Chatbot «Eva»

PERSONALIZATION OF LEARNING IN AN EDUCATIONAL ONLINE ENVIRONMENT

Maksym Poltoratsky
Olha Konnova

Our team

Member 1
O. V. Spivakovskyi
Author of the idea and methodology
Member 2
M. O. Vinnyk
Responsible for financial planning
Member 3
O. I. Lemeschuk
Project manager
Member 4
M. Yu. Poltoratsky
Knowledge System Developer/Training artificial intelligence models
Member 7
M. M. Fomichev
QA engineer
Member 5
O. V. Konnova
Full stack developer
Member 6
I. V. Karpov
Specialist in training artificial intelligence models
Member 8
T. O. Cherkashina
Expert in the field of educational management

Project idea

Our project introduces a comprehensive methodology for creating intelligent assistants powered by artificial intelligence using the example of an education management chatbot. However, the proposed approach is universal and can be used to create intelligent assistants in any industry. In this presentation we describe the concept of development of the assistant and the technologies that were used to create it.

An intelligent chatbot for personalized educational management «Eva» was created

Eva Logo

An educational management chatbot is used at Kherson State University to perform various administrative and educational tasks:

  • Assistance in the formation of an individual educational trajectory and the study plan of a student of higher education, its implementation during the entire period of study;
  • Informing about the conditions and possibilities of practice-oriented training and research work according to the educational program chosen by the applicant;
  • Recommending to the applicant of higher education to take into account the results of informal education and academic mobility in higher education institutions of Ukraine or abroad when forming an individual study plan;
  • Advising students on the need to observe the norms of academic integrity and scientific ethics;
  • Informing the student of higher education about the procedure for solving and preventing conflict situations;
  • Advising students on the university's internal business processes.

TECHNOLOGIES

To develop an intelligent ChatBot we will use the following list of technologies:

Knowledge model: RDF
Development of the educational management model

Translator RdfToJsonl: Python, SPARQL
Automatic translation of RDF model to JSON. Azure AI Studio accepts files in JSONL format for creating and configuring artificial intelligence models, including chatbots

Platform: Microsoft Azure ML
Microsoft Azure Machine Learning is a cloud platform for creating, training and implementing artificial intelligence models. It provides tools for developing and experimenting with various machine learning and deep learning algorithms, as well as for automating model training processes.

Microsoft Azure Machine Learning

The advantages of Microsoft Azure include:

  • Wide functionality, a wide set of tools for the development, training and implementation of machine learning models. From data processing to creating models and monitoring their performance - all this is available on one platform.
  • Integration with other Azure services: Azure Machine Learning integrates with other Microsoft Azure cloud services, such as Azure Data Lake, Azure SQL Database, and Azure Cognitive Services. It allows you to easily process, store and analyze your data.
  • Simplified work with structured and unstructured data: The platform has tools for working with different types of data, including structured, semi-structured and unstructured data.
  • Advanced scaling capabilities: Azure Machine Learning provides advanced scaling capabilities for processing large amounts of data and training complex models.
  • Wide selection of integrated models and frameworks: The platform supports various machine learning frameworks such as TensorFlow, PyTorch, Scikit-learn and others.

Preparation and presentation of data

To create a model of educational management, we used data obtained from the following sources:

  • Job search sites (Work.ua, Robota.ua, and others)
  • Educational programs implemented at Kherson State University
  • Syllabi of educational components/learning disciplines

Having systematized the indicated data from job search sites (Work.ua, Robota.ua and others), as well as educational management documents, we developed a matrix of correspondence of vacancies with the necessary skills that are put forward to it, and educational components that cover the formation of these skills and competencies.

Matrix of correspondence between disciplines, skills and vacancies
Specialty: Software Engineering

Skills/Disciplines Formal methods of software engineering Full stack development Software modeling and verification Databases of distributed information systems
Create and write technical documentation
Modeling language:UML
Modeling language:sysML
Frontend development
Backend development
Verification and optimization of software
Developing database

The process of collecting and preparing data for chatbot training

The process of collecting and preparing data for chatbot training includes the following stages:

  • Data collection and preparation
  • Development of a knowledge engineering model (RDF model)
  • Broadcasting of the RDF model in the JSONL format
  • Chatbot training in the Microsoft Azure studio

RDF notation


					//Subject
					
					
					Formal methods of software engineering  
					
						24
					
					 24 
					
						A course focusing on formal methods used in software engineering
						for rigorous software development
					
					https://Link to the syllabus
					  
					

					//Vacancy
					
					
					Software Engineer  
					
						A professional involved in the design, development, testing,
						and maintenance of software systems or applications, utilizing programming languages, tools, 
						and methodologies to meet project requirements.
				    
					  
					

					//Skill
					
					
					
						Modeling languages:UML
					  
					
						A proficiency in Unified Modeling Language (UML), a standardized 
						general-purpose modeling language in the field of software engineering for visualizing, 
						specifying, constructing, and documenting the artifacts of software systems
					
					
				

Discipline<SFMSE> --(Provide)-->Skills[CUML,CSML,…]--(necessaryFor)--> Vacancy<VSE>

A fragment of the RDF model of Educational Management

The matrix is an important tool in the context of knowledge engineering for a bot. It helps to determine which specific skills and competencies are necessary for various vacancies. In addition, this approach will make it possible to improve the integrity of the model.

To train the algorithm, we chose the babbage-002/davinci-002 format for data presentation. Therefore, we are faced with the task of converting the resulting model into RDF, a JSONL format that Azure AI Studio accepts. Example of working JSONL, valid for training in Azure AI Studio:

{"prompt": "Good afternoon, please tell me at which address to pick up the diploma from the KSU?", "completion": "Good afternoon, Ivano-Frankivsk, 14 Shevchenko Street."}

To transform the model, a special translator was developed in the Python programming language using special SPARQL queries.

					
							# SPARQL query
							query = f"""
							PREFIX rdf: 
							PREFIX rdfs: 

							SELECT ?subclass_label ?comment
							WHERE {{
								?subclass rdfs:subClassOf  ;
								rdfs:label ?subclass_label ;
								rdfs:comment ?comment .
							}}
							"""
						
					

The next step is to save the query results in JSONL format, which was presented above:

						
						# Execute SPARQL query and get results
						result = g.query(query)

						# Save results to a text file
						with open(f"{search_criteria}.txt", "w", encoding="utf-8") as file:
							for row in result:
								result_dict = {
									"prompt": str(row.subclass_label),
									"completion": f"It is {search_criteria}
									                    .{str(row.comment)}"
								}
								file.write(str(result_dict) + "\n")


						
					

Training

The result of the parser was a JSONL file ready for use in Azure AI Studio. Below is a fragment of such a file:

 {'prompt': 'Modeling languages:UML', 'completion': 'It is Competencies. A proficiency in Unified Modeling Language (UML), a standardized general-purpose modeling language in the field of software engineering for visualizing, specifying, constructing, and documenting the artifacts of software systems.'}
 {'prompt': 'Modeling languages:sysML', 'completion': 'It is Competencies. A proficiency in Systems Modeling Language (SysML), a general-purpose modeling language for systems engineering applications.'}

Models page

Playground

Playground page

Result of work

User interface of the assistant

In the course of the work, a methodology for creating an intelligent chatbot in the field of educational management was developed. This chatbot will allow automating students' search queries, and will also be useful for advising students in choosing an individual learning path. This methodology can be used not only in educational management, but also in other fields.

Certificate The project won in the nomination "Improvement of distance learning systems" of the XIII International Festival of Innovative Projects "Sikorsky Challenge 2024" within the International Forum "Artificial Intelligence: Global Dialogue"

KSU.AI

You can find more details in the relevant articles:

For more information, please contact us:
KSU.AI@gmail.com