Skip to content

Resource list for improving search results by incorporating images, additional language, audio

Notifications You must be signed in to change notification settings

stoneyv/multi-modal-search-resources

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 

Repository files navigation

Multi Modal Search

Resources for incorporating image embeddings, joint image/word embeddings, multi lingual word embeddings for learning to rank or as a replacement for things like BM25 or TFIDF.

  1. Kamelia Aryafar (Overstock, formerly Etsy) Learning to Rank in Ecommerce AIWTB 2017 https://arxiv.org/abs/1511.06746 (Nov 2015) https://www.youtube.com/watch?v=QjTi1qcLTQw
  • In production retrieve top k documents using SOLR and bm25. Then rerank the the returned results. Concatenation of transfer learned image embeddings from a VGG model trained on ImageNet with bag of words Learning rank from search logs using click and ignore signals

  • Great examples of overcoming user labels for Wedding dresses that are not relevant.

  1. Ethan Rosenthal (Dia & Co, Birchbox)
    Ethan writes very accessible content and prefers methods that are intuitive.
  1. Ivona Tautkute
  1. Han Xiao (Zalando)
    Building Cross-Lingual End-to-End Product Search with Tensorflow
    https://hanxiao.github.io/2018/01/10/Build-Cross-Lingual-End-to-End-Product-Search-using-Tensorflow/

  2. Babylonpartners / fastText_multilingual
    alignment of multiple fastText word embeddings into a single vector space
    https://github.com/Babylonpartners/fastText_multilingual

  3. Aleksander Movchan

  1. Shutterstock
    Great examples of overcoming disambiguation due to stemming.
    hawk -> some 'Stephen Hawking' along with 'hawk' that is a bird.
    angel -> 'Los Angeles' more than angel
    https://tech.shutterstock.com/2017/03/08/image-search-using-joint-embeddings-part-one/
    https://tech.shutterstock.com/2017/04/13/image-search-using-joint-embeddings-part-two/

  2. Lyse clothing search blog post
    https://making.lyst.com/2018/01/10/a-machine-learning-model-to-understand-fashion-search-queries/

  3. Deep Fashion
    http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion.html

  4. Dan Gillick (Google NLP group, instructor at Berkeley)
    Student talk at Berkeley

  • learning embeddings in the same vector space for more than one media type multi lingual search
  • Increasing relevancy by returning more results from tail of documents.
  • They mostly use Google search logs
    https://www.youtube.com/watch?v=JGHVJXP9NHw
  1. Google AutoML announcment
    Use cases that are mostly about learning product attributes https://www.blog.google/topics/google-cloud/cloud-automl-making-ai-accessible-every-business/

About

Resource list for improving search results by incorporating images, additional language, audio

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published