Makeup Match Demo

Insight Data Science Fellowship

Transcript

1. Chanin Woods
2. Friend recommended
3. ...but contains silicone Friend recommended
4. ...but contains silicone ...the top recommendations on Sephora.com all contain silicones too Friend recommended
5. Gather Data BeautifulSoup, MySQL
6. Reviews (300,000) Product Ingredients (14,000) Products (500) Gather Data BeautifulSoup, MySQL
7. Ingredients (25,000) Products (500) Reviews (300,000) Product Ingredients (14,000) Gather Data BeautifulSoup, MySQL
8. Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
9. Pre-process The lipstick colors are an amazing color lipstick | colors | amazing | color lipstick | amazing | color Stemming Stop Words Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
10. Pre-process The lipstick colors are an amazing color lipstick | colors | amazing | color lipstick | amazing | color Stemming Stop Words Term Freq. – Inverse Doc. Freq. High word count Rare Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
11. The lipstick colors are an amazing color lipstick | colors | amazing | color lipstick | amazing | color Stemming Stop Words Rare Latent Semantic Indexing 51,000 Words 100 Words to High word count Term Freq. – Inverse Doc. Freq. Pre-process Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
12. Pre-process The lipstick colors are an amazing color lipstick | colors | amazing | color lipstick | amazing | color Stemming Stop Words Rare Latent Semantic Indexing 51,000 Words 100 Words to Cosine Similarity Item 2 Item 3 High word count Term Freq. – Inverse Doc. Freq. Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
13. Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
14. Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
15. Oh No! Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
16. Reviewer Model Collaborative Filtering Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
17. Sentiment Analysis Reviewer Model Collaborative Filtering Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
18. Sentiment Analysis 1 0 1 1 0 1 1 1 0 Item 1 Item 2 Item 3 User-Item Matrix Reviewer Model Collaborative Filtering Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
19. Sentiment Analysis User-Item Matrix 1 0 1 1 0 1 1 1 0 Item 1 Item 2 Item 3 Cosine Similarity Item 2 Item 3 Reviewer Model Collaborative Filtering Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
20. Linear Combination d = a*Reviews + b*Reviewers Ensemble Distances (d) Reviewer Model Collaborative Filtering Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
21. Ensemble Distances (d) Reviewer Model Collaborative Filtering Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
22. Ensemble Distances (d) Reviewer Model Collaborative Filtering Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
23. Real-Time Filter Ingredients SQL queries Ensemble Distances (d) Reviewer Model Collaborative Filtering Gather Data BeautifulSoup, MySQL Reviews Model NLTK, gensim
24. What is happening? Dimension 1 Dimension 2 Lipstick Mascara Conditioner Blush Foundation
25. What is happening? Dimension 1 Dimension 2 Lipstick Mascara Conditioner Blush Foundation
26. Discovered New Category!
27. Discovered New Category!
28. Discovered New Category!
29. Chanin Woods PhD in Bioinformatics
35. Similarity Same Category Different Category Similarity Liked Both Items Disliked Either Item Reviews Model Validation
36. Reviewers Model Validation Similarity Liked Both Items Disliked Either Item Similarity Same Category Different Category
37. Ensemble Model Similarity Same Category Different Category Similarity Liked Both Items Disliked Either Item
38. Reviews Model Alternatives Dimension 2 Dimension 1 TF-IDF and LSI Dimension 1 TF-IDF Only Dimension 1 LDA
39. Reviewers Model Alternatives Sentiment Frequency
40. Dimension 1 Dimension 2 Lipstick Mascara Conditioner Blush Foundation Products (red = single brand)
41. Term Frequency-Inverse Document Frequency FiloTechnologia.blogspot.com
42. Latent Semantic Indexing iv.slis.indiana.edu Singular Value Decomposition (SVD) Reduced Dimensions
43. Item-Based Collaborative Filtering salemmarafi.com
44. Textblob Source WordNet Pattern TextBlob