A mvp (web app, through flask/dash/streamlit etc) that takes input (upload a skin image) and generate an output (recommend helpful products that works for their skin type after classifying the skin type).
Data source: amazon and Sephora reviews of different skin care products – users often upload photos with problems described. Dermnet website also contains pictures with skin problems such as facial acnes, eczema
To do: 1. Scrape those websites (e.g., Amazon, sephora) to get skin images with product reviews, as well as websites with labeled data (e.g., dermnet). Parse data, do image analysis and nlp and create labels (skin type) from the data, such that classification of those different skin types e.g., acne prone; redness; oily) can be done when input image is passed in. 2. Case-based retrieval from scraped and parsed data to recommend skin care products that work for the user input based on the stage 1 classification.
- only an mvp to have something up and working so do not need extensive data and labels, or elegant models – maybe 3 skin types for now and more data can be added for the next assignment, and models can be refined.
Objective: An application to classify the type of skins that users have based on input images. Then recommend suitable products based on skin type. · Build a mvp (web app, through flask/dash/streamlit etc) that takes input (upload a skin image) and generate an output (recommend helpful products that works for their skin type after classifying the skin type). · Data source: 1. Amazon reviews of different skin care products (Pick at least 5 products for 3 different skin types e.g., acne-prone; redness; aging, etc.). Users often upload photos with problems described. Create labels from analyzing images with user reviews. E.g., redness on face are gone after using this = redness-prone skin · Example links: · e.g., https://www.amazon.com/Eucerin-Sensitive-Redness-Relief-Soothing/product-reviews/B0013L3XMM/ref=cm_cr_dp_d_show_all_btm?ie=UTF8&reviewerType=all_reviews · e.g., https://www.amazon.com/CeraVe-Foaming-Facial-Cleanser-Washing/product-reviews/B01N1LL62W/ref=cm_cr_arp_d_paging_btm_next_2?ie=UTF8&reviewerType=all_reviews&pageNumber=2 2. Sephora website. Products are categorized with skin concerns and what skin type is the product good for. Customer reviews also listed their skin type with feedback. Example link: https://www.sephora.com/product/the-ordinary-deciem-hyaluronic-acid-2-b5-P427419?icid2=products%20grid:p427419 3. Dermnet websites contains pictures with skin problems such as facial acnes, eczema http://www.dermnet.com/dermatology-pictures-skin-disease-pictures/ https://dermnetnz.org/image-library/ To do: 1. Scrape those websites (e.g., Amazon, sephora) to get skin images with product reviews, as well as websites with labeled data (e.g., dermnet). Parse data, perform image analysis, NLP and create labels (skin type) from the data, such that classification of different skin types e.g., acne prone; redness; oily) can be done when input image is passed in. 2. Case-based retrieval from scraped and parsed data to recommend skin care products that work for the user input based on the stage 1 classification. *We’re only building an mvp this week, which is to have something up and working so we do not need extensive data and labels or elegant models. We can add more data and refine models for next week’s assignment.