthermore, the architecture supports modular integration, enabling future enhancements such as personalized responses, context retention, and multilingual support.

[0012] Fig.Figure 101 illustrates the workflow architecture of an AI-based system for answering educational questions according to an embodiment of the present invention. The process begins when a user submits a natural language query via an input interface. This query is forwarded to a multi-query generation module based on a large language model. This module generates several semantically varied versions of the original query to improve the robustness and coverage of the search. The set of reformulated queries is then processed by a vector-based semantic retriever. This retriever compares the queries with document embeddings stored in a vector database and identifies semantically similar documents or passages from an academic corpus.In parallel, the original query is forwarded to a keyword-based retriever that uses the BM25 algorithm. This retriever searches for documents containing exact or very similar terms, based on traditional inverted indexing methods, and delivers results that show a high lexical match with the original query. The results from both search methods—semantic and keyword-based—are then combined by a results evaluation and combination module. This module weights each search method equally and combines the best results into a unified list. This hybrid ranking approach leverages the advantages of both semantic similarity and term-based matching to improve the relevance of the results.The consolidated set of documents is then passed to a prompt construction module, which formats the selected passages, along with the user's original request, into a structured prompt. This prompt is then processed by a large language model to generate a comprehensive and contextually appropriate response. Reference symbol list 100 System

Claims

[1] An AI-based system for answering educational questions (100), comprising: a user interface module configured to receive a natural language query from a user; a module for generating multiple queries, configured to process the received query using a large language model to generate a variety of semantically reworded queries; a semantic retrieval module that includes a vector database retriever configured to retrieve relevant documents based on semantic similarity to the reformulated queries; a keyword-based retrieval module that includes a BM25 retriever configured to retrieve documents based on keyword matches with the original query; a ranking and aggregation module configured to combine and evaluate the outputs of both the semantic retrieval module and the keyword-based retrieval module using a weighted scoring mechanism; a response generation module configured to process the evaluated documents and the original request using a large language model to generate a contextually relevant response for the user. [2] System (100) according to claim 1, wherein the module for generating multiple search queries is configured to generate at least three reformulations of the original search query. [3] System (100) according to claim 1, wherein the semantic retrieval module uses context-related vector embeddings to identify semantically similar documents. [4] System (100) according to claim 1, wherein the keyword-based retrieval module uses an inverted index to find documents containing exact or near-exact keyword matches. [5] System (100) according to claim 1, wherein the ranking and aggregation module assigns equal weights to the results from the semantic and keyword-based retrieval modules. [6] System (100) according to claim 1, wherein the response generation module comprises a generative, pre-trained, transformer-based language model. [7] System (100) according to claim 1, wherein the user interface is provided via a web-based or mobile chatbot platform configured to receive feedback or additional requests.

Images