Make the Internet Smarter at Helping Us

Posted on Feb 1, 2011 | 96 comments


Recommendation Slicing

The Internet has a problem.  It’s one you know well. The signal-to-noise ratio is getting out of hand.  In non-geek speak this means that there is so much information out there now it’s hard to separate the good stuff from the rest.

We all now visit UGC sites to learn about products & services before we use them. I’ve been thinking a lot about this over the past two years and bending the ears of any entrepreneur who will humor me to hear what I think the solution needs to look like.  I call it “recommendation slicing.”  I guess I sort of want “Pandora for everything else.” Why can’t I have it?

Here’s the problem and the solution set.

Problem
So I go to Trip Advisor and I want a recommendation to stay near Laguna Beach. I was thinking about the St. Regis so I went to see what people thought. Years ago we had been to the Ritz Carlton nearby and we were saved by Trip Advisor because a commenter had mentioned that one of the wings was under renovation so we were able to book the right side of the hotel. So I’m drawn to UGC reviews. But, oy, is it serious effort.

So there it is – the St. Regis, the number 28 recommendation in Orange County.  Hmmmmph. I thought it would be nicer?

And then I notice in the list of hotels above it: Best Western, Doubletree, Residence Inn, Howard Johnsons … all score higher!  Now, I’m not a hotel snob (ok, I kind of am) because I wasn’t born with a silver spoon in my mouth, but at 42 when I want to take a nice vacation I’m looking for a nice vacation.

And it dawned on me – every freaking recommendation website out there is just one big amorphous mass of group recommendations that don’t relate to me. They are the wisdom-of-the-crowd, sure, but a different crowd than I hang out with these days.

So I had to find out exactly what WAS wrong with the St. Regis. I started reading the comments of reviewers. It seems that people were complaining that there was a kiddie pool too close to some of the rooms and there were people with their kids yelling and screaming at 8am.  “Never stay here again!”  This was a few years ago when I was in the kiddie-pool phase (now we’ve luckily graduated to boogie boarding!).

All I was thinking was, “Hallelujah! Yippie! My kids can yell at 8am!  Sign me up, baby!”

And of course I know that some 25 year-olds are thinking that they’d rather stay in somewhere more hip, more alive at night & probably a bit cheaper.  That was once my demographic, too. We’re the same in some ways (my wife says my real age is still 13) but we’re different in others.

Food is no different. Whenever I’m trying out new restaurants I always feel compelled to check Zagat & Yelp and then STILL email friends to ask their POV on my top 3-4 choices. Cobbling together the historic wisdom (Zagat) with the edgier wisdom (Yelp) with my friends input gives me a pretty good triangulation and seldom fails me. If I listed to Yelp alone I’d mostly eat at the 25-year-old hangouts all the time.

Neither Zagat nor Yelp seem to do a good job of capturing what the new trendy, up-and-coming restaurants are, which is important for people who live in big urban cities where restaurants change like fashion styles. Luckily I have access to Kelly Amoroso in our office who is more knowledgeable about restaurants than anybody else I know. She’s the GRP food ninja.

And there you have it. UGC caters to the masses and isn’t “sliced” for us as individuals. I see a few companies at the edges building intelligence into their products including people like Lunch.com (share & discover opinions most relevant to you), Gogobot and Hunch (although I’ve never really asked Chris where he’s taking things longer term – maybe I should!).  I like his tagline a lot, “Hunch personalized the Internet.” That sounds right. If he can achieve that it will be big. Really big.

So either the future is a Lunch.com like sacking of websites like Trip Advisor (which given that he sold them his last travel venture for some serious bank sounds very plausible) or somebody developing technology to enable all of the existing products including Yelp, Zagat, RottenTomatoes – even Amazon.com – to be sliced better.

My Wish List:

I want at least three slices for any UGC sites.

1. Sliced by social graph – The obvious place to start is my social graph.  Sure, I’d like to know what my Facebook friends think about the wine, books, films, hotels, airlines, etc. that I’m thinking about using.  So I’d find this slice very handy. Necessary, but not sufficient.

A few years ago I planned an outing to Santa Barbara for my Dad’s 70th birthday. It was quite difficult to find the right place to stay. I needed somewhere nice enough that people wouldn’t complain about lumpy beds but not so nice that people would complain about the prices – it IS Santa Barbara after all. At the time I had young kids but my siblings didn’t. (now we’re in the reverse situation). So I needed a place baby friendly but not too family focused.

Ugh.

Our social networks are filled with our friends, sure. When we’re all 25, they’re all pretty much exactly like us.  The older you get, the more people you meet, the more diverse your friends become and even your lifelong friends diverge. My closest friend from childhood (and best man at my wedding) decided not to have kids. Believe me our hotel & travel preferences are different!

2. Sliced by influencer graph – The second slice I’d like to see is by the “influencer graph.” I’ve talked before about the power of Twitter for information discovery. If I want to know about any topic I can follow experts who like to Tweet & link to that topic and I expand my horizons. It’s why I like following people like Shervin Pishevar, for example. We’ve only met once but I love tracking what he’s into politically as well as his startup views. He’s one of those guys that everybody should follow. Beats only reading the NY Times everyday.

So what if I could know who were the “scouts” for all the hottest new restaurants in NY, SF or LA? It would be a sort of crowd-sourced Kelly Amoroso. What if I knew who wrote the best stuff for kids dining in cities? What if I knew who the biggest influencers were of independent cinema (for which I’m a voracious fan), the fashion guides for men’s clothing, the eco-travel wizards who span the globe with kids.  I’d love to know people to “follow” on websites to help expand my tastes the way that Twitter expands my news sources.

No, seriously. I’m stuck in a music rut. I still love the classic rock I grew up with, grunge from the 90′s and rap/hip hop from the 90′s and naughties (00′s). But I’m sooo out of date. I kept asking the guys who are like me but way more hip musically (like Ian Rogers) than I what to listen to. I need a better system of music discovery. But music IS probably the leading category at providing these types of services. We need them for every vertical.

3. Sliced by people like me – But mostly I want my recommendations sliced by random people I don’t know. That sounds counter-intuitive so let me explain. I want to know people who are just like me. They like similar food, they have similar politics, they’re in the same age & stage bracket, they have similar financial situation, watch the same movies, etc. In fact, this data set is probably totally different by vertical. I might love the same films as 50 people who for whatever reason don’t like the same hotels.

So my ask is topical slicing by people like me. Help me cut through the clutter. And I don’t want you all who are different from me to be bothered with my eclectic taste in films (I hate Hollywood blockbusters and I drag my wife to foreign films of shepherd herders in Iran or EVERY Mike Leigh film) or my poor taste in music.

I look forward to the death (or improvement) of Trip Advisor and all the other crappy UGC sites out there that have no intelligence or targeting whatsoever. I want Lunch or Hunch or whoever is going to solve this problem to hurry up and sack the incumbents and spill the blood of the outdated, crappy UGC Internet.

  • http://20minus.com Thanasis Polychronakis

    So, you want to follow separate people for each and every activity in your life? That would be perfect, but selecting who to follow on each and every site / activity would be a drawback for most of the users…

    How do you beat that?

  • http://twitter.com/L1AD LIAD

    Here here.

    Reading hotel reviews or even movie reviews for that matter – is akin to taking the road to hell.

    Forceful divergent opinions leads to action paralysis. Try using trip adviser to search for a hotel in NYC. With so many axis of differentiation and opinions – left me wanting to cancel my trip, crawl under my bed and cry.

    'Sliced by people like me' – also isn't the holy grail though. I know people/have friends who have comparable upbringing, education, political outlook etc as me – but their movie, music or dining tastes would translate to me into huge turnoffs rather than turnons.

  • http://kirstenwinkler.com KirstenWinkler

    Great food for thought, thanks Mark!

    How would you collect the data of people who are like you? Based on checkins, likes, subscriptions etc – hence with an open sourced API services can use to feed their data in?

  • http://bothsidesofthetable.com msuster

    1. have a way of discovering experts in each field
    2. be able to follow people broadly like me (across all categories)
    3. be able to refine by category if I choose to go that granular. I know that on Lunch.com you answer questions in a fun, game-like way about which films you like (think Netflix), restaurants, etc. and it matches you with people by category. You can then see why it matched them (you both liked Scarface & Deer Hunter) and you can purge people.

  • http://bothsidesofthetable.com msuster

    people like me is not = people who grew up like me or think like me. It's about preferences. I want to know people who have a similar taste in film and get reco's from them. I want to know people who are similar in hotels so when I go to NYC I can stay where they do. See my comments above to Thanasis about Lunch.com, for example.

  • http://bothsidesofthetable.com msuster

    Can be via quizzes, click on pictures of what you like (think Netflix) and/or via compiling information about likes, check-ins, similar. The algorithm is the intelligence that will make a couple of these players really stand out.

  • http://twitter.com/L1AD LIAD

    understood. it was just when you wrote “they have similar politics, they’re in the same age & stage bracket” – it stood out to me that those generalities don't dicate diddly squat about tastes and preferences.

  • http://20minus.com Thanasis Polychronakis

    I believe and know you can do that.

    However i also believe that at this point you have lost 90% of the internet users. In other words, i do not believe common users have the agility to do that, don't forget we are deep geeks and throve from staff like that – organizing the world around us…

    So if the majority of the users will not follow on this trend what is the incentive for the web companies to adjust?

    However pointy i may sound, i totally agree with the problem / issue you describe and believe that we are on the verge of flipping to the next level of the web, because of this exact issue.

    How it will be done and in what manner is what eludes me and i'm trying to debate it… Great post to do just that! Thank you :)

  • Stephen

    The obvious way is to look for people whose ratings of places correlates well with yours. This solves the problems of having to correctly pre-define rating categories and of people not really knowing what they want. If you worry about changing tastes, as mentioned in the original article, you can weight more recent reviews by the user more highly. The software could also see that your tastes correlate well with two different groups, and give a couple of different groups of suggestions (one group correlating to when you're traveling with the family, one to when you're traveling alone). The drawback is that you have to rate a fair amount of stuff to get a good correlation.

    But isn't this the same as what Netflix and Amazon do to make recommendations now? Am I missing something?

  • http://twitter.com/sgauss Stanley Gauss

    Mark.. that is actually what we are taking on as we build localhabit.com. Our algorithm is at a very early stage but we have figured out a way of connecting multiple social graphs into one, localizing it and slicing it up into location, interests (people like me) and habits. Keep an eye on our progress and as we mature, I am sure we will help users make smart decisions about things to do and places to go.

    The idea came from my own pain.. I have two young sons with Aspergers Syndrome and sometimes it is hard to find a place to take them that both stimulates them and is private and quiet at the same time. No matter how I researched I couldn't find recommendations and then gathered a list of 20,000 people like me with the same problem. I then found other segments with the same issues… Home Schoolers, Religious Groups, Single Parents and about 30 more.

  • http://kirstenwinkler.com KirstenWinkler

    I think the system should also take pure consumption into consideration. For example if I watch a certain series on TV every week I don't need to rate it every time to show my interest. It could also happen that I don't like the second season and stop watching but forget to “unlike” this show.

    The pure data would be more relevant here. Same if I stop visiting a restaurant because they changed etc.

  • http://hdemott.wordpress.com Harry DeMott

    Mark – I could agree with you more on this. I have very much the same frustrations and have written about them in comments on a number of blogs – as well as writing directly with Chris Dixon of Hunch to talk about it.

    What's interesting is that you learn to parse UGC sites pretty well over the years.

    For example – when I go on Trip Advisor, I know for certain that low reviews that are written by people in the UK are invariably problems with service. (Read through them, you'll see) People have an issue with breakfast not coming quickly enough, or a mistake on the bill, or their drinks not being refilled quickly enough in the Caribbean. All of these are valid complaints for the person writing – but not for me. I could care less about any of these – so I mentally parse them.

    When I lived in NYC with my wife (pre-kids) we ate our way through the Zagat guide year after year – and guess what, you can easily parse that guide as well – despite the lack of granular information (at least back then) Whet many people thought were great restaurants had great food, but didn't have great dining experiences. At 26, you don't want to eat with blue haired people who look as if they are having no fun. We drank a lot and joked out loud – and were probably mortifying to the patrons, but more than once the waiters told us that they wish they saw more folks like us in the restaurant.

    The question is how we fix the problem. Is it an overlay filter with check boxes that allows you to overlay and weight your social network, your key influencers, and others who have demonstrated tastes like you?

    Or is it a whole new system built to be social and malleable from the very start (I like this idea much better).

    The issue with any of these UGC review sites is that there is no way to really filter on what is important to you.

    For example – I went to Hawaii this summer with my two daughters (10 and 8). Where to stay? We wanted to be as close to the ocean as possible – okay that narrows it down, but Kona or the Kohala Highway – or do you go over to Hilo, or Puna? The sites aren't set up for that query. What's most important to us? Dining? Shopping? Fanciness of the hotel? Amenities? Proximity to Activities?

    In our case the key was staying near the best beach snorkeling spot on the island for kids. That made it eassy – and we chose the Outrigger Keahou Hotel. It is right on the ocean, and you can walk from your room to a awesome snorkeling spot with a large fringing reef – so there are almost never any waves or currents – filled with turtles and perfect for kids.

    Is that the hotel I would choose otherwise? No. It is really no better than a Best Western. No comfortable beds. No flat screen TV's. No great room service. No workout facilities. No great breakfast buffet. No plush towels. No fancy bathrooms with interesting toiletries.

    But it had the two key attributes I needed. Every night I went to sleep listening to the endless sounds of the ocean and woke to the same – grabbed my fins and mask and walked downstairs to head in.

    So how does any site deal with that?

    You need to be able to filter deeper against more constraints.

    Pandora works great – but if I want classic rock that I can work out to because it has 110 beats per minute or higher – I am SOL. Can't do it – even though the music genome has the information that would allow me to do it.

    The answer to your question is far more control – which makes the sites harder to navigate.

    Tough problem – but one well worth solving.

  • http://joshuadance.com/ Joshua Dance

    I like it. Often times, it is too much work to cobble together your friends info, the reviewers info, and actually make a decision. Slicing would be a great solution.

  • Stephen

    For things which are often repeated (TV, restaurants, maybe hotels, websites), this is a very good point. FourSquare probably does or could do stuff like this for restaurants. Taken to the privacy-destroying extreme, my phone or credit card could notice that I go to a certain set of restaurants or hotels regularly and recommend them to people similar to me. I wonder if there's a less creepy but similarly automatic way to gather the info…

  • http://twitter.com/_girishrao Girish Rao

    Slicing by verticals is what makes this challenging. Firstly, there needs to be a way to define “like” for each user. Maybe a user has up-voted a band, or maybe a user has checked in to a restaurant. Secondly, currently there is no one ubiquitous (well, maybe Hunch has this), magic data set that will find people like you in terms of food (restaurants) AND hotels AND music AND politics etc.

    Each of these verticals require specific user information tied to that vertical. An obvious one is checkins to similar places and defining similarity using the price/food type/time/neighborhood/etc — I believe 4sq will implement something like this for place recommendation (or collaborative filtering, both user A and I checkedin at Places X,Y,Z. User A also checked into Place ZZ, so recommend ZZ to me).

    Aggregating this data across categories as a one-stop service becomes trickier and this is what Hunch is trying to do, but how successful have they been? They have partnered with 3rd party websites to plug in their recommendation service, but many other (most?) websites prefer to keep their user data in-house and build their own recommendation engine.

  • http://kirstenwinkler.com KirstenWinkler

    Exactly, the Zuckerberg approach. You don't actively checkin anymore, your mobile device does it on its own.

    As you say, creepy but on the other hand lots of potential here as with the downward spiral of laziness even pulling out the iPhone and pushing the button can be too much.

  • http://bothsidesofthetable.com msuster

    I doubt we're at odds. More refinement of my thinking that didn't make the post but forms my in-person debates:
    - I want flexibility for advanced users to play with their personal lists
    - I want the masses to be auto-matched by algos that are “good enough” looking at inference data like FourSquare check-ins, Facebook likes, booking data from travel engines, etc.

  • http://bothsidesofthetable.com msuster

    True. Except when they do. They can form part of the intelligent algo – but not the whole. As I wrote above:
    - I want flexibility for advanced users to play with their personal lists
    - I want the masses to be auto-matched by algos that are “good enough” looking at inference data like FourSquare check-ins, Facebook likes, booking data from travel engines, etc.

  • http://bothsidesofthetable.com msuster

    Netflix does a good job. Amazon could do better, IMO. But I'm talking about matching intelligence for the rest of the web. Travel, food, movies, books, airlines, etc. Yes, I think it's a mathematical correlation problem.

  • http://twitter.com/_girishrao Girish Rao

    Facebook's Social plugins (like buttons/recommendations) also aim to achieve this — a ubiquitous graph across a variety of verticals. Greater distribution of their plugins will lead to a better quality graph.

    Might Facebook use this data to serve both personalized content and targeted sponsored contents/ads? Like in the FB NewsFeed, personal content and branded content are starting to blend and they could make this happen across the web.

  • http://bothsidesofthetable.com msuster

    Cool. I'll check it out. And I have 2 friends with Asperger's kids so I know what you're talking about. The best companies come out of solving personal problems. Good luck.

  • http://www.thegetdownguy.com willcole

    We're working on solving some of these problems with Know About It. Right now the focus is on content that flows through your social streams (links). We are recommending content and giving you views into your friends personal collections as well. The basic idea is what you mentioned above; there is so much content coming through these streams on facebook, twitter, tumblr etc… and we want you to find the good stuff.

    We are also working on some views now that will extract certain types of links like location sharing etc… that may solve some of the restaurant recommendations you were looking for in your social graph.

    We'd love for you to sign up for the alpha and let us know what you think.

    http://knowabout.it

    alpha code = alpha

  • http://twitter.com/LutzVA Lutz Villalba-Adorno

    How about http://www.google.com/hotpot ?
    Seems like the ideal solution if it takes up.

    I actually dont care to much, in my life of hostels, cafes, car sharing and whatever, try and luck is pretty interesting.

    And I didnt use HotPot yet, but I saw it on TWiS and the UI is great to review a lot of stuff with fun (you dont go into to much detail). I guess at some point they know how your taste fits with others and can push some interesting stuff and I remember that they also provide to check instantly Yelp and others to get a more detailed picture.

    And they of course provide you to follow “friends”.

  • http://twitter.com/revoline Woosung Ahn

    “Recommendation Slicing” would make a lot of sense, particularly if you can also add ethnicity filter. Great post. Thanks!

  • http://bothsidesofthetable.com msuster

    Thanks for long and nuanced reply. My answer? Yes.

  • http://bothsidesofthetable.com msuster

    I think that recommendation sites all need faceted search that allows us to slice & dice our views on different dimensions. I think that the data set by vertical matters. I think being able to get inference data (with no input from users) is key where possible b/c 90% won't input.

    Any partial solution would be better today. I'm going to dive deeper and learn about what Hunch is up to.

  • Juha Huttunen

    Great post and topic – getting the internet personalized to your liking is indeed one of the holy grails. You used a lot of travel examples so let's talk about that as it is something I know something about. First a disclosure: I'm the founder and CE O of a travel startup called TripSay (http://www.tripsay.com) which tries to solve exactly the problem (in travel only) you're talking about in exactly the way you described: allowing you to follow other travelers and trying to help you find the ones that are most like you so that you could find exactly the people whose advice you want to listen to. We use collaborative filtering to show how well you match with the other users and then you can read their stuff and follow them if you like. This leads into a personalized travel guide on each place you want to travel to which means that I get different recommendations on what places to see/eat/sleep in New York than someone with a different profile.

    Sounds nice enough doesn't it? And we're not by far the only site in this space in travel. I guess it's the same in many other verticals as well. Why is it then that TripAdvisor still rules travel even though many people are, for good reasons, dissatisfied with it? The simplest answer is SEO. When ever you want to find some content related to travel, you're bound to end up on TripAdvisor, Expedia etc. Travel is not something you do regularly. Thus people are bound to go to Google and do a search. That again means that they'll end up in the few biggest travel sites that have been out there for a decade and have tons of content. There are at least a dozen recommendation sites for travel out there but I guess few people have heard of them as people start the process at Google when the travel bug bites. This means that getting users is hard for new players making it hard to build a company. The other thing making personalization hard is the fact that to get something you need to give something. To make the recommendation engine work well you need to register and build your profile so that the service provider has a chance of knowing what you like and can provide you with good recommendations. Otherwise the “recommendations” are just the typical average stuff that you can find on any travel site. These two things make it hard in travel to create what you are asking for. I guess it's not far from truth in other verticals either.

    Now if you multiply that by the many different verticals this could and should be applied in you quickly get to the conclusion that it is simply too much work for the consumer – even though they might say they want it but when they need to act they don't have the time, remember the site name, etc. etc. Most don't want to sign up to single new site, let alone many. Most don't want to contribute making them hard to profile.

    Recommendations are working nicely for e.g. Amazon and Netflix. The way these work is not through users but through objects like books and movies. Those objects relate to buying/consuming the item in question which means that you automatically leave a trace of yourself in the system that system can then use to refine its recommendations. In many cases like in travel you might just travel a few times a year on a vacation and purchase the trip/hotel/whatever on the site with the cheapest prices which means that no one is able to build that kind of a profile of you as e.g. Amazon is able to do.

    So I guess you could say that the reason you don't have the “Pandora for everything else” is that it is so damn hard to build. :) But we and many others are trying… It's a tough nut to crack but great rewards await those who can solve it!

  • http://twitter.com/hongdquan Hong Quan

    Mark,

    Amazing analysis of the problem we're trying to solve with product paralysis for Parents. I've been thinking about a lot of the same solutions, just didn't have a fancy name for it. Mind if I borrow your terminology? We were just in Hawaii with two young children and our best recommendations came from a friend with kids in similar age range.

    Cheers,
    Hong

  • http://hdemott.wordpress.com Harry DeMott

    I like the nuance!

    Would love to catch up on the phone if you have some time. My office is 212-356-2918.

  • http://twitter.com/billmcneely Bill McNeely

    You should check out my friend Ryan Kuder's site called http://www.bizzy.com. The site allows you to get personalized recommendations for great local businesses from people with tastes like yours.

    When you first sign up you are asked about 20 questions in order to get a feel for your tastes.

    Then Bizzy compare how you answered those questions to others who answered in a similar fashion and produces dining recomendations

  • http://www.robinsonxd.com Barrie Robinson

    The challenge here is one of approach. Rather than trying to extract applicable data from what exists today, you need to think more so of activities that can be enabled that in-turn deliver a richer more 'sliceable' data set. Once you have this data, you can drive a more intelligent internet experience in a number of ways.

    We currently live with a toolbox web. There's a task overhead to this. It would be nice to see more focus on products that proactively service our needs intelligently, rather than tools we need to use to deliver our own results.

  • http://www.tastesavant.com Sonia K

    Hi Mark,
    Your insights are dead on. I'm actually working on a product that is trying to solve these EXACT problems. Stay tuned over the next few months for more information from http://www.tastesavant.com. And in fact, I'll be in California in a few weeks if you want to talk in person!

  • http://www.nektra.com Sebastian Wain

    Mark, I am working on something related in my spare time. I think an alpha may be ready in six months, it will be interesting to have feedback from you.

  • http://twitter.com/_girishrao Girish Rao

    Seems to me there are a couple use cases being discussed here. #1, in Mr. Harry DeMott's response above, the user actually knows what he wants very specifically. This, as you state, feels like a search problem. Selecting check boxes and radio buttons seems tedious. Maybe TripAdvisor could index their data/content more thoroughly and provide the user a more thorough search experience on the site.

    The second use case occurs when you don't know exactly what you're looking for, you need a recommendation. This type of content should be pushed to you based on inference data describing you and people like you.

    Hah, Wanderfly – a travel recommendation site – just came up on TC: http://techcrunch.com/2011/02/…/

  • Kevin G.

    Mark,

    This problem has already been solved, we're just in stealth mode right now for the final stages of our build. I've been following your blog for a while and was waiting for the right time to get in touch with you. Perhaps this is the right time? Would love to show you what we've got. How can we best get in touch?

    Kevin

  • http://twitter.com/danilod Danilo Durazzo

    Hi Mark! I'm glad to say that here at Mappyfriends are working on exactly that and, and what you described looks just like our roadmap!

    People want relevant results. Where can they get them?

    1) Your friends: you know them, you know how similar and how different they are from you. Gogobot and hotpot are doing interesting work on this but it's not deep enough.

    2) Experts who you might not know: Sometimes we don't realize how much we know about something. Any of us could be an expert on something and there is currently no platform that leverages that long tail of experts.

    3) Crowdsourcing: sites like tripadvisor and yelp did great work in the past but are getting old very fast. If they don't adapt quickly enough they will be replaced by other more innovative sites.

    We have a very strong vision of how we think people will make places-related decisions in the future and we are working to reflect that on Mappyfriends' upcoming versions.

    Feedback appreciated!

  • http://www.victusspiritus.com/ Mark Essel

    Defining relevance is a behavioral, social and algorithmic problem, and all are a function of time.

    1) Behavioral because we need to passively or actively encourage interested users to create a taste graph for areas of interest

    2) Social because recommendations from friends that are familiar with our styles and persona defy the limited knowledge of automated systems

    3) Algorithmic because our specific tastes in niche areas can be clustered to other people. While we're all unique, there are many (at least some) folks who share our style tastes in a narrow category

    A company may focus on all three areas by narrowing cluster size, or they may specialize in one area for a larger audience.

    I've jotted down notes on relevance a number of times, but it's not something I have an answer to now. What I do know is that the right solution will evolve quickly through adopted interfaces once it's proven, and it won't come from a single company.

    A taste protocol will provide any business or group (open source/side projects) with the means of organizing user generated content into the proper form to promote intelligent recommendations. Individual organizations can be the keeper of the most recent high quality taste clusters for hotel chains/by region, restaurants with a specific type of food, or the best mystery novels. They can compete on relevance and receive scoring quality from contributing users.

  • http://www.victusspiritus.com/ Mark Essel

    Loved that answer Harry, and it sounds like you found precisely what you needed based on your specifications (behavioral component I spoke to in my comment).

    “Or is it a whole new system built to be social and malleable from the very start (I like this idea much better)”

    I think we can use the existing structure of web apps and databases, but additional work must be done to refine the exchange of taste information (protocols/specifications). Jumping to an entirely new “web” is too drastic. There needs to be a path with stepping stones from A to B, considering browser vendors/information systems. Navigating that path, at least the macro milestones, and adapting it on the fly are what will separate the winners from the losers.

  • http://hdemott.wordpress.com Harry DeMott

    I don't want to recreate the internet etc…. just adapt the current sites.

    So if we take travel, how do you look at a Trip Advisor and make it better?

    One way would be to start with a pretty comprehensive list of hotels. You can get that off of Trip Advisor so a lot of that work is done.

    but then, how do you do a better job mapping them, seeing pictures of them – locating and scoring them relative to proximity to dining, activities, tourist sites, friends/relatives?

    How do you then further analyze the properties based on the hotel amenities: pools, spas, workout areas, dining opportunities, toiletries, 350 count sheets etc…

    Could be that all you care about is price? Fine. You should be able to sort on that.

    so you do end up with the Pandora or travel.

    The issue with the solution is that Pandora is built on a music genome – with over 400 separate properties quantified.

    You would need to do the same thing with hotels – which require actually visiting the hotel, seeing the layout, seeing the rooms etc… as well as plotting all sorts of other variables – some of which are quite subjective.

    That's what makes it hard.

    How do you do a test of the comfort of the beds other than sleeping in it?

    A lot harder than telling whether the song is in a major or minor key.

  • http://twitter.com/sixwing Rand Fitzpatrick

    A system to develop Interest and Expertise vectors that can be modulated by the location, domain and social graph they're being utilized in would be quite useful. I worked on a small scale version of that while at AT&T Interactive, and am not rolling a discovery and data curation tool to extend those principles. The team over at klout.com is also building up some excellent tooling on the infrastructure side of this problem.

  • philsugar

    Well written.

    I think the reason why travel gets written about so much is that there is a small subset of people (like me) that travel 100,000 miles a year. I don't wonder if we're a small use case that gets concentrated on because naturally most of us work in the tech space.

    We obviously spend a boatload of cash when we travel, and we have all sorts of different needs: what I want to eat or where I want to sleep will vary tremendously with whom I am eating or traveling

    You are right, when I Google I get back the big sites where I have some Yahoo telling me that sushi at Sagami's (best on the east coast) might as well come from the Supermarket.

    You are also right in that we (at least me) are lazy and selfish and just want to see the results and not contribute. I think Zagats has it the closest but they don't have the right categories.

  • http://twitter.com/sixwing Rand Fitzpatrick

    Going a bit further, I think it's important to leverage the networks and activity already out there, and being able to mine for explicit, implicit and emergent interests algorithmically, and then refining that with human-driven tuning or curation of the interest points and UGC.

  • http://tonepedia.com/blog Danny Strelitz

    Question is if the internet is ready for that?
    The web is an active medium unlike TV, we are actively engaged with the content, like I am posting a comment now. So if I am actively looking for content, is there real value in my social graph, or influencer graph?
    Don't get me wrong, I think the idea is great, but I don't think that the mess market out there is ready to consume data passively. But than again, I might be wrong.
    If there is a way to make a passive flow of data active on the user end?
    I loved facebook events, when I could see my friends events, until the noise was bigger than the value.
    So, how can you make it both accurate, with real value, active, and user excepted?
    Hope I am wrong :)

  • philsugar

    You know Elie Seidman (guy with the all time most popular post on A VC) is working on this at http://www.oyster.com

    I am in no way affiliated with Elie or his company other than to say his post about being just an entrepreneur pushing the ball up the hill inspired me to blog.

  • http://hdemott.wordpress.com Harry DeMott

    Thanks Phil.

    I do know Elie – and I've met him in person – looked at Oyster etc…

    I like what he is doing – and think it is a good start toward a much better and more complete system.

    Wasn't in the position to invest at the time – but hopefully that will change over time.

    Oyster has the chicken and egg problem. Good ideas but not enough content yet – and the content is expensive to acquire as he is doing a real thorough job vetting the hotels. If he can build it up – it will be great.

    BTW: what was the all time popular post on AVC?

  • http://twitter.com/starttowonder S Jain

    Mark, for “Sliced by people like me”, The companies will need data. And that data will have to be provided by you the user. The company that makes the user enter the data will be the winner.

    I had sent you an email sometime back on this topic(social comparison and social database). I think your today's post validates the need for that product.

    Thanks

  • http://swombat.com Daniel Tenner

    You might be interested in checking out (and who knows, perhaps even investing in!) a London-based startup called Rummble. They do exactly what you describe – i.e. match you with recommendations from people who like the stuff you like.

  • http://twitter.com/starttowonder S Jain

    Also, I think the best placed company to do this currently other than Facebook is twitter. I don't know if they are doing this right now though. For any start up looking to solve the above problem should also target twitter to get the masses. Once you start getting the masses people will be more willing to share more data….but yeah add amazingly customizable privacy settings.

  • http://mileslennon.com Miles Lennon

    A couple of comments:

    1) Interestingly, you can seg this topic into the “one graph to rule them all” topic. I'm not sure there is going to be one graph to provide all of these slices. Take restaurants for example:

    Let's say the whole experience started on Yelp. Your “Taste Neighbors” graph (citation: Fred Wilson) on Foursquare will provide your people like me slice. Then if you could take your Food following list from Twitter to find Yelp reviewers with “influence” you could get your influencer slice. And if the site had Facebook connect you could get your social graph slice. But all of these separate graphs will have to be very interoperable to be delivered in the same place the right way.

    2) You started your essay talking about Pandora. Please feel free to disagree, but I believe the massive uptake on Pandora wasn't as a result of taste neighbor slices, collaborative filtering, or machine learning (are they even doing much machine learning for radio stations?). Part of what's beautiful about a Pandora radio station is how the music is editorially sewn together. By editorially, I mean with rules. The product worked well with ONE slice – the music genome. The genome for my food intake is largely NYMag. I enjoy yielding to their editorial control just like I enjoy yielding to the editorial control of Pandora's musicologists.

    I think that awesome influencer/editorial slices can get us 80% of the way there. Maybe it's the other slices that will provide the next 20.

  • http://mileslennon.com Miles Lennon

    “To make the recommendation engine work well you need to register and build your profile so that the service provider has a chance of knowing what you like and can provide you with good recommendations. Otherwise the “recommendations” are just the typical average stuff that you can find on any travel site. These two things make it hard in travel to create what you are asking for. I guess it's not far from truth in other verticals either.”

    Nice point. However, I think that if Travel is a multiple times per year event (as opposed to Netflix, which is multiple times per month/day), then users might be more amenable to giving TripAdvisor information to filter results and reviews. For example, I'm not sure it's a long shot for Mark to enter the following information:

    1. I have kids
    2. I don't like chain hotels
    3. I've been to Orange County (California) before

    Not sure if this meta-data could be IMMEDIATELY used to users' advantage. But at scale, with enough participants, it would be enough to even flag the review from the user who clearly did not have kids with them.