HKMU FYP · COMP 4570SEF

Cantonese-English code-mixed marketing made smart

SlashMarketer is the research prototype for Cantonese-English code-mixed marketing NLG — LoRA fine-tuned Qwen3-8B-Base over the Hong Kong Marketing Corpus, with an FFT + CWT + DWT + Kalman signal-attribution module.

Generate copy Attribution pipeline

Code-mixed marketing NLG

Generate Cantonese-English code-mixed ad copy via SlashAI, driven by a standardized marketing brief.

Try generate

Signal attribution

FFT + CWT + DWT + Kalman composite transient detection on daily marketing series.

Run attribution

HKMC ad library

1,437-record curated Cantonese-English marketing corpus across 19 sectors, licence-labelled.

Learn more

About the research

Two threads: a primary NLG thread (LoRA on Qwen3-8B-Base over HKMC, RAG with BGE-M3, evaluated by BLEU-4 / ROUGE-L / CSI / Burstiness + a Likert rubric) and a secondary signal-processing thread (transient detection validated on synthetic ground truth).

HKMC v2

669 → 1,437 records · 19 sectors · CC BY / CC BY-SA