Merken

Music Recommendation - Future Casting

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
but a a a a shape and
a welcome back to stage form
to this 2nd parts of their music recommendation topic and this 1 is called future casting and there is a great team of foreign speakers um and that will be talking about technological development what's uh up in research at the moment regarding recommendation algorithms um or new trends in how a recommendation will can the the how we do you want to work hard work is in the future and to the panel is organized a by Italy guide to NAQ on come on stage had
and they are a key is uh doing lots of PI in uh music and technology and is also part of 4 on the part of uh all together now and had the idea for this panel and is also going to moderate and I think and um as this so we ask the other speakers who have multi high from air to it Spotify we have Amélie Anglade From sound clout and Stephane Baum on um who is a researcher at the D. F. you whatever that's in English German Research Center for the allied you have to take a K but you will all anyway yummy going this is of celsius now and around island Alonso governing this year 2 nights myself tonight but it's like Knight of the so the another might but
excellent here once again thanks for coming here and I'd say um we do little introduction around them and it has introduced myself and um I'd like to go on with Emily millenium about your role and and what you doing at some clout right so many seminal on that and I am a developer at some plants and I have a research background and so i come from academia and I was doing a lot of music information retrieval i which is this field at the border between musicology mashine learning and signal processing and and I don't sound like 2 years ago and since then i've been working on machine learning algorithms for also itself as sound discovery stuff on here on of on I'm heading the more 10 people and we are doing research about music recommendation engines and other stuff about flight 2 to 3 people working on music recommendation engines only and I'm into this kind of recommended this is Vince 10 years so I have seen a lot coming disappearing coming again so all you have do this is basically what is happening over the last 10 years and it is the next you be the research keep on this 1
of 2 of kids I marry I'm I'm working for us but if i in Germany and part of the team that looks after Germany
Switzerland Austria and is manager labor relations which means that I'm looking after their relations to I licensing they have so licenses that means that I'm in contact with all the labels of a new labels all the bands aggregators who want to promote their records and these butterfly so money will also give us some some deep inside of the music industry perspective on this whole recommendation on topic that we can discuss today yeah yeah
vessel and thanks again um to stand and all go for this so great of introduction to the curatorial music recommendation and since we are here for the rather techy part now um I to like to firms starts them with a little
um with a little postulation that we did in the paper before before we can actually handed in from this session here and what he said in this paper was like umm inadequate recommendations of the biggest hurdles facing commercial success of streaming services so i'm meaning and is this really still true since we have some really come from new figures so for instance but if I have some 6 million paying users by now the user says it has 3 million music industry said there was like a big turning point in 2012 them the 1st year of them actually um corrode since the 1999 so this this postulation still come true or do we have to just put the panel at this point no recommendation is all fine and we can just go on and on the day of and way we did before so I'd say um was the problem with this whole music streaming topic and recommendation there 20 million tracks systems so why don't we just pick the effect we like and off we go what was the problem stuff I'm from I would say to produce a take home message already now all the can leave if you don't like it but the point is actually how you design the experience so far to say we have all this wonderful data are like 10 15 20 million songs wonderful all 4 5 box amongst the effect uh and people craft all this front interfaces which you like more by stationary whatever um and sometimes there comes a good emotion into it because you're so happy that you find this rare tune and you can listen to it immediately sometimes there's bad in notion because uh some people just don't know where to start so in the end of our year our go as researchers and for sure the goal of the guys industry to is somehow to shape this experience in a way that all these people will be happy but this is difficult because of some as many people as we have here I guess most of you would have different experiences when you think about it good listening experience when using such services and this is just the statement that maybe there is still a little bit to do and as you can see on the slide here we have this scored of the
guy of teaser which is a huge streaming Service and he stated like uh I think we have invented five-percent of what should be invented to try vectors discovery so I don't know if he's talking about user or if he's
talking about the music streaming industry as a whole and but the court is not so bad the Eric 2 we have a clique
engine of remote play on the slick ends in a fit of the data the and just to get the discussion started an open we get into discussion with you 2 and the 2nd 1 is the of Don Norman was the usability cool since 30 years is like but I don't like recommendation services recommending the error read stuff because I guess I'm not an average guy I just want something which is special for me and this is kind of the Treaty regarding recommendation algorithms that there might be some
problems in the algorithm itself
so for me personally this is very subjective I when I was like 16 or 17 years old this device created the greatest change in my music experience ever some with the Walkman you could listen to music on the go in a city you put I do all these things which are nowadays quite normal usage if you go to the subway or
elsewhere everybody seems to live in his own world for us this was really shaking and the
point which was so nice about this the wisest there you still could use the play with all these wonderful mix tapes that you ex-girlfriends generated for use hate tapes I don't know what there's even a very nice book about the whole mix tape story and all the pointers I wonder sometimes what do we still speak a lot to the past because we love insist still are in the very focus of everything but so
far I had not disperse shaking experience again and some if I look at my plate is now
on the Apple iTunes it's a more like a so what found and I would like to discuss a little bit so what Spotify SoundCloud and we could
do to goal maybe some steps ahead do you decrease so far somehow nobody agrees everybody
wants something the but anyway and linear I jump in with some common sense I mean no don't we usually get like the best human music recommendations from our friends Andrew maybe something to discuss and I think maybe I will I'm actually talk about this a broad so things so that the latest um interesting inventions them on the recommendations like inferences from from Twitter Twitter music was just lounged and the idea was basically um that you follow your friends and just pick up some nice music recommendations from them and that's very simple and and I think that the approach that um Spotify um started all will actually roll outs pretty soon that goes into this direction but he mean if you wanna um just shoal of some of the stuff on that you that you have read have to switch computers now but there could Dickinson Tecumseh man yeah the you can you can continue discussion we just fix them the so we should discuss about the rights K I was here so sample artists working with play this while we have sets and so that's on on the interpretation of data is found the answer anybody can actually create the yeah sets of stuff sounds and they wanna put
together and some and some matches use
that to create the album that ladies but some others just to yeah and maybe something to play for your friends or anything so yeah we we also do have that when and focusing a well I'm and managers like yeah for users to be able to build something is any to right OK but I'm S and some
input to this rather social recommendation idea and I would like to just so social what sporophyte has in mind on the recommendation from not only what to what we have in mind but also give you a short overview of what we did in the past and how they service developed over the past years so I think that you know maybe all the way out how the service looks
like at the moment and how it works and how you see I have friends are listening to certain by by or what we call a tracking system on the rights
and then the I'm going to show you how this whole strategies developing in the future and also showcase uh within the employer version I have on my laptop that's the reason why we needed to change it and how it's going to be like in the future and it would change a lot because we have seen and when we looked at the users and we did when we looked at a product that's what if i is the great and 2 for everyone who knows what's here she actually wants to listen to so searching is great works great you just in what you when I listen to and you hear it immediately but it's not so great if you have more like a lean back user who actually wants to browse and not actively search for something and this is really really important was also from a from a matter of from our from from the group on the perspective we have to be a mainstream service so we wanna reach a mainstream audience and not all the the mainstream users actually know what they want to listen to so what we have now of of a user base is quite the early adopter user but we wanna each everyone everyone in the market everyone in Germany everyone over the words and it is the index search and browse mode is very important for us so we developed a few other things and possibility to them the to discover new music you can always of course search for music on your own when you have an idea but recommendation means for us we actually providing new things so user so what we did in the past is that
we had this section called friends and sharing and social conversations so this is what you know from phase look and this is no this is what you know from the right side of the client at the moment so you see what your friends that you have on Facebook and said that you of the following right now I listening to at the moment few examples of this for example if you follow Bruno Mars at the moment you would see what he's listening to this is a kind of a global example but of course we have that German artists we have German media and were active actively listening Spotify and you can see in a life experience model what they are listen to at the moment then we have something else and plays that we call professional curation and professional creation is being made being done by apps with this modifies apps are likely the program's just like an iPhone would which 1 you can install in the clients and those apps helped to um to recommend to new stuff and also to give an overview of what songs we have in the catalog so a few
example of these are for example enable apps say from pious or maybe universal then we have apps from
artists like and from from other magazines like Rolling Stone and also some played as partners that to go was just as quiet that our company is what the and then as the last parts which we think them even more important and which is also based on algorithms which we call intent and
recommendation that means that we're looking at them interesting and stuff that we learned from my users that means for example what do users listen to that other users isn't as well so this so this local Amazon recommendation I would say and then we look at them what's our people listening to and maybe try to get them into buying tickets as well by merchandise so I would show you a small video voters will look like in the mail it in the very near future so this hasn't been rolled out yet but it's going to be there very soon no it's not when taken the
you need a um of Angola help the but yeah this is the blue we have this given unfortunately so am I tell you with the uh with this slide what it's going to be like the that social discovery means that we're going to have an endless feet of recommendation stories so what could and air recommendation story be that could for example be uh a new release that's 1 of your favorite artist has just released or it could be a plate is that 1 of your friends the
following has shared or it could be uh opportunity to abide for concept from another you may like and all these social needs and social stories will pop up under new we coded what's new page I would show you in a minute and also earn these stories will pop up as notifications for example on your as a push notification the born or maybe as a push notification within the client I'm going to show you this in real time so this is my and quantify profile evidence of it all too because on his into a lot of crappy pop music as you will see um and this means who receive the universe castle pitiful but you to the in in of find not find the right window do that the this OK that's good yeah so this is what we will have in the future instead although which you may know about the what's new carousel with their records on this will completely go away and so that we will have these soldiers always looks a bit awkward with this layout looks better normally so what I can see here is that I've heard a song from Cayley's and it's recommending me a song from a doorknob because it's similar to that from an algorithmic view but also I see that my friends as crimes has share the a song which you would see on the right if it was possible and below of that's and the system recommends me and you reuse of uh being delegates because I listen to a bunch of voice which is uh similar to that according to a system and then you will also have the possibility for example if you are at the moment listening to the luminious you look and and you ground to me on their feet in always checks whether you might like to also a donation as a kid a review of our thanks for the note is as long as you want and if you like it and maybe say appear when it is to that later on in 1 of his mind his name is at the moment but maybe save it for later I can I can save it and updates then there is of course not party and input as well so we have them reviews some pitchfork was sample for phosphorus and here or I have played a remained recommendations for songs that I've already listened to which all featured in a certain playlists and then there's of course them ticket recommendations as well and check if there is 1 at the moment the so this is like the an endless feed you can do this for weeks or years if you want to the but it doesn't look like as if I have a ticket recommendation at the moment but I could have um if I have listen to a song artist recently which is now being on toward this system with suggests to me that I could bite you can go to the show and share it to my friends and ask them to come with me so this is where how it's going to be in a couple of weeks and say as so the whole um recommendation system all what if I will develop very quickly and into this what we call social discovery I think that that cool so the thanks money the for them this little demo so um this experiments the becoming like in a couple of weeks what you that's what you said right and you think and we we talked about ah think also Christian talked about um the Filter Bubble problem them uh and um what we had in discussions before I was like umm if you always like get great recommendations from your friends and the friends of your friends which I thought of culturally related to and a part of your social environment could be a trap because of this won't help to get actually new input or all we call it this maybe um the destruction needed in this room social recommendation stream and so what's the
what's the solution to that yeah I mean this problem is well known the are we know it from Amazon we know it from last of uh you have seized the vector loops and biases nothing new comes in with such a system of the Lotus long tailed the poor independent artists will never end up in the playlist because everybody is adding the stuff of friends and celebrities well this fishing in the same code so this is kind of also very strong argument against it but I mean it's it's fair enough if you trick to mainstream market you you have to ask what up people listening to all this is rather a kind of philosophical question do we need algorithms which are fair and objective to each independent artists updates on something else so for me as an academic and doing science the answer is pretty much clear we we favor algorithms who are completely objective and rely on musical facts and try to get the user into experience exploring this by himself but I'm not the own else but if I I don't have to drive a commercial company so um nothing against these algorithms but they are the ups and downs shows this just a smaller many here focus on on a mainstream market or would you say like there's also like to hope for for the long tail so to speak well it always depends on any of the individual user how who and whom he wants to follow so if I choose only to final Trudeau podium and rock-art I get only recommendations by these I say me magazines ebenso so never they were never be Arianna popping up in my feet so it's always my own Charlie choice who I want to follow and whom suggestions I wanna have my feet so it's always I have to of course they have to know the ones that I want to follow the swimmers of where exactly the the point if you try to sum all educated people you can't you know and I mean the independent Gaza very happy because they control each other and never drop something in from the mainstream which could be interesting and vise versa this is kind of a very strict philosophy either you call this direction all or into a new 1 which may be only in the lab so far commonly to and you know I don't think it's on in the lab honestly and like and so on SoundCloud and we've been trying to bring discovery and and serendipity to people on and I guess a trying to focus on making sure that they can discover new things and things that are trending or or you know getting some traction on the site and and and that's that's what we do is the explore page and then I'm also making sure that when they get stack or when they are on a sound and we don't know where to listen to next week we suggest some sounds to listen to that are somehow similar opted in term of of some properties in listening time of effusive properties on so yeah you use food all songs saluted them on inferno that this is necessarily Nobel like we use a lot of things that that yeah it's more like yeah connecting sounds rather than this and people know just a deadly at get the next to the other machine online against people start to look yeah so and if the I agree I think like it's a I think we need to have posts and we we should be also like making sure that people get more of what they like have that also probably sometimes make some discoveries and that doesn't mean that it needs to being very in always in long tell about the long tail can be nice as well and again like the long tail could be the alters of things it could be slightly different shards then would have been listening to or and completely new Jorry's or or on some wheels are likely to have but also some continent music content and maybe sometimes you just like to listen to the radio show something in the that would just as again be able to digging in these huge amount of content and allied and not just having a search box to deal with if you already experiences of the whole people in which kind of choices they use maybe a lot rather a social bond or a SOM-based 1 because you have the opportunity to track the users with saying actions right now we've been starting to leak into that and and that's not my team directly so I can share any any results of that but yeah we were really interesting as well again should I like you to little demo can get share from rates and yeah so I like to demo explorer and of some of of the the other recommendation features of that so I already said some of that but I I want to emphasize on the fact that our some in might seem that we call the discouraging we deal with search exploring recommendation
and so on might be a bit strange but like actually search was the 1st starting point because and we need we actually need to make sure that people actually the get what they're looking for at 1st but which is not always easy um and so yeah so making sure that all the the search results are relevant for the users and for each and every user and from there on we actually decided to also work on on Explore so this possibility of discovering new things and and then recommendations so I'm not even
having to dig into any any sort of page beginning the recommendation straight in your in your listening experience riding and
so I think I'm going to switch to the live demo if you can do that from
yes OK so I mean and I don't know if I need to reintroduce some pervades the world's largest sound and social some platforms are so and again I insist on that we don't just do music we have all sorts of songs and and that means
also known music and content radio shows and podcasts and and and and everything you can imagine from also like your friend sharing is there an everyday life sounds and so on and so
this is the the base you get when you um when you are not lending so gonna login and the the rights and and then so as as a user
and all users are the same for us so you could be a used a very famous artist so just um someone
that is more into ordered the listening experience from them you can always have an actually share your own sounds and so we call them and share them on the stage of the stage and this is my own page and I am not a heavy user of the recording and functionality because I I don't do that kind of thing but I have I have a few things in there which are sums of recording myself like this 1 or uh some set every posted from other people and years and some from 1 of my colleagues pull on the recorded a song from the bending get
sycophants and arm so did the concept is that and I can
affect image that any better of well and is that yeah you so you have
this overload function where you actually recalled aurochs files that you upload and and then you can share anything you you need in their army
entered but that means also nature of much more famous artists like and
Snoop Dogg on here for instance by an army as he get new name so you can use it yeah you have much more traction on your on the and sites of in your page and on the so the
your outer space as well and then you are actually indeed I have to trace if we have more time there um yet
for every every sound you have actually uh sound page as well and I'm this is where I can actually so 1 of the functionalities of the recommendation of 1 of the recommendation functions we have on this side is the related sounds so once you've done was 1 it sounds are listening to 1 side and 1 son page like here you would actually switch directly um due those sounds that even like noticing at 1st probably and those are sounds that we think are related and you should buy these into next after that sound so that's a good way of not
getting stuck in in your listening experience and discovering new things but more interestingly I like what I would like to show today is on or explore page on the information that
and the and so the principle of explorers that you could be I'm unsure of what you want to listen to now I know what used in mind specifically no sound in mind and you just want to listen to something
new and so we have a stage where we we have also and and
again in music and non music content so uh musical content on top and then on non-musical counted here and and you can decide on 1 specific degree for instance from here and then you
would have said Jorry's in there or and
in subcategories in the non music and and things as well
and but if you go in there and then
you for each subcategory or category you have a bucket of sounds that you can listen to you like here and and features so and those sounds are the sounds that trending on end by trending we don't mean all these like being sounds that everybody's listening arm and I getting out at you know like the goddess like it it's safe Mattaniah releases a you sound and it would have a lot of traction of course but we also wanna show those like suddenly sounds that are getting attention and anatase Sicily isn't by everybody but I'm getting a lot of attention suddenly and then we
would be would automatically be on that page because all that is automatically created
by and by algorithms and will buy people's on gap so trend analysis and detection is is the basis of those algorithms and thank you
so much Saul trendiness them as a factor so what does the term it was just the LEP guys say about this recommendation criteria here again it's it's about how you implement
trendiness in the algorithm tho in the mathematics it slide is it a very obvious scores thing like the stuff was getting a lot of traction in a short amount of time this 1 will be the 1st to propose and that's all of them will either be more fine-grained and looking to 0 I don't know peaks in very very small amounts of time so I could foresee anomaly for sure tool thousands of possibilities how to implement the algorithm and this algorithm is then driving the users to pick up stuff or not so the responsibilities in the team was developing the algorithm right and yeah but again we honored like decorating that by hand so the I agree I mean we've been tuning those other a lot experimenting to see me I should be should show period of times over much longer ones and so yeah we we we've been still working on this to be honest and I agree you get very very uh different results depending on that on barrier or o concept which is that you give something fresh to the user and and that's that's what we want as as a way we rephrase that a lot of we kiss and we want people to it's time to go to that page even if it's 1 hour later should get something new and something that he would like to listen to and um trendiness or what you own actually generated from like this recommendation based on trend in a so what would what would be the report to hear some more examples to and the moment where well we have something that we call this butterfat top lists of course it would just simply chance of what is based on the plate is placed but also of course within playlists we have we see that if it if the the track is is featured in and out of Paterson a lot of other influences pick the pick it up it's of course and volume and an and other approach but we see these these trending topics as well but we don't measure it at the moment we don't really bring it up to the user 1 would that be an interesting future for the future you have to offer developers of the vendors here OK the but I guess so and then I was wondering and
since sound rather or used to be like a platform for creative wasn't a policy was rolled out was the initial idea to to share your own productions for instance this this like a reason why you have different recommendation criteria
and so no we don't have different
recommendation country out so we want all creators to have their sounds are found and and M. and discovered so we don't bias anything at all all creators I
call even the very small artist uh there because as yet we want them to be recognized as well so um concept is that we want every listener to find interesting things in every creator to get their sounds found as well that that's there conflicts like for instance of a new world order new user interface and that I have been tolerate that the 1st day the delta him some criticism from the creative side thing orchids much more listen the focus now Lord has changed a lot you can see the commons as well if you've seen it before hands the at was honesty and I work on those issues specifically but and I still think that the yeah so the creators are still a very important to us and it's still 1 of 4 focus but it's true that we also 1 the listeners to be happy and and enjoying the experience as well in right and so on told us them again like and coming back with this with the with the same question but how do we get the destruction and into this some system like the surprise effect on yeah I think so Omn 8 it have to some extent you still have surprise because you could discover things that you don't know about ensures that you still
um enjoy a lot already from yacht I think it's a it's a step 2 words uh serendipity but you you can probably even have more like it if it's a son is the trending then it wouldn't appeared is any could still be a very surprising and interesting and sound to listen to most definitely them to have a look into the left here but you have to change in my OK I through the the change so I I had been criticizing the other approaches telling me if you rely on people you might be not objective enough if you rely on algorithms maybe the developer is not objective enough so at least I should understand another approach so
we don't need this so I called it the break graph so get result rid of the play with assume a user which is willing to be active i and then you can do something there is
so a data collection which is called the Linked Open Data and there you have it millions of facts the vault songs spans movies actors Wardell and all these concepts are related to each other by a very simple mechanism which is like a subject is related to an object for example you see that the Thomas masters so the thing all the inmates Phoenix's band was France's area is related to orange the to a manager of orange was kinda crazy and so on and as you can see the if you start at a certain point you can travel through this entire graph for always fall lifetime so actually we use this craft and put it into the machine
and we build Luo demo system which is called Austin austin short for ballistic recommendation and storytelling technology just to get costs and and I have some examples
and I need to play this area so since we are talking about music we should play more music that's right is it already running was not
every you are such such moderated you know what do we have sound the and it has a certain influence on the reports of you the man all the law of England and all of that was an awful lot of of all of the things had
all the words in
the in the in the
kind of model the hearing
and the goal this and
and and the time the you know and
and
through the if all this 1 of the major when they have been very young like 10 years ago or so from the 1st along course called if I have a few that and the point is just by contrasting the sound examples of these I listen to both of them and the point is this this somehow a relation in between these 2 artists which might year bring some surprising new facts to me and engage myself into something different and here for example we automatically create this structure here and you can see that they're in my God what little niche the what um the Thomas bond there is minimal stuff plant and he was member of Stalin's I have never heard of Stalin before using this tool and
the parental which is a member of the needs and he was a member of thought into and then I was checking a little bit and indeed when these guys beyond and meeting in Versailles and having the 1st bands they already play together in this band and it could be given to me to say come on let's say what darling is about maybe it's a mixture of the songs I like already
so I have a different example say in then can be and you I don't know what it if you got this media campaign of stuff upon their releasing the new album by the end of made and they put this wonderful teacher
documentaries with all these legendary perfuses like sort of mode of how this guy actually they i and j if you have more solitary and that some you hear the funding itself calculating and it was done by Ni watches so I as if the relation between i watches and this stuff punk tune I played before and again you can see OK not autos has been performing the fund it up on so many tracks the played for these David Bowie tu'm that stands and then
suddenly released this David Bowie area and see the Robert what song was sampled in a recording called come again and this 1 contains samples of under pressure which is the sum of David Bowie so you can argue yes but come once this is a very strange story we start with this funky guitar player and the funky electrifying all stuffed bond and then we end up somewhere with the David Bowie song with samples but we don't know what kind of user we could expect maybe there are some freaks out so what
exactly into this the or something
different and ask for after that so I think you listen to it it's a very similar baseline the I can do that if it's like but bomb bomb and then in the good times thing it's like followed up up up and in the other example it goes like but up but part about and here we are by another 1 bites the dust and I had both humans in my blade is paid in those times and I now make a look up how is another 1 bites the dust related to good types besides this funky baseline all off tells us OK this is a song by creating uh and stated other stuff like the Republic of June Bohemian Rhapsody either this thing it's sometimes or then it yeah from time to time I actually area and in the end both songs ended up owner release which is called the direct house and sold strange enough but that's things that another 1 thank at
this is the i isn't coming up later and here this is actually if you know the whole thing here but and this guy in the world and so if you have not here to the child alone you pick up yeah your best friends coalport uses a former times great singers and you create an entire album by creating songs with these guys and this is what exactly the cozy did and he did 1 song this year and I mean did stop at 20 December issue will how does this relate to Berlin experimental deep balls they are from Hamburg he cut those from much you in this well let's let's so the the the questions answered they from the same
city replied in what about it will be an I yeah so so what is it's much smarter than me because as you can see the connection people what this release it was the case that already cozy differs therapy when they have been together with the small this units of Czech thought and this is the topic from on so if you go to the actual album of cozy with the you might get on to this there will be 3 meters to and this is the entire trade
maybe I can show something funny if you are interested in the they get at and 9 inch nails just to see if all of us something even in this case there is a relation so I don't have sound as in this case
but I I mean the performance the come up with this tool at force on the October 1st it's where
99 9 knowing no understood by no world was on and there's this English version called red balloons and this guy here was an additional member of mine a Nate's played the best space so the point is that there is no 1 ever was made magic in I mean for us it's very tough to navigate this graph it's slide 9 millions of this notes and 22 millions of these edges and it's like sitting in main memory we like to to speak by so we have to do a little bit of stuff but it's not so much about tweaking the trending Paramita while relying on the social play by other guys it's just by getting in this open data which is out there and then playing around with so this is our answer to put it another component into the entire play how to create recommendations and as I was saying in the beginning I think it's very essential there not 1 solution it's really about what data do you have wow how might you sign your interface to get it very emotional to create really a good user experience and this is done great by Spotify from create by some plot it's done rate by us but not for the entire world you have just to see certain use cases certain people and this seems to be much more complicated than that we think at the moment and actually you know some for
me yeah very traditional experiences thing if you look at the
picture you see a a 12 inch and you see me quite happy and I'm sitting in a blind eye there's the in the background and I'm kind of recommending with some sort of this 12 inch vinyl um yet because this is creating for me a very special experience and I guess we have to work harder on on these issues so the discussion is open great thanks the way another 10 minutes to go is to signal to him so I think we also would like to um involved you guys into the discussion that I should we have some open the Q and a so at this point what the um tested on you and I'd like to ask you how you mean how do you come fill your own a system but the data I mean who was further bandleader of whom and who I mean because that this thing is because some plot had this and training the algorithm to to detect trendiness and by shot the periods where some place uh let her that had to you that because you have to set an onion desktop and end or the new little time I was not precise enough at this point These LOD linked open
data thing from is the certain all work which computers can read which is collecting open data which is around for example there is a machine readable form of Wikipedia which is called DB pedia as you can imagine if you go to Wikipedia to feed to find the diagram page all facts and this is the computer readable form with having all these so-called triplets like subject predicate object and this is also available at MusicBrainz at Freebase and we're just using some of these bubbles but it's already 9 million nodes 22 million edges and the year we have to see what we can do with a lot of research but anything may I I don't think that the long tail is out there in those datasets yet I right so that's the problem so it's it's really interesting and I agree I I think this is kind of the future of their music recommendation that yes some new artists wouldn't have any anything reference at their own not searchable that easily and and so they wouldn't be in the system in and
zone uh might remember some of this music user but if I a very much like it and some of them and staggered lost because there is such a big space of recommendations and whatever what I could choose and do you have something in mind like a personalized factorizations so that I can go in this direction or in this direction that a kind of canned the see what the different factors for my music this our not to that understand correctly but M I want answers that we offer like an active system of recommendations with 3 different pillars so he can choose if you want to be uh and if you want to have recommended music based on algorithms or if you want them on the final function or features in the just wanna see what your friends are listening to the views of the options you have and you can always scale these options to your needs and of course you can always don't use any of these you can always search for music however you want to to be like back in the days um I think do you know OK what I want us in a
certain situation I want to narrow down these recommendations just to some specific direction but I'm interested in so I don't care about all recommendations at a certain point in time because I just wanna listen to certain the Kneser-Ney in a specific genre are you have a but a drummer
is like a given category so maybe something of more brotherless so that the point is we have to guess
who you up and in what context you iron and we have to do this in a
digital way which means if you and that have been very active we know already for example that you like these edges which have contains samples all of it and maybe in the future you'll have kind of devices which detect you about emotional response and stuff like this and you might guess OK this guy is and the sad mood and in previous times he always favor this kind of songs female artists and then the system itself the prediction but it can be completely wrong because if you are in deep of emotional depression about the ex-girlfriend maybe you like to read reading false statement and you want to have so sad songs the set songs ever and maybe the systems there's this guy so the press you know we need something uh which really makes them happy so but the point is really how much does the user want to be predicted and giving up this free will or how much time is he willing to spend on in action and this is for me the most important question for future interfaces how much will be done automatically without asking you quite effective or how much are you willing and not lazy to do some things right making it so so good point because they can there is a lot of
research being done on on context in using the context information in time is 1 of them so if we don't know anything about you context that is no which time the babies out where you are of the apparently and and sound and then maybe we can infer that you are commuting to work or you are joking or you are just waking up and that kind of thing and and we commend things based on that yeah so that's that's 1 direction that we haven't covered here but there is also a research fellow lot I think I don't know they are an actual products using that lot yet but if it's getting there we have time
for maybe 1 more question then we're through any other questions from the audience OK I take this as a could to put it so and
so we do try to do a little of summary norms is boring we all want to to go and drink beer but I'm still and I'd say um as long as the sum of these new technologies are not most fully implemented and deployed I mean them and actually 1 point that I want to add to to harvest them as long as the most doesn't tell me why he actually gives me that certain recommendations recommendation it at the end at the neighbors still will have to stick to our own research capabilities and that brings us back to the system and all Connecticut has but them they're gone already can but that the any final common so in this round however think a bit more note of I think that the users off of fine so far we always have this kind of discussions about the poor long tail the poor independent artists and of fall yeah spending my leisure time I'm independent artists to um but I don't care too much if people were gone algorithms to do this it's it's always yeah I'm not sure we like to do this kind of research but I don't know if it's such a old of random problem for the everyday people they use 45 they use ontologies that big successes and some I don't know it's in and I think it's a it's a good point because like we we probably are um like listening to a lot of music every day and but like there are people that just listen to a few tracks of a week and that nothing so we we should also kind mind that we need to accommodate for the different I use cases and and not everybody needs to discover new music every day might actually be overwhelming for many people to do so and some people just wanna listen interest at the end they really like ants end up over and over so we also need to keep that in mind actually yeah I really love the the awards idea a lot and
it's very entertaining as well but I think that's man may be more for we dt is people and and
most of us almost all of of mainstream users that we wanna approach are maybe more interested in simply a work of plate is they 1 to turn on and I don't know why they have this has been recommended to to them they just simply want to listen to some music that's it all right then let me say thank you to Emily Štefan and muddy and of course to them and they are and the beautiful Fico organized to have us here tonight and
them if you want to we will
continue this discussion also with some other people I met here Berlin music we coming up in september and if you want to have some hand-selected music later on i'm gonna play records would unveil here at the refill bomb for your amusement bigger much for coming
and have fun here
you know if you if you
Bildschirmmaske
Gerichtete Menge
Algorithmus
Twitter <Softwareplattform>
Momentenproblem
Gruppe <Mathematik>
Mereologie
Elektronischer Programmführer
Auswahlverfahren
Information Retrieval
Virtuelle Maschine
Multiplikation
Analoge Signalverarbeitung
Algorithmus
Datenfeld
Rechter Winkel
Gruppe <Mathematik>
Mereologie
Softwareentwickler
Schlüsselverwaltung
Arbeit <Physik>
Datensatz
Datenmanagement
Perspektive
Gruppe <Mathematik>
Relativitätstheorie
Mereologie
Soundverarbeitung
Befehl <Informatik>
Bit
Punkt
Quader
Objektverfolgung
Rechenschieber
Arithmetisches Mittel
Dienst <Informatik>
TUNIS <Programm>
Gruppe <Mathematik>
Vektor <Datentyp>
Systemprogrammierung
Figurierte Zahl
Axiom
Message-Passing
Dienst <Informatik>
Algorithmus
RPC
Benutzerfreundlichkeit
Gruppe <Mathematik>
Eins
Clique <Graphentheorie>
Fitnessfunktion
Fehlermeldung
Punkt
Algorithmus
Gruppe <Mathematik>
Magnetbandlaufwerk
Mathematisierung
Besprechung/Interview
Mixed Reality
Vorlesung/Konferenz
Zeiger <Informatik>
Fokalpunkt
Interpretierer
Bit
Inferenz <Künstliche Intelligenz>
Twitter <Softwareplattform>
Menge
Rechter Winkel
Gruppe <Mathematik>
Stichprobenumfang
Vorlesung/Konferenz
Metropolitan area network
Computervisualistik
Richtung
Dienst <Informatik>
Datenmanagement
Momentenproblem
Objektverfolgung
Matching <Graphentheorie>
Rechter Winkel
Vorlesung/Konferenz
Systemprogrammierung
Ein-Ausgabe
Videospiel
App <Programm>
ATM
Facebook
Momentenproblem
Mathematisches Modell
Gruppenkeim
Versionsverwaltung
Online-Katalog
Biprodukt
Strategisches Spiel
Client
Dienst <Informatik>
Automatische Indexierung
Perspektive
Gruppe <Mathematik>
Notebook-Computer
Hypermedia
Strategisches Spiel
Garbentheorie
Wort <Informatik>
Optimierung
Phasenumwandlung
App <Programm>
Abstimmung <Frequenz>
Algorithmus
Internationalisierung <Programmierung>
Mereologie
Vorlesung/Konferenz
E-Mail
Videokonferenz
Sichtenkonzept
Momentenproblem
Güte der Anpassung
Keller <Informatik>
CAM
Ein-Ausgabe
Homepage
Rechenschieber
Demoszene <Programmierung>
Streaming <Kommunikationstechnik>
Datensatz
Client
Echtzeitsystem
Rechter Winkel
Bildschirmfenster
Mereologie
Stichprobenumfang
Pay-TV
Vorlesung/Konferenz
Systemprogrammierung
Grundraum
Programmierumgebung
Hilfesystem
Resultante
Punkt
Quader
Gruppenoperation
Code
Richtung
Eins
Homepage
Demoszene <Programmierung>
Virtuelle Maschine
Loop
Algorithmus
Gruppe <Mathematik>
Vorlesung/Konferenz
Auswahlaxiom
Parametersystem
Content <Internet>
Kategorie <Mathematik>
Stochastische Abhängigkeit
Bitrate
Arithmetisches Mittel
Vektor <Datentyp>
Systemprogrammierung
Ultraviolett-Photoelektronenspektroskopie
Normalspannung
Steuerwerk
Tabelle <Informatik>
Lokales Netz
Formation <Mathematik>
Webforum
Videospiel
Algorithmus
Softwareentwickler
Content <Internet>
Virtuelle Maschine
Vorzeichen <Mathematik>
Systemplattform
Quick-Sort
Computeranimation
Fisher-Information
Homepage
Demoszene <Programmierung>
Einheit <Mathematik>
Gruppe <Mathematik>
Information Retrieval
Benutzerführung
Prototyping
Mailing-Liste
Datensatz
Gewichtete Summe
Element <Mathematik>
Funktional
Login
Computeranimation
Homepage
Overloading <Informatik>
Elektronische Publikation
Weg <Topologie>
Natürliche Zahl
Funktional
Elektronische Publikation
Raum-Zeit
Computeranimation
PROM
Lesezeichen <Internet>
Gruppentheorie
Vorlesung/Konferenz
Laufwerk <Datentechnik>
Scherbeanspruchung
Normalspannung
Streaming <Kommunikationstechnik>
Instantiierung
Fisher-Information
Funktional
Ikosaeder
Computeranimation
Homepage
Umwandlungsenthalpie
Klassische Physik
YouTube
Euler-Winkel
Content <Internet>
Radikal <Mathematik>
Spieltheorie
Telekommunikation
Computeranimation
Minimalgrad
Gruppe <Mathematik>
Zellulares neuronales Netz
Schwebung
Instantiierung
Magnettrommelspeicher
Klassische Physik
Kategorie <Mathematik>
Ordnungsbegriff
Vorlesung/Konferenz
Telekommunikation
Hill-Differentialgleichung
Normalspannung
Computeranimation
Homepage
Resultante
Objektverfolgung
Momentenproblem
Mathematik
Mailing-Liste
Bericht <Informatik>
Frequenz
Term
Teilbarkeit
Eins
Homepage
Rechenschieber
Fraktal
Algorithmus
Twitter <Softwareplattform>
Rechter Winkel
Endogene Variable
Basisvektor
Vorlesung/Konferenz
Spezifisches Volumen
Softwareentwickler
Normalspannung
Einflussgröße
Analysis
Feuchteleitung
Soundverarbeitung
Subtraktion
Benutzeroberfläche
Besprechung/Interview
Versionsverwaltung
Systemaufruf
Biprodukt
Systemplattform
Fokalpunkt
Inverser Limes
Fraktal
Rechter Winkel
Vorlesung/Konferenz
Systemprogrammierung
Maßerweiterung
Ordnung <Mathematik>
Instantiierung
Fraktal
Algorithmus
Graph
Mathematisierung
Kontrollstruktur
Wort <Informatik>
Monster-Gruppe
Softwareentwickler
Demoszene <Programmierung>
Linked Data
Objekt <Kategorie>
Virtuelle Maschine
Gerichtete Menge
Punkt
Datenmanagement
Graph
Flächeninhalt
Gruppe <Mathematik>
Stichprobe
Flächeninhalt
Systemprogrammierung
Computeranimation
Gewöhnliche Differentialgleichung
Fraktal
Klassische Physik
Flächeninhalt
Rechter Winkel
Gruppe <Mathematik>
Ordnungsbegriff
Bericht <Informatik>
Gesetz <Physik>
Ähnlichkeitsgeometrie
Computeranimation
Metropolitan area network
Subtraktion
Punkt
Extremwert
CALIBAN <Programm>
Mathematisches Modell
Ähnlichkeitsgeometrie
Ranking
Computeranimation
Keller <Informatik>
Spannweite <Stochastik>
OISC
Grundsätze ordnungsmäßiger Datenverarbeitung
Speicherabzug
Datenstruktur
Rungescher Approximationssatz
Zusammengesetzte Verteilung
ATM
Bit
TUNIS <Programm>
Objektverfolgung
Verbandstheorie
Gruppe <Mathematik>
Relativitätstheorie
Hypermedia
Ähnlichkeitsgeometrie
Quick-Sort
Computeranimation
Druckverlauf
Datensatz
Gewichtete Summe
Flächeninhalt
Datentyp
Güte der Anpassung
Mereologie
Stichprobenumfang
Ähnlichkeitsgeometrie
Computeranimation
Einfach zusammenhängender Raum
Einheit <Mathematik>
Meter
Addition
Bit
Punkt
Momentenproblem
Graph
Relativitätstheorie
Güte der Anpassung
Versionsverwaltung
Plot <Graphische Darstellung>
Bitrate
Ähnlichkeitsgeometrie
Raum-Zeit
Data Mining
Rechenschieber
Forcing
Festspeicher
Zusammenhängender Graph
Algorithmus
Punkt
Vorlesung/Konferenz
Plot <Graphische Darstellung>
Systemprogrammierung
Frequenz
Quick-Sort
Sichtenkonzept
Fächer <Mathematik>
Funktional
Zeitzone
Teilbarkeit
Raum-Zeit
Richtung
Computervisualistik
Homepage
Konfiguration <Informatik>
Objekt <Kategorie>
Prädikat <Logik>
Virtuelle Maschine
Bildschirmmaske
Knotenmenge
Diagramm
Gruppe <Mathematik>
Systemprogrammierung
Steuerwerk
Umwandlungsenthalpie
Prognostik
Befehl <Informatik>
Punkt
Menge
Kategorie <Mathematik>
Gruppenoperation
Stichprobenumfang
Endogene Variable
Videospiel
Systemprogrammierung
Kontextbezogenes System
Richtung
Fisher-Information
Biprodukt
Kontextbezogenes System
Richtung
Bit
Punkt
Algorithmus
Gewichtete Summe
Objektverfolgung
Ontologie <Wissensverarbeitung>
Gruppe <Mathematik>
Stochastische Abhängigkeit
Unrundheit
Systemprogrammierung
Normalvektor
Steuerwerk
Metropolitan area network
Datensatz
Gruppe <Mathematik>
Besprechung/Interview
Gerichtete Menge

Metadaten

Formale Metadaten

Titel Music Recommendation - Future Casting
Serientitel re:publica 2013
Anzahl der Teile 132
Autor Eitel, Eric
Baumann, Stephan
Anglade, Amélie
Heimer, Marie
Lizenz CC-Namensnennung - Weitergabe unter gleichen Bedingungen 3.0 Deutschland:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen und das Werk bzw. diesen Inhalt auch in veränderter Form nur unter den Bedingungen dieser Lizenz weitergeben.
DOI 10.5446/33587
Herausgeber re:publica
Erscheinungsjahr 2013
Sprache Englisch

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract The selection of music online is enormous. But many users would like better recommendation systems. Experts say inadequate recommendations are the biggest hurdle facing commercial success of streaming services. Die Auswahl an Musik im Netz ist riesig. Doch viele Nutzer wünschen sich bessere Empfehlungsfunktionen. Das ist die eigentliche Hürde, die dem Durchbruch von Streamingdiensten noch im Wege steht. Waiting for the UberRecommender Streaming established itself as one of the top distribution channels in addition to downloads and phonorecords. Spotify and Deezer are caught up in arat race for world domination; an IPO or other glamorous exit seem to be only a matter of time. But are the products currently being offered really sophisticated enough? One hears users and experts alike complaining about the lack of creative and above all more powerful recommendation features. Many users see this shortcoming as one of the main blockers for music streaming services to make a commercial breakthrough.

Ähnliche Filme

Loading...