Two research papers: one paper accepted for publication in Springer Nature conference proceedings (FICTA 2026) on spotting rare emotions in student chatbot conversations, and one systematic literature review on generative AI in marketing presented at the 10th Pan-IIM World Management Conference.
Multi-Class Affect Detection in Educational Chatbots: Evidence from Imbalanced Dialogue Data
Accepted for publication in Springer Nature proceedings, Smart Innovation, Systems and Technologies series (Scopus indexed) · FICTA 2026
Context and Problem
Educational dialogues contain severe class imbalance. Critical emotional states like confusion and frustration indicate when a learner is struggling, but they appear far less frequently than neutral or positive acknowledgments. Conventional classification algorithms optimize for overall accuracy, which allows them to score well while failing to detect the minority emotional states where pedagogical intervention is most needed.
Methodology
I paired RoBERTa, a pretrained language model, with SVMSMOTE, a method that creates extra examples of the rare emotion classes so the model sees enough of them. I judged the models by how many of the rare Confused cases they caught (recall), not by overall accuracy.
Key Findings
The proposed RoBERTa plus SVMSMOTE architecture achieved a 74.19% recall on the rare Confused emotion class, compared to less than 52% recall for baseline architectures without synthetic oversampling.
Venue
Springer Nature (FICTA 2026)
Primary Metric
74.19% Rare-Class Recall
Architecture
RoBERTa + SVMSMOTE
Adoption and Implications of Generative AI in Marketing: A Systematic Review and Research Agenda
Presented at the 10th Pan-IIM World Management Conference 2025 · IIM Ranchi
Context and Objective
Enterprise interest in generative AI for marketing automation has grown rapidly, but academic literature remained fragmented across practitioner whitepapers and narrow technical benchmarks. The review pulls together what published studies say about how companies use generative AI in marketing.
Review Protocol
Following the PRISMA review method, I reviewed 29 Scopus articles published from 2020 to 2025 across technology readiness, consumer adoption, and regulatory considerations.
Synthesis and Proposed Agenda
The review synthesised research priorities across algorithmic transparency in personalized marketing, consumer trust in synthetic media, workflow hand-offs between marketers and generative models, and data governance for automated campaigns.
Venue
10th Pan-IIM World Management Conference (WMC 2025)