Code of Conduct
Results 1 to 19 of 19
  1. #1

    Default Netflix of Scents

    One of the common themes that I see people post about here is "would I like X if I like Y?"
    Netflix has a history of recommending movies you may like based on what you have liked.
    Amazon does a similar thing by recommending what other customers with interests similar to yours have purchased.
    Pandora does a better job at music recommendation by incorporating more meaningful features (genre, instrumentation, mood, etc.) in their analysis.

    I have searched around and found a couple of scent recommendation services, but they were neither comprehensive (usually have a limited database of mainstream fragrances) nor did even a remotely decent job at recommending relevant fragrances. which brings me to these points:

    1- If you know of a decent scent recommendation service please enlighten me.

    2- Assuming none already exists, how well will people receive such a service? contrary to movie and music, perfumes seem to entertain a rather small community.

    3- Are there any Basenoters who would be willing to collaborate with me? My knowledge of perfume world is very limited. But I'm fairly well versed with machine learning algorithms that can be applied to this task. The crucial thing is to arrive at a comprehensive set of features that the program will use to determine similarity of different scents. Some candidate features I can think of are perfume house, perfumer nose, notes (base, heart, top), and molecular structure.

    Send me a PM, or a leave comment below.
    Thanks :)

  2. #2
    Sound Scents
    drseid's Avatar
    Join Date
    Jan 1970
    Location
    Northern VA/DC Area
    Posts
    6,690

    Default Re: Netflix of Scents

    I just filled out a form from Lucky Scent where they ask you a bunch of questions about your tastes and then they send you back some recommendations accordingly. It is a free manual service they offer from a link on their site. I have yet to get my response back, so we will see what they come up with, but I will say the questions they asked were quite probing and detailed. It should be interesting...
    Current Top Favorites:
    1) Portrait of a Lady original formula (EdP Frédéric Malle)
    2) Giorgio for Men vintage/V.I.P. for Men (Giorgio Beverly Hills)
    3) Dia Man vintage edt (Amouage)
    4)
    Anat Fritz Original Formula and Classical (Anat Fritz) - tie
    4) Lalfeorosa (O'driù) - tie

    6)
    Les Nombres d'Or Vetyver (Mona di Orio)
    7) Captain vintage (Molyneux)
    8) Tzora (Anat Fritz)
    9) Javanese Patchouli (Zegna) - tie
    9) Monsieur de Givenchy vintage (Givenchy) - tie
    9) Coeur de Vetiver Sacré (L'Artisan) - tie
    9) X for Men (Clive Christian) - tie
    9) Patou pour Homme Privé (Jean Patou) - tie
    9) Oud Shamash (The Different Company) - tie

  3. #3

    Default Re: Netflix of Scents

    Quote Originally Posted by drseid View Post
    I just filled out a form from Lucky Scent where they ask you a bunch of questions about your tastes and then they send you back some recommendations accordingly. It is a free manual service they offer from a link on their site. I have yet to get my response back, so we will see what they come up with, but I will say the questions they asked were quite probing and detailed. It should be interesting...
    Interesting. I'd really like to know your assessment of their recommendations.

  4. #4

    Default Re: Netflix of Scents

    Fragrantica has such feature.

  5. #5

    Default Re: Netflix of Scents

    Quote Originally Posted by zliang View Post
    Fragrantica has such feature.
    If I'm not mistaken Fragrantica's approach is similar to Aamazon's and Last.fm's: They recommend what others with similar interests have liked. That is the least intelligent way of doing this.

  6. #6

  7. #7
    hednic's Avatar
    Join Date
    Oct 2007
    Location
    McLean, NYC, & Búzios
    Posts
    83,755

    Default Re: Netflix of Scents

    Quote Originally Posted by zliang View Post
    Fragrantica has such feature.
    Not a bad one at that.

  8. #8

    Default Re: Netflix of Scents

    I tried a couple and the results were not impressive. For Encre Noire it suggested La Nuit de l'Homme, and for Noir Epices, Incense Oud.

  9. #9

    Default Re: Netflix of Scents

    Quote Originally Posted by Reza View Post
    I tried a couple and the results were not impressive. For Encre Noire it suggested La Nuit de l'Homme, and for Noir Epices, Incense Oud.
    What makes you say that? Do you understand how these things work? Are you expecting 'smell alike' fragrances?

    More importantly, given people's inability to use the search function, I doubt they'd use it.

    http://www.sephora.com/browse/me/index.jhtml

    http://www.fragrance-editions.com/fa...id=6140&type=M

    http://www.fragrance-editions.com/fa...id=6140&type=F

  10. #10

    Default Re: Netflix of Scents

    I think this is a great idea, but I think you'd have to start a new database. Even with the huge volume of reviews on Basenotes, I don't think the thumb system has sufficient granularity to support recommendations. And people don't necessarily assign thumbs as a reflection of their own preferences. Often, people give a thumbs up for good construction even while disliking the smell, or give a thumbs down for overpricing while saying that they like it.

    Even in Netflix's system, the most important component is "average review score".

    What I suspect may be more doable is a fragrance similarity map. Just present random pairs of fragrances to respondents and ask them to rate similarity from 1 to 10. From that data you should be able to make a nice clustergram.

  11. #11
    Basenotes Plus
    knit at nite's Avatar
    Join Date
    Sep 2009
    Location
    CA, Central Coast
    Posts
    4,842

    Default Re: Netflix of Scents

    I've used sephora's, Michael Edwards and Fragrantica's and thought Fragrantica's the most accurrate.
    A Scent Rescuer
    Every great perfume deserves a good home

  12. #12

    Default Re: Netflix of Scents

    As suggested already, it's the algorithm. Here, I suspect there's bias; let's call it the bias of The New. I say this because the site's "recommendation engine" is "up-to-the minute" and that it "features the latest US fragrance releases." Hmmmmm.
    But giving it a go, I simply recalled what I'd put on this morning, and clicked on "H" for Houbigant, as in "Houbigant Fougere." Alas, no such fragrance listed.

  13. #13

  14. #14

    Default Re: Netflix of Scents

    Do you understand how these things work? Are you expecting 'smell alike' fragrances?
    Chill man, I teach how these things work.

    given people's inability to use the search function, I doubt they'd use it.
    See my longer comment below.

    I think you'd have to start a new database. Even with the huge volume of reviews on Basenotes, I don't think the thumb system has sufficient granularity to support recommendations.
    That's a good observation. But there are ways of extracting valuable info implicitly even from the conversation threads on BN -think lateral factorization.

    Even in Netflix's system, the most important component is "average review score".
    Not quite. Netflix is continually funding research on more progressive feature selection techniques.

    What I suspect may be more doable is a fragrance similarity map. Just present random pairs of fragrances to respondents and ask them to rate similarity from 1 to 10. From that data you should be able to make a nice clustergram.
    That's indeed what I meant to say. Perhaps it was not clear: Trying to guess what a user may like is doomed to fail, even in principle. The better alternative is to represent graded similarities along different dimensions and let users make an informed decision. For instance, if I like Musc Ravageur in virtue of its musk, I will need a different recommendation than if I like it because of its oriental character. Services like Fragrantica and Michael Edwards fail on this very ground, and simply suggest what you may like without any clue as to what features of the recommended scent makes it likable to you.

    because the site's "recommendation engine" is "up-to-the minute" and that it "features the latest US fragrance releases." Hmmmmm.
    My/ our interests being non-monetary, financial concerns will not be a factor.

  15. #15

    Default Re: Netflix of Scents

    Its a great idea - I would like to collaborate with you. I have an advanced background in web applications, "Big Data" technologies such as NoSQL, large-scale distributed computing, recommendation and search engines etc.
    -

  16. #16
    Dependent
    Join Date
    May 2011
    Location
    Los Angeles, CA
    Posts
    3,652

    Default Re: Netflix of Scents

    Netflix's algorithms are amazing; I find it very accurate to suggest what I'd like to watch.

    Pandora is another service which has developed a huge database -- The Music Genome Project -- with the help of teams of dedicated musicologists. They got funding for the idea and then made the music "radio" product based on it. As a musician, I find it an integral service which helps me research and stay on top of the huge volume of music in the world.

    A "Fragrance Genome Project" would be amazing. I wonder if some commercial end could be found so as to allow more resources to be dedicated to it. Either way, I'm all for it. How can I help?

  17. #17

    Default Re: Netflix of Scents

    Quote Originally Posted by Dr_Rudi View Post
    What makes you say that? Do you understand how these things work? Are you expecting 'smell alike' fragrances?
    Quote Originally Posted by Reza View Post
    Chill man, I teach how these things work.
    I'm perfectly chilled - and you haven't answered the question. These 'fragrance finders' are based on Michael Edwards' fragrance wheel - not an algorithm.

    And it is not designed to deliver 'smell alikes' if that's what you want. If you want an oriental, look up orientals.

    http://www.sniffapaloozamagazine.com...-MAGAZINE.html

    Pay up for a subscription: http://www.fragrancesoftheworld.info/

  18. #18

    Default Re: Netflix of Scents

    Thanks zztop and MonkeyBars.
    What I have in mind is using some sort of web crawler for mining raw text from threads. I'll then parse the text to find latent features (more precisely, using a variant of Word Order model). I don't know how BaseNotes would feel about us using the info over here. Perhaps a moderator can comment on that?
    In case you're wondering, for the decision part I'm planning on building a Dirichlet process mixture model.
    If this sounds interesting to you please email me: rs689 at cornell dot edu

  19. #19

    Default Re: Netflix of Scents

    Quote Originally Posted by Dr_Rudi View Post
    I'm perfectly chilled - and you haven't answered the question. These 'fragrance finders' are based on Michael Edwards' fragrance wheel - not an algorithm.

    And it is not designed to deliver 'smell alikes' if that's what you want. If you want an oriental, look up orientals.

    http://www.sniffapaloozamagazine.com...-MAGAZINE.html

    Pay up for a subscription: http://www.fragrancesoftheworld.info/
    Good. Thanks.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  



Loving perfume on the Internet since 2000