DEVELOPMENT OF AN INTELLIGENT TEACHING ASSISTANT BASED ON THE KNOWLEDGE GRAPH

Zozulya N.M., Kuksina O.I.
e-mail: nikitadonputilov@gmail.com

Abstract. The development of intelligent teaching assistants has gained attention due to their potential to enhance learning and improve educator efficiency. This paper explores the design of an intelligent teaching assistant based on a knowledge graph, which represents relationships between concepts. The assistant can answer questions, explain material and support teachers in managing student queries. An example of a knowledge graph for a computer science course has been provided. It has also been demonstrated how the assistant answers queries and discussed the integration of natural language processing (NLP) for better interaction and accuracy.

Keywords: intelligent teaching assistant, knowledge graph, natural language processing, educational technology, question answering, teacher support, course content, concept relationships, NLP techniques, student queries, graph-based learning, computer science education.

Introduction

The concept of intelligent teaching assistants has emerged as a response to the increasing need for personalized and scalable educational support. Traditional classroom settings often face challenges in addressing individual student needs, especially in large classes where teachers are unable to provide one-on-one assistance to each student. With advancements in artificial intelligence (AI) and natural language processing (NLP), it is now possible to create digital assistants that can assist teachers and students in answering questions, explaining content, and guiding students through the learning process.

One of the most effective ways to model and organize educational content is through the use of a knowledge graph [1]. A knowledge graph is a network of concepts or entities linked together by relationships, making it possible to visualize and access information more efficiently. Knowledge graphs have been used in various fields, including search engines, healthcare, and e-commerce, and their application in education is a promising area of research.

This paper presents the development of an intelligent teaching assistant based on a knowledge graph. We explore the design and architecture of such an assistant, the creation of a knowledge graph, and the query-answering capabilities it offers. The goal is to demonstrate how knowledge graphs can be used to enhance the interaction between students and teachers, streamline the process of answering questions, and provide accurate explanations of theoretical material.

Knowledge Graph for Educational Content

A knowledge graph is a data structure used to represent knowledge through entities (or nodes) and relationships (or edges). It is particularly powerful in representing interconnected information, making it an ideal solution for organizing educational content. In the context of an educational course, a knowledge graph can be used to visually and structurally map the relationships between different concepts, terms, and topics. This structure allows for easy navigation, query answering, and comprehension of how different topics are interrelated. The connections between concepts can be classified using different types of relationships such as "is a type of," "depends on," "includes," "is related to," "is a part of," etc.

In educational settings, especially in fields like computer science, mathematics, or physics, topics often build upon one another. This hierarchical and interdependent nature makes knowledge graphs highly valuable for both students and educators. The graph not only helps visualize the content but also provides a basis for answering questions, elaborating on concepts, and even discovering new information that wasn't previously explored.

Structure of a Knowledge Graph

A knowledge graph can be divided into two core components:

Nodes (Entities): these represent the key concepts or terms within the course or domain. In the case of a computer science course, for example, nodes might represent terms like "Algorithms," "Sorting," "Data Structures," "QuickSort," "Complexity Analysis," "Recursion," "Graphs," and "Trees".
Edges (Relationships): these represent how the concepts or entities are related to one another. In an educational knowledge graph, relationships can be varied and nuanced. For instance, "QuickSort" might be connected to "Sorting" through the relationship "is a type of," while "Sorting" itself might be linked to "Algorithms" through the relationship "is a subset of".

These edges help to form the network of connections between various concepts [2]. Such networks are particularly valuable because they allow a teaching assistant based on this structure to answer questions that require an understanding of multiple concepts in relation to one another.

Example of a Knowledge Graph for Computer Science

To illustrate how knowledge graphs are applied in the context of education, let’s consider a portion of a knowledge graph for a computer science course. In this graph, we’ll model several key concepts from the subject of algorithms (Figure 1).

This knowledge graph can be broken down as follows:

"Algorithms" is the core concept and is connected to "Sorting" through the relationship "is related to". This indicates that sorting algorithms are a type of algorithm.
"Sorting" is a subfield of algorithms that includes various algorithms for sorting data. In this case, "QuickSort" is included as one specific sorting algorithm.
"Algorithms" "depends on" "Data Structures", highlighting that algorithms are often designed to operate on different types of data structures.
"Data Structures" is a broader category that includes both "Array" and "Linked List". These are two commonly used data structures, and they are connected to "Data Structures" through the relationship "is a type of".
The knowledge graph also illustrates Big O Notation, which is a concept linked to "Complexity Analysis", and provides a complexity measure (e.g., O(n log n)) for both QuickSort and Sorting.

Relationships in a Knowledge Graph

The relationships in a knowledge graph can serve various purposes, such as:

Hierarchical Relationships: these can indicate whether one concept is a more general term for another or whether one concept is a specialized version of another. For example, in the graph above, "Sorting" can be viewed as a more specialized area within the broader concept of "Algorithms".
Dependency Relationships: these illustrate which concepts or techniques depend on others for understanding. For instance, the fact that "Algorithms" "depends on" "Data Structures" means that students need a solid understanding of data structures to effectively understand and implement algorithms.
Containment Relationships: these indicate that a concept contains or includes sub-concepts. For example, "Sorting" includes algorithms like "QuickSort", which are specific implementations or instances of sorting algorithms.
Complexity Relationships: these relationships represent the performance characteristics of concepts, such as computational complexity. The connections from "Sorting" and "QuickSort" to "O(n log n)" show that these sorting algorithms, on average, have this time complexity.

Natural Language Processing (NLP)

When a student submits a query, the first task of the assistant is to process the question and extract the relevant information. This process is done through Natural Language Processing (NLP), a subfield of artificial intelligence focused on enabling machines to understand, interpret, and generate human language [3]. NLP techniques are critical for transforming a raw text into structured data that the system can work with.

Key Tasks in NLP:

Tokenization: the first step in NLP is tokenizing the text, which involves breaking down the sentence into smaller components (tokens) such as words or phrases. For example, the sentence "What is QuickSort?" would be tokenized into: ["What", "is", "QuickSort"].
Named Entity Recognition (NER): in this step, the system identifies key entities or concepts within the sentence. For the query "What is QuickSort?", the system recognizes "QuickSort" as a named entity, which refers to a concept in the knowledge graph.
Intent Recognition: understanding the intent behind the query is a crucial task in NLP. In this example, the assistant must determine that the student is asking for a definition or explanation of "QuickSort". Techniques such as text classification or semantic role labeling are often used here. The intent can be classified into categories like definition request, relationship request, or example request.
Part-of-Speech (POS) Tagging: this step identifies the grammatical structure of the sentence, labeling each word as a noun, verb, adjective, etc. For example, in "What is QuickSort?", "What" would be a question word, "is" a verb, and "QuickSort" a noun.

Graph Traversal

With the key concepts and intents extracted from the student's question, the assistant needs to navigate the knowledge graph to find relevant information. The knowledge graph stores relationships between concepts, and the assistant utilizes these relationships to retrieve accurate answers.

Steps in Graph Traversal:

Identifying Relevant Nodes: the assistant searches for the nodes in the graph that represent the concepts mentioned in the query. In our example, the concept "QuickSort" is identified as a node in the graph.
Exploring Relationships: once the relevant node is located, the assistant explores the relationships that connect it to other nodes. These relationships can be labeled as "includes", "depends on", "has", or other context-specific labels. For instance, "QuickSort" might be connected to the "Sorting" node, and "Sorting" may be connected to "Algorithms" through relationships such as "is a type of".
Breadth-First Search (BFS) / Depth-First Search (DFS): the assistant can apply graph traversal algorithms like BFS or DFS to explore the graph and gather all the relevant connections. These algorithms systematically explore all nodes linked to the queried concept. If the student is looking for a broader explanation, such as how "QuickSort" fits into "Sorting" or its relation to "Complexity Analysis", the assistant may traverse the graph to provide such connections.
Filtering Irrelevant Nodes: not all the nodes in the graph will be relevant to the query. The assistant uses context from the question (e.g., the intent and entities) to filter out irrelevant information. For instance, if the user asked specifically about "QuickSort", the assistant wouldn't need to delve into unrelated concepts like "Data Structures" unless they are explicitly mentioned or relevant.

Example Interaction

Consider the following example of an interaction with the intelligent teaching assistant.

Student: "What is a linked list?"

Assistant: "A linked list is a data structure consisting of a collection of nodes, where each node contains data and a reference (or link) to the next node in the sequence. Linked lists are used to implement other data structures such as stacks and queues."

In this interaction the assistant recognizes "linked list" as a concept and retrieves its definition from the knowledge graph. The assistant responds by explaining the concept and providing a brief description of its usage.

This process involves querying the knowledge graph to retrieve relevant information and using NLP to generate a coherent, context-appropriate response.

Conclusion

The development of an intelligent teaching assistant based on a knowledge graph represents a promising approach to enhancing the educational experience for both students and educators. By leveraging the power of knowledge graphs, it is possible to organize and access educational content more effectively, allowing the assistant to answer student queries and provide explanations in a personalized manner. The integration of natural language processing techniques further enhances the ability of the assistant to understand and respond to questions in a human-like manner.

This approach can significantly improve the scalability of educational support, especially in large classes where direct interaction with the teacher may be limited. Furthermore, the assistant can be continuously updated and expanded as new concepts are introduced, ensuring that it remains a valuable tool throughout the duration of the course.

Future work can focus on expanding the knowledge graph to include more advanced topics, integrating machine learning techniques for personalized learning, and enhancing the assistant’s ability to generate more complex responses. With continued advancements in AI and NLP, intelligent teaching assistants have the potential to revolutionize the way we approach education and support teaching and learning.

References

Volkova I.A., Shamaeva E.D. Excursion into knowledge graphs / I.A. Volkova, E.D. Shamaeva // International Journal Of Open Information Technologies. – 2023 – №1 – P. 75-83.
Gasparian M., Korneev D., Titov V., Karmanov M., Golkina G., Sekerin V., Gorokhova A. Conceptual model of a smart integrated educational environment / Mikhail Gasparian, Dmitriy Korneev, Valery Titov, Mikhail Karmanov, Galina Golkina, Vladimir Sekerin, Anna Gorokhova // Wisdom – Special Issue 3 (4) – 2022 – P. 32-39.
Geldieva G., Durdyyeva O., Atayeva O. Everything we need to know about the artificial intelligence (AI) / Galina Geldieva, Olga Durdyyeva, Oksana Atayeva // International Scientific Journal "Innovative Science" – № 5-1 – 2024 – P. 166-167.