5 ms·

So you think Google knows if a piece of content is instructional? Or know that it is a Review? (not just has a rating) And you have to have all the words or a

by drakaal 14y ago

So you think Google knows if a piece of content is instructional? Or know that it is a Review? (not just has a rating) And you have to have all the words or a synonym of the words. But more importantly Google doesn't know what kind of content something is. Or what questions the content answers. Our system knows "this document answers what do aardvarks eat"

gibybo 14y ago
>So you think Google knows if a piece of content is instructional? Yes. >Or know that it is a Review? (not just has a rating) Yes. >But more importantly Google doesn't know what kind of content something is. You don't think Google knows how to classify content?
- drakaal 14y ago
  Then you think wrong. Google can't tell that something is instructions. That's not something they do. It may know that ehow has a lot of pages with How To on them, but it doesn't know if those are pages with step by steps for how to do something, nor does it know that Rotten Tomatoes has pages with Opinions and Points, and conclusions that make up a review.
  - joedevon 14y ago
    Well their move to semantic technologies e.g. Schema.org is a step in the direction of understanding these things better. Like if you mark up the page with http://schema.org/Review http://schema.org/Review wouldn't you agree that they know it's a review? Are you using semantic web btw?
    - drakaal 14y ago
      No. They know you tagged it a review. I could tag it review and have it be a sales copy page for a product. Our stuff knows that a review expresses an opinion, backs that up with facts, and makes a conclusion.
      - joedevon 14y ago
        And are you using Semantics? Going to Semtechbiz in June? If so, I'll buy you a beer ;)
  - pilooch 14y ago
    For the sake of realism only: any basic supervised machine learning algorithm does that with proper labeled dataset. I have built many. I can assure you this is so classic that Google has it, that many other companies have it. And there are much better and accuratd solutions in place for combining all signals of this kind.
danso 14y ago
I will happily concede that you and your colleagues know more about search than I will ever know, and so I find it strange that we're having this argument...your perception of Google's limitations seems so far off that it's as if you've mistakenly referred to them instead of AltaVista. To answer your questions, yes and yes, Google can derive the meaning of my search without relying on literal interpretation of the search terms. In fact, Google can return what I want even if I deliberately spell every word in the query incorrectly: "couk besr ribz" instead of "cook best ribs" brings up recipes of how to do good ribs: https://www.google.com/search?q=couk+besr+ribz https://www.google.com/search?q=couk+besr+ribz I'm willing to take you at your word, that there is a better way to interpret context and meaning from a search query than however Google does it...but if that is the entire raison d'etre for your search engine, can you come up with at least a few case examples where this is the case? The "cupcakes" you've contrived is clearly hypothetical (and not at all close to what happens in reality), and the ones that I've tried don't seem to show any improvement on human-friendly results. Which is not to say that samuru is a bad product...what you claim to do is incredibly difficult and is exactly the feature that makes Google such a useful, ubiquitous engine...I would love to be surprised but I'm skeptical that a new engine with a fraction of Google's processing power, nevermind the resources for test engineering and algorithm design, can compete with Google here...this isn't a "well, why hasn't anyone ranked search queries in this way before?" in the same way that PageRank/BackRub was 15 years ago...Search engines have been analyzing queries for intent, and their shortcomings in this area are due to it being a very hard problem, and not for lack of desire.
- DanBC 14y ago
  Your Google link gives me "Searching for couk best ribs". The 10 results give me: 2 hits for boat related stuff, with no food. 4 hits for restaurants, with no recipes. 1 hit about cervical ribs. And then, in positions 5, 8, and 9 there are recipes. So, even with all the smart people and computing power at Google this is still a very hard problem. It's great to have different people trying different approaches.
  - danso 14y ago
    That's strange, I get links from amazingribs.com and food.com, which is more or less what I expect http://www.amazingribs.com/recipes/porknography/best_BBQ_ribs_ever.html http://www.amazingribs.com/recipes/porknography/best_BBQ_rib... Either way, it's a deliberately contrived example to show that Google was more sophisticated than the GP indicates...I have no comment on whether amazingribs.com really does have the best ribs preparation tips
hexagonc 14y ago
But more importantly Google doesn't know what kind of content something is. Or what questions the content answers. I'm not sure I agree with this. Google may not know it explicitly but it can effectively know it, perhaps in some cases even better than the average human would if given the same search keys. It is the difference between knowing what a word means by the way people actually use it versus by looking it up in the dictionary. Google essentially crowdsources its results, taking advantage of the fact that people entering similar queries probably have similar intentions. If your users are only using keywords, with little regard for word order, I don't see how you can do better than Google on average. You may do better for obscure queries where the best search result can not be easily inferred from the keywords present, but how pages are like this? Furthermore, if there are a set of keywords where semantic analysis suggests page A is the best result but the page that most people actually click on is page B, which do you return first? What if your index doesn't even have B? This will be challenge you will face when trying to do better than Google on average. Nevertheless, I applaud your work and will definitely keep Samuru at the ready for the queries that Google struggles on.
- drakaal 14y ago
  Who said we don't use word order? Who said we don't use similar search to infer intent? We can't really use the click throughs until we have users. We need feed back to improve results. But this also powers the related search in our TLDR Products ( http://www.tldrstuff.com http://www.tldrstuff.com ) and those have to work with much more abstract queries, because they aren't user queries they are generated queries.
  - hexagonc 14y ago
    I'm not saying that you don't use word order. I think you must to some extent if you are doing a comprehensive semantic analysis of the keywords and websites. I'm saying that your users may not use correct word order because they are lazy and because they are used to the behavior of search engines that don't require it but still give okay results. Perhaps you have solved this problem, but I just don't see how you can offer better results than Google on average (without having a similar sized index) when users are just throwing together a bunch of words related to what they are looking for. It seems to me that if we want to take advantage of systems like yours for search and if we want to get better results than Google, we need to change users' behavior; they need to learn to give more precise queries.

by drakaal 14y ago

gibybo 14y ago

>So you think Google knows if a piece of content is instructional? Yes. >Or know that it is a Review? (not just has a rating) Yes. >But more importantly Google doesn't know what kind of content something is. You don't think Google knows how to classify content?

drakaal 14y ago
Then you think wrong. Google can't tell that something is instructions. That's not something they do. It may know that ehow has a lot of pages with How To on them, but it doesn't know if those are pages with step by steps for how to do something, nor does it know that Rotten Tomatoes has pages with Opinions and Points, and conclusions that make up a review.
- joedevon 14y ago
  Well their move to semantic technologies e.g. Schema.org is a step in the direction of understanding these things better. Like if you mark up the page with http://schema.org/Review http://schema.org/Review wouldn't you agree that they know it's a review? Are you using semantic web btw?
  - drakaal 14y ago
    No. They know you tagged it a review. I could tag it review and have it be a sales copy page for a product. Our stuff knows that a review expresses an opinion, backs that up with facts, and makes a conclusion.
    - joedevon 14y ago
      And are you using Semantics? Going to Semtechbiz in June? If so, I'll buy you a beer ;)
- pilooch 14y ago
  For the sake of realism only: any basic supervised machine learning algorithm does that with proper labeled dataset. I have built many. I can assure you this is so classic that Google has it, that many other companies have it. And there are much better and accuratd solutions in place for combining all signals of this kind.

danso 14y ago

I will happily concede that you and your colleagues know more about search than I will ever know, and so I find it strange that we're having this argument...your perception of Google's limitations seems so far off that it's as if you've mistakenly referred to them instead of AltaVista. To answer your questions, yes and yes, Google can derive the meaning of my search without relying on literal interpretation of the search terms. In fact, Google can return what I want even if I deliberately spell every word in the query incorrectly: "couk besr ribz" instead of "cook best ribs" brings up recipes of how to do good ribs: https://www.google.com/search?q=couk+besr+ribz https://www.google.com/search?q=couk+besr+ribz I'm willing to take you at your word, that there is a better way to interpret context and meaning from a search query than however Google does it...but if that is the entire raison d'etre for your search engine, can you come up with at least a few case examples where this is the case? The "cupcakes" you've contrived is clearly hypothetical (and not at all close to what happens in reality), and the ones that I've tried don't seem to show any improvement on human-friendly results. Which is not to say that samuru is a bad product...what you claim to do is incredibly difficult and is exactly the feature that makes Google such a useful, ubiquitous engine...I would love to be surprised but I'm skeptical that a new engine with a fraction of Google's processing power, nevermind the resources for test engineering and algorithm design, can compete with Google here...this isn't a "well, why hasn't anyone ranked search queries in this way before?" in the same way that PageRank/BackRub was 15 years ago...Search engines have been analyzing queries for intent, and their shortcomings in this area are due to it being a very hard problem, and not for lack of desire.

DanBC 14y ago
Your Google link gives me "Searching for couk best ribs". The 10 results give me: 2 hits for boat related stuff, with no food. 4 hits for restaurants, with no recipes. 1 hit about cervical ribs. And then, in positions 5, 8, and 9 there are recipes. So, even with all the smart people and computing power at Google this is still a very hard problem. It's great to have different people trying different approaches.
- danso 14y ago
  That's strange, I get links from amazingribs.com and food.com, which is more or less what I expect http://www.amazingribs.com/recipes/porknography/best_BBQ_ribs_ever.html http://www.amazingribs.com/recipes/porknography/best_BBQ_rib... Either way, it's a deliberately contrived example to show that Google was more sophisticated than the GP indicates...I have no comment on whether amazingribs.com really does have the best ribs preparation tips

hexagonc 14y ago

But more importantly Google doesn't know what kind of content something is. Or what questions the content answers. I'm not sure I agree with this. Google may not know it explicitly but it can effectively know it, perhaps in some cases even better than the average human would if given the same search keys. It is the difference between knowing what a word means by the way people actually use it versus by looking it up in the dictionary. Google essentially crowdsources its results, taking advantage of the fact that people entering similar queries probably have similar intentions. If your users are only using keywords, with little regard for word order, I don't see how you can do better than Google on average. You may do better for obscure queries where the best search result can not be easily inferred from the keywords present, but how pages are like this? Furthermore, if there are a set of keywords where semantic analysis suggests page A is the best result but the page that most people actually click on is page B, which do you return first? What if your index doesn't even have B? This will be challenge you will face when trying to do better than Google on average. Nevertheless, I applaud your work and will definitely keep Samuru at the ready for the queries that Google struggles on.

drakaal 14y ago
Who said we don't use word order? Who said we don't use similar search to infer intent? We can't really use the click throughs until we have users. We need feed back to improve results. But this also powers the related search in our TLDR Products ( http://www.tldrstuff.com http://www.tldrstuff.com ) and those have to work with much more abstract queries, because they aren't user queries they are generated queries.
- hexagonc 14y ago
  I'm not saying that you don't use word order. I think you must to some extent if you are doing a comprehensive semantic analysis of the keywords and websites. I'm saying that your users may not use correct word order because they are lazy and because they are used to the behavior of search engines that don't require it but still give okay results. Perhaps you have solved this problem, but I just don't see how you can offer better results than Google on average (without having a similar sized index) when users are just throwing together a bunch of words related to what they are looking for. It seems to me that if we want to take advantage of systems like yours for search and if we want to get better results than Google, we need to change users' behavior; they need to learn to give more precise queries.