I understand the quest for a larger userbase but please don't use this sort of FUD that is being peddled by those who don't understand the technical aspects of search and are trying to sell their books.
A search engine's prime function is to filter the millions of results for each query down to the most relevant results for each individual user, and never the same 10 results are relevant to each and every user.
There is little difference between personalization and the relevance of search results.
How would you go about ranking then? alphabetically?! it's a matter of tuning the relevance 'dials' and it's all in early stages so a solution to this imaginary problem is more research and not to hide behind bullshit terminology.
And as a bonus a user (regardless of any personalization) can dig through any initial set of results if she seeks more information.
So please don’t buy into this misleading and ignorant bullshit that is being disguised as a real issue.
Search queries have different purposes. Personalization based on my preferences and past search works wonders for known-item seeking and for re-finding.
But for exploratory search and exhaustive search personalization works as an echo chamber (or filter bubble if you will) .
Then there is the problem of change and inertia. People change, their preferences vary in time. Personalization has an inertia that causes the recommendation engine to always be behind the actual preferences. It's more visible the better the recommendation engine.
I'm not saying personalization is problematic. There are problems with using it all the time and stealthily, without a clear possibility to turn it off.
All can be remedied by tweaking the algorithms and that happens all the time, you can also use a certain vertical directly for certain type queries (scholar search etc).
The same way that what a user might enjoy can be inferred the inverse of that can be used as a signal in the algorithm as well. Personalization is just relevance and all is just a matter of tuning.
I agree, it can be remedied by better algorithms - such that can recognize when I'm doing a search for a known item and when I'm performing exploratory search. In the first case, personalize away, in the second case, please show me everything.
The challenge is that people switch very quickly between these search modes and don't even think about it.
"There is little difference between personalization and the relevance of search results."
Even if you're a Sailor Moon fan, you might actually be looking for information about the rock rotating around this planet when searching for "moon". In the same vein I might be a liberal but be looking for a multitude of opposing views and ideas when googling "core values of NRA members".
Oh btw, we engineers, our algorithms, aren't smart enough to foretell people's inclinations based on their past habits, we also can't read minds. The increasing use of personalization as a factor for relevancy will, imho, lead to user dissatisfaction, precisely because they make shoddy assumptions.
Believe it or not, but some people also want to be surprised and learn new things. YES. We exist!
The article makes the assumption that personalized search results are bad for you, then goes on exemplifying it, but does not say or demonstrate in any way why personalized search results are bad, especially since it doesn't give any context about the person making those queries.
Search ranking is a matter of context. "Egypt" does not mean anything, other than the name of a country, and different people mean different things. When you're searching for "Egypt" and want to get travel tips, if you don't get them you can always expand your query to "Egypt travelling tips". But even that is a pretty shallow request and you can further expand it like "Egypt travel guide to the pyramids".
The search engine's job is to find what I'm looking for. And IMHO the current state of the art is a little behind my expectations - I would have expected these personalized results to be far more effective than they are by now.
It's true that sometimes I'm looking for specific things, but other times I'm searching to get an idea of "what's out there", to fill gaps in my knowledge and make sure I'm appraised of what's going on in the world. It'd be nice to at least have some control over what kinds of searches I'm doing. For example, if I search for my own name, what I want to know is something like: what are the most common search results other people get when they search for my name. The same is usually true if I'm doing "related work" type searching for research; I explicitly want to find stuff outside my immediate area of specialization, in order to make sure I'm not missing anything that other people would've expected me to cover.
I understand that, but if the search results are not personalized for you, then they must be personalized for the common denominator, as in the hive, the majority's opinion, the status quo.
For general search queries, the long tail goes out the window anyway, and I haven't done or seen any quality metrics for DDG, but I doubt their search results are better for exploration or getting opinions different from your own.
I agree. I think people will surround themselves with a filter bubble with or without personalized search results. Ever tried to convince an avid Fox or CNN reader to read the opposing view's site?
From my personal experience, people rarely search for something as simple as "Egypt," and instead, search for "Egypt travelling tips" like your example.
If your Google-fu is strong, you'll write a query that targets your desired results much better than a generic shot-in-the-dark query.
I expect my search to be faithful to the terms I enter. If the results aren't satisfactory, I'll narrow the search with additional terms. Withholding information based on some guess at my political leanings is dishonest. I'd be appalled if a librarian did it; why should I expect less from a search engine?
If you went to a librarian and said "show me Egypt", she would have nowhere to start. A conversation would then ensue about what you are interested in.
Keep going to the library, keep having those conversations, and eventually, the librarian will take you straight to the books you want when you ask for Norway. If you want something different, you'll have to explicitly ask.
What Google is doing is no different. It just happens to have a lot more conversations with you.
What Google is doing is no different only if you can, in your ensuing "conversations," correct a misinterpretation that may have happened somewhere along the way. If your ability to do that is hampered because the only way to do it is by clicking on something, you don't see the thing to click on which will do this, then the filtering is not helping (and actually works against Google's and my mutual interests).
Not sure where you are going with your dichotomy- if you are unable to correct in either case (because you don't get the option of clicking on a correct answer) then it fails. I suppose with enough perseverance you could trick the search engine into showing you what you want- if you know it exists in the first place- but then, (1) just enter the very specific term to start with, and (2) this defeats the attempts to become more relevant.
I agree with you that these seem to be small changes. So far.
That's a good point, but I think it kind of misses the forest for the trees. It's true that personalization might sometimes lead Google to mistakenly show you less relevant results, but it seems pretty certain that the opposite would also be true — a generic page will sometimes show less relevant results than a personalized one. One search result is going to be shown ahead of another either way. If showing a less relevant result ahead of a more relevant one is considered "hampering," I don't see any reason this "hampering" is unique to personalized search.
I think it would come down to a matter of degree. A slight change in the order of results is pretty meaningless either way. But major alterations, where some pages are relatively inaccessible, could obviously be an impediment to getting the results we want.
Let me try this example: Suppose your searches are primarily academic. Let's say that whenever you search the term "momentum" you are looking for something scientific- ballistics, elementary particles, whatever. But one day you are writing a blog in which you want to search for background, but you need to use a non-scientific meaning of the same word. Perhaps a psychic used the term and you are debunking. The particular case is irrelevant. The point is tha if personalization is too aggressive you may not find the info most relevant to your interests.
This isn't privacy FUD or anything, just a pragmatic warning. The FUD comes when you start thinking about how their algorithms might actually decide which results you are "interested" in. Thinking of Google, what bu siness are they in? What about you behavior on the web most interests them? Would they decide that the most interesting things to you are the ones on which you clicked the most ads?
But whatever change personalized search makes relative to generic search, the opposite change will occur going from personalized to generic. To penalize personalized search when the change is of equal magnitude going either direction doesn't seem fair. As long as the personalizations are just a transformation and not a subtraction, the two options are just mirror images of each other. They have the exact same kind of failure condition. The question is just which one's failure conditions are more likely to occur.
As a counter-example: Suppose your searches are primarily academic. Let's say that whenever you search the term "momentum" you are looking for something scientific- ballistics, elementary particles, whatever. But most people aren't looking for scientific info, so you constantly have to dig and dig to find anything relevant on Google. The point is, if the search technology is too impersonal, you may not find the info most relevant to your interests.
As long as the personalizations are just a transformation and not a subtraction
This is the crucial point. As long as it's accessible, I can get what I want by search refinement either way (but I would imagine that it would be easier for me to refine in the space in which I am familiar than to refine in the unfamiliar space). Giving me the (easy) option to turn it off is a good idea for search.
What does faithful mean? You want German or Chinese results? You want global or local results (in your language of choice). There is always personalization in search.
The 'political leaning results' are just a tool the author used to try to explained his (ill-informed) point, there are no evidence to any such degree of personalization specially not to any 'bubble' inducing extent, it's all very sensationalistic.
And for the record the political example was about facebook wall.
I think you missed the point. Yes DDG ranks, but it may rank differently. Of course, there is an implicit slam on them here: instead of showing you what you want the vast majority of the time, they show you other stuff. This is good on occasion (so you don't live in an echo chamber), but if this was all the time it would mean lower quality search results for simple things.
This is exactly why I stopped using DDG as my main search engine. My workflow was often like this:
- Search for term on DDG
- Look at results, find nothing
- Try to find better term
- Give up and use original term in Google
- Find result
I tried to use DDG exclusively two time (once 8 months ago, once ~4 months ago), but the result was the same. I don't know how much personalization affects this, but Google just gives me the best results.
My own standing is that most of the time you need what you have to have to complete task at hand. Then you could go at YouTube and refresh "Recomendations" to your heart content.
The other way around is to wreak havoc to important parts of your life.
Why do you think it is an issue? I think everyone already filters intuitively. It's called having a "bullshit filter". I believe it makes not much difference whether it's automated or everyone does it manually.
I think the chance that I would click on a bullshit source like Fox News is very low because I don't trust them. Even if Google brought it up as the first result.
My point is that people already only click on search results that they agree with. Nobody wants to see stuff that they disagree with. That is human nature.
So living in a filter bubble already happens without automatic filters. It happens in our brains and we call it personality.
It's not an issue just because you want to believe it is.
Media outlets always filtered stuff for you but now the algorithms are 'filtering' results according to your interaction to find what's relevant to you, so you have a 'vote' here.
You also can dig through any initial results (click the news vertical to find more news for example) you couldn't do that with the media of yore.
Maybe we don't want a filter that sucks, plain and simple? Whether the filter sucks or not is completely subjective, but it sucks for me, so there.
Changing the language not that intuitive in Google, it's a drag. I have to do it all the time when I clean my cookies, and sometimes I still get crappy "pt.wikipedia.org" results on top instead of "en.wikipedia.org" just because my browser isn't in english and I don't want to change it just because of Google.
I might want to type www.google.it and search for Berlusconi news in Italia, but it won't let me, because it just show me brazilian news. So yeah, I can't even get to another bubble.
Also, apparently Google filters by IP, since searching via Tor or a VPS renders very different results. I dislike this for personal reasons because it breaks the internet for me.
Also, I might not like the fact that Google collects stuff about my search habits, which is also a valid concern.
So, there, it's an issue for me. Might be an edge case, might be an exception, but I think that I deserve to know why this happens and the alternatives.
"There is little difference between personalization and the relevance of search results. How would you go about ranking then? alphabetically?!"
Oh, you know, maybe using Google's much vaunted "PageRank" algorithm, which is supposed to take in to account things such as links on other pages back to the page being considered for inclusion in the search results.. the more such links there are, the better the rank is supposed to be.
The above description is obviously an oversimplification, and I don't have access to Google's PageRank algorithm anyway, so couldn't tell you what it actually was if I wanted to, but it's something along those lines, and need not take in to account your previous search history in any way or "personalize" the search for you.
That agrees with what I said. Unless you don't think that much growth in searches over such a short period is significant.
It's even more significant when you look at where it was a year ago. 1.2 million in April 2010, 5.9 million in April 2011, 6.3 last month, and already 3.8 a little over halfway through this month. This is significant when you consider the lack of marketing and that DDG is run by one person.
It's not significant. In the world of search engines, you need to be the #1 or #2 player to earn major revenue. People aren't gonna wanna waste any time/energy advertising on your platform if you're getting just a handful of searches per month for say... mp3 players or cellphones. DDG is just betting on an acquisition, but it's just a wrapper around the Yahoo BOSS API.
Nothing starts at #1. Expecting DDG to displace an oligopoly built over a decade and a half in a year or two of existence is asking too much of anyone.
I understand the quest for a larger userbase but please don't use this sort of FUD that is being peddled by those who don't understand the technical aspects of search and are trying to sell their books.
A search engine's prime function is to filter the millions of results for each query down to the most relevant results for each individual user, and never the same 10 results are relevant to each and every user.
There is little difference between personalization and the relevance of search results.
How would you go about ranking then? alphabetically?! it's a matter of tuning the relevance 'dials' and it's all in early stages so a solution to this imaginary problem is more research and not to hide behind bullshit terminology.
And as a bonus a user (regardless of any personalization) can dig through any initial set of results if she seeks more information.
So please don’t buy into this misleading and ignorant bullshit that is being disguised as a real issue.