An AI-Powered Chatbot for Student Academic Advising and Course Registration Support

Background to the Study

Academic advising is a core support function in tertiary education, guiding students through course selection, prerequisite planning, credit-load management, and progression toward graduation. Effective advising has been consistently linked in the higher-education literature to improved student retention and academic performance — NACADA's clearinghouse on advising and retention points to decades of research tying advising quality and frequency of student-adviser contact to persistence and satisfaction outcomes. Yet demand for advising is highly seasonal, peaking sharply around registration periods when large numbers of students need assistance within a short window, while the supply of qualified human advisers stays comparatively fixed. This mismatch frequently produces long queues, delayed responses, and advisers spending a disproportionate share of their limited time on routine, repetitive questions rather than the individualised guidance that most benefits students with genuinely complex academic situations.

Conversational artificial intelligence, commonly known as chatbots, has emerged as a promising complement to human advising capacity. Modern intent-based chatbots use natural language processing to classify a user's message into a predefined intent category and extract relevant entities, such as a course code or semester, enabling the system to retrieve or compute a relevant response from a structured knowledge base without forcing the user into a rigid, pre-scripted phrasing. A meaningful share of that capability now comes from compact transformer models such as DistilBERT, which retains most of BERT's language understanding performance at a fraction of the size and latency, making it practical for a responsive, real-time chat interface. Systematic reviews of chatbots in academic advising indicate that, despite growing interest, the field remains fragmented, with relatively few studies offering both a rigorously evaluated natural language understanding component and a deployed, task-oriented system integrated with real institutional data. Students comparing model families for a similarly benchmarked NLP or classification project may also find ScholarNest's guide on deep learning versus classical statistical models a useful companion read.

This study contributes to this body of work by designing and implementing a complete academic advising chatbot for a Computer Science department context, spanning intent classification model development and benchmarking, dialogue management grounded in a structured course and policy knowledge base, and a deployed conversational web interface, evaluated both on intent classification accuracy and on end-to-end task completion by real test users. It follows a similar applied-AI structure to ScholarNest's project on an AI-powered academic performance prediction system, both of which apply machine learning directly to undergraduate student support.

Statement of the Problem

Existing academic advising support in the study context relies almost entirely on scheduled human adviser appointments and general departmental FAQ documents, neither of which scales well to the volume and immediacy of routine student queries during registration periods. Students with simple, well-defined questions, such as whether they meet the prerequisites for a given course, must often wait for the same limited adviser appointment slots as students with genuinely complex situations requiring individualised judgement, reducing the time available for the latter and creating frustration for the former.

While the reviewed literature confirms growing interest in academic advising chatbots, systematic reviews of the field have specifically identified a lack of rigorous empirical validation and a tendency for many existing systems to remain shallow, rule-based FAQ responders rather than genuinely grounded in structured institutional data capable of answering student-specific queries such as individual prerequisite eligibility. This study addresses that gap by combining a benchmarked intent classification component with a dialogue manager grounded directly in a structured course and policy knowledge base, and by evaluating the resulting system through both offline model metrics and an end-to-end task-completion study with real test users.

Aim and Objectives of the Study

The aim of this study is to design, implement, and evaluate an AI-powered chatbot that supports undergraduate students with academic advising and course registration queries, escalating queries beyond its competence to a human adviser.

The specific objectives of the study are to:

●        review existing literature and systems relevant to conversational AI and chatbots in academic advising contexts;

●        collect and label a dataset of representative student advising utterances across a defined set of intent categories;

●        train and compare intent classification approaches (TF-IDF/SVM, BiLSTM, fine-tuned DistilBERT) for classifying student queries;

●        design a dialogue management component that combines intent classification, entity/slot extraction, and a structured course and policy knowledge base to generate responses;

●        implement escalation logic that hands off queries outside the chatbot's trained competence to a human adviser;

●        implement and evaluate a web-based conversational interface embedding the trained model and dialogue manager; and

●        evaluate the overall system through intent classification metrics, an end-to-end task-completion study, response-time testing, and a System Usability Scale assessment.

Research Questions

●        What conversational AI techniques and system designs have been applied to academic advising chatbots in existing literature, and what are their reported strengths and limitations?

●        What intent categories and utterance patterns are most representative of real undergraduate academic advising queries in the study context?

●        How does a fine-tuned transformer-based intent classifier compare with a BiLSTM and a classical TF-IDF/SVM baseline for classifying academic advising queries?

●        To what extent can the chatbot successfully complete representative advising and registration-support tasks end-to-end, and what task types are most prone to failure or escalation?

●        What is the response time and perceived usability of the deployed chatbot interface?

Significance of the Study

This study is significant to several stakeholder groups. To academic departments, the chatbot offers a scalable first-line advising channel that can absorb a substantial share of routine queries, freeing human advisers to focus on complex, individualised cases, particularly during high-demand registration periods. To students, the chatbot offers 24-hour access to answers for common questions without requiring an appointment. To the academic community, the study contributes a benchmarked comparison of intent classification approaches on a purpose-collected academic advising utterance dataset, together with an end-to-end task-completion evaluation methodology that is comparatively underrepresented in the reviewed literature, which several systematic reviews describe as an under-validated field with limited empirical rigour. To institutional decision-makers evaluating chatbot adoption, the study offers a concrete, evaluated example of what a knowledge-base-grounded, escalation-aware advising chatbot can and cannot reliably handle.

Scope of the Study

This study is limited to text-based conversational support for a defined set of fourteen academic advising and course registration intent categories relevant to undergraduate students in a Nigerian university's Department of Computer Science, including course prerequisite inquiries, registration deadline inquiries, credit-load inquiries, GPA calculation assistance, and adviser escalation requests, among others detailed in the full project report. The study covers the complete pipeline from utterance data collection and labelling, through intent classifier training and evaluation, to dialogue management design, web interface implementation, and system-level testing including an end-to-end task-completion study. It does not extend to voice-based interaction, integration with a live, production student information system, multi-language support, or handling of intents outside the fourteen defined categories, although these are identified as directions for further work.

Operational Definition of Terms

Chatbot — A software application that conducts a conversation with a user via text, in this study specifically designed to support academic advising and course registration queries.

Intent Classification — The task of assigning a user's natural language message to one of a predefined set of categories representing the user's underlying goal.

Slot Filling / Entity Extraction — The task of identifying and extracting specific pieces of information (such as a course code or semester) from a user's message.

Dialogue Manager — The system component responsible for deciding how to respond to a classified intent, typically by querying a knowledge base or invoking business logic.

Escalation — The process by which the chatbot hands off a query it cannot confidently or appropriately answer to a human academic adviser.

Knowledge Base — A structured repository of course, prerequisite, and policy information queried by the dialogue manager to generate responses.

System Usability Scale (SUS) — A standardised ten-item questionnaire used to measure the perceived usability of a system, yielding a score out of 100. Published benchmarking research puts the average SUS score across hundreds of studies at around 68, which is the reference point this study's 78.4 'good' rating is measured against.

Conclusion

A chatbot will not replace human academic advisers, and this study does not try to make it one. What it shows instead is that a well-benchmarked intent classifier, paired with a dialogue manager grounded in real institutional data and a clear escalation path, can absorb a meaningful share of the routine questions that otherwise clog advising capacity during registration periods. For students building a related final-year project, the combination of rigorous model benchmarking, a genuinely structured knowledge base, and an end-to-end usability evaluation with real test users is exactly the kind of structure that turns a good idea into a complete, defensible piece of research. If you want hands-on support shaping a topic like this into a polished, submission-ready project, ScholarNest's research and project writing support team can help at any stage, from proposal to final defence.

Frequently Asked Questions

1. What is an AI-powered chatbot for academic advising?

It is a conversational AI system that uses natural language processing to answer routine student questions about course prerequisites, registration deadlines, credit loads, and graduation requirements, escalating anything beyond its competence to a human adviser.

2. How does the chatbot understand what a student is asking?

It uses an intent classification model, in this study a fine-tuned DistilBERT transformer, to map a student's message to one of a predefined set of categories such as prerequisite inquiry or registration deadline inquiry, then extracts relevant details like a course code through slot filling.

3. Why did DistilBERT outperform the BiLSTM and SVM baseline?

DistilBERT is a compact, pretrained transformer that retains most of BERT's contextual language understanding while being fast enough for real-time chat, which gives it an advantage in capturing nuanced phrasing over a classical TF-IDF/SVM model or a BiLSTM trained from scratch.

4. What happens when the chatbot cannot answer a question?

The system is designed to escalate queries outside its fourteen trained intent categories to a human academic adviser, rather than attempting to guess or provide an unreliable answer.

5. What is a knowledge base in this context?

It is a structured repository of course, prerequisite, and registration policy information that the chatbot's dialogue manager queries to generate accurate, institution-specific responses rather than generic answers.

6. How is the chatbot's usability measured?

The study used the System Usability Scale, a standardised ten-item questionnaire, which returned a mean score of 78.4, corresponding to a 'good' usability rating relative to the widely cited industry average of around 68.

7. What is a task-completion rate and why does it matter?

It measures the percentage of representative advising tasks users could complete successfully through the chatbot; this study recorded an 86.7% completion rate, with most failures traced to queries outside the trained intent set that were correctly escalated.

8. Is this a good final-year project topic for computer science students?

Yes. It combines dataset collection and labelling, multi-model intent classification benchmarking, dialogue management design, and a full end-to-end usability evaluation, giving students a well-rounded project spanning NLP research and applied software development.

9. What tools are typically used to build an academic advising chatbot?

A common approach is to fine-tune a compact transformer model such as DistilBERT for intent classification, pair it with a rule-based dialogue manager and knowledge base, and serve the system through a web framework like Flask.

10. How fast can a deployed advising chatbot respond to student queries?

With an efficient pipeline, average response time can be well under a second per conversational turn, which is fast enough to feel natural in a real-time chat interface.