Project typeContent analytics case study
Featured case, content analytics

Data-Driven Content Analytics for LVMH

My team analyzed 23,805 Red Note posts to help LVMH's beauty teams refine their content strategy on the platform. We looked at which posts people liked and which they saved, then used those patterns to develop 6 content briefs across fragrance, makeup, and skincare.

SkillsMultimodal analysis, content strategy
TimelineNov 2025 – Apr 2026
Methods & toolsNLP, ViT, Random Forest, LLM features

What makes a beauty post worth saving?

On Red Note (Xiaohongshu), someone might like a fragrance video for the mood and save a different one to compare scents later. LVMH's beauty teams needed to decide what to film and which response to measure.

Likes

We used likes to measure immediate audience response.

Saves

We used saves as a signal that people wanted to revisit a post, potentially while considering a purchase.

My contribution

I co-developed the pipeline that connected post text, visuals, and engagement. Our team turned those patterns into six briefs, one aimed at likes and one aimed at saves for each category.

How we compared posts

  1. 01

    Segment the dataset

    We started with 10,854 image posts and 12,951 video posts. We compared formats within fragrance, makeup, and skincare, excluding posts labeled “Other” from those comparisons.

  2. 02

    Extract content signals

    We recorded posting time, titles, captions, hashtags, and mentions. For video, we also used transcripts, visual features, and AI-generated content labels.

  3. 03

    Compare engagement patterns

    We analyzed keywords and topics, grouped similar visuals, and used Random Forest feature rankings to see which traits were most informative for likes and saves.

  4. 04

    Write the briefs

    We looked at which content choices appeared together, then turned those combinations into suggestions for what to film, how to present the product, and what to say.

Original work

Where video had the edge

What we would make

For the same fragrance, a lifestyle video could invite a quick reaction, while a review could give someone a reason to save the post. The briefs apply that distinction across all three categories.

Fragrance

For likes

Film the scent outdoors in natural light and frame it around an occasion, like a summer evening.

For saves

Put the bottle and packaging in close-up, then have a reviewer describe the scent notes and compare the fragrance with a similar one.

Makeup

For likes

Show an everyday eye look in facial close-up so viewers can see the application and the finished look.

For saves

Film a step-by-step concealer tutorial with close-ups, a before-and-after, and tips for adapting the technique.

Skincare

For likes

Open with a concern such as visible pores and show the product texture, how to apply it, and the relevant ingredients.

For saves

Have a creator show where the product fits in a routine and explain what they layer it with and in what order.

Would more saves lead to sales?

The briefs give LVMH a starting point for a test. We observed how existing posts performed, and the available outcomes were engagement metrics. We had no purchase data.

What could explain the gap

Posting time, trends, and how the platform distributed each post could all affect likes and saves. A comparison of existing posts leaves those influences in the results.

How I'd test it

Start with the two fragrance videos. Use the same product and comparable audiences and posting windows. Track likes and saves per view, then use tagged product links to measure clicks and attributed purchases.