evidoria

← Back to browse

Good practice Imported

NEU-Chatbot — National Economics University's AI Admissions Assistant on Facebook, Hanoi

Vietnam · Hanoi · See the Vietnam profile · See the Hanoi profile

Evidence: Observational / pre–post Top 89% 27/100 · Ask Evidence Copilot about this practice

National Economics University in Hanoi built an AI admissions chatbot (Rasa deep-learning intent models, 97.1% test accuracy) for its official Facebook page. Over three months it fielded more than 50,000 student and parent questions and answered 90.3% of them correctly.

97.1%
Intent-detection accuracy on held-out test set
50,000+
Questions received via Facebook in three months (2020-2021 deployment)
90.29%
Questions answered appropriately without staff intervention (2020-2021 deployment)
50+
Distinct admissions intents modelled
~1,500
Labelled training examples
NEU-Chatbot — National Economics University's AI Admissions Assistant on Facebook, Hanoi

Details

Promoter
National Economics University (NEU)
Period
2020–2021
Keywords
higher education, admissions, student services, chatbots

Context

National Economics University (NEU), one of Vietnam's leading business universities, built an AI-driven chatbot to handle the surge of admissions questions it receives every year on Facebook — the most widely used social network in Vietnam. Researchers Nguyen, Le, Hoang and Nguyen describe the system in a 2021 peer-reviewed paper in Computers and Education: Artificial Intelligence (Elsevier).

Activities

The chatbot uses deep-learning intent-classification models built on the Rasa open-source framework, with a custom Vietnamese-language preprocessing pipeline. It was trained and tested on a curated dataset of over 50 distinct admissions-related intents and roughly 1,500 labelled examples covering curriculum details, admission procedures, tuition fees and IELTS score requirements.

Results

The model reached 97.1% intent-detection accuracy on the held-out test set. After NEU put the chatbot live on its official admissions Facebook Fanpage, it received more than 50,000 questions from prospective students and parents in three months and answered 90.29% of them appropriately, without a human staff member intervening.

Conclusions

The published evaluation stops at answer accuracy — it does not report downstream outcomes such as application completion rates, enrolment yield or applicant satisfaction, and it does not describe any data-protection or bias review of the system. Those are reasonable next steps for anyone replicating this in another institution or language.

Implementation
Indicative cost
Low (< €50k)
Time to results
Short (< 1 year)
Staffing & skills
Built and evaluated by NEU-affiliated researchers (Nguyen, Le, Hoang, Nguyen) using the open-source Rasa framework

Conditions for success

  • Custom Vietnamese-language preprocessing pipeline
  • Curated training dataset of 50+ intents and ~1,500 labelled examples covering the most common admissions questions
  • Deployment on Facebook, Vietnam's most widely used social network, where applicants already congregate

Common failure modes

  • Published evaluation stops at answer accuracy — no downstream outcome data (application completion, enrolment yield, applicant satisfaction) is reported
  • No data-protection or bias review of the deployed system is described

Commonly funded by

Own resources / municipal budget

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Replication kit

Reusable artefacts from this practice — as published by their sources.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful