AI-Powered Personalized Learning Platform: NLP-Driven Article-Centric Chatbot with Sentiment Analysis
Dr. Pavithra N., Dr. Sapna R., Dr. Preethi, Dr. Manasa C. M., Dr. Ashwitha A., Vishesh Goyal
Overview
Most AI learning tools answer from a giant pre-trained knowledge base, which means they hallucinate, drift off-topic, and can't be controlled. This IEEE-published system takes a different approach: upload any article, and the chatbot answers only from that content. Built with NLP, TF-IDF vectorisation, cosine similarity, and an SVM-based sentiment classifier, the platform achieved 90% question-answering accuracy and 90.84% precision, with zero reliance on large language models. Designed specifically for educational environments where transparency and controlled knowledge sources matter.
About This Research
The core problem with most AI-powered learning tools is that they don't know what they don't know. General-purpose chatbots hallucinate answers, pull from stale or irrelevant knowledge, and give students no way to verify where the information came from. In educational settings — where accuracy and source transparency are non-negotiable — this is a serious problem. This system solves it with a strict constraint: the chatbot can only answer from the article the user uploads. Nothing else. No pre-trained knowledge base, no external retrieval. If it's not in the document, it doesn't answer.
Technical Architecture
The pipeline is deliberately lightweight and interpretable — a conscious choice over heavier transformer-based models.
| Stage | Technique |
|---|---|
| Text Preprocessing | Tokenisation, lowercase normalisation, stopword removal, stemming / lemmatisation |
| Vectorisation | TF-IDF on both query and article content |
| Relevance Scoring | Cosine similarity to find the most relevant passage |
| Sentiment Analysis | SVM classifier running in parallel on content tone |
Why Not Transformers
Heavier transformer-based models introduce hallucination risk, opacity, and latency — all unacceptable for classroom use. This architecture is fully auditable: every response traces directly back to a passage in the uploaded document.
Multimodal Output
The chatbot surfaces relevant image links alongside text responses. This improves student engagement without adding meaningful computational overhead.
Why This Matters
| Property | General-Purpose Chatbot | This System |
|---|---|---|
| Hallucination Risk | High | None — bounded by source |
| Source Transparency | None | Full — every answer is traceable |
| Knowledge Freshness | Stale training data | Always current — user uploads |
| Classroom Trust | Low | High |
For educational applications, a lightweight, interpretable, document-bounded system consistently outperforms over-engineered solutions. Students and teachers need to trust the source. This system makes that possible.
Published at the 2026 International Conference on Next-Gen Quantum and Advanced Computing (NQComp), IEEE.