DEV Community

Sologub
Sologub

Posted on

Google annoys me insanely.

Google annoys me insanely.

When I search for information, I want to get not 14,900,000 links found on the Internet pages, among which, probably, there is information that I need.

I need an answer.

Clear, understandable and structured. It would be great if it also fits into the context of my previous search and matched my needs based on my previous actions.

It's very simple! If I go to a search engine and type β€œMask” - before that I did not look for the address of the nearest one (which excludes the possibility of coronavirus), I am not fond of cosmetics and there are no appointments in my calendar for the next three hours (and indeed it’s the evening) ... Hence I am looking for a movie with Jim Carey.

The situation I described is very primitive - but I think it shows where I'm going.

We need to (1) avoid providing information in the form of a large number of links and (2) make tools friends with each other at the system level, not interfaces.

Let's start doing it!

Discussion (7)

Collapse
joelbonetr profile image
JoelBonetR • Edited on

It's a double edge sword. What about looking for a mask for coronavirus and being spammed with the movie, thematic forums, merchandising and fandom posts about it just because you searched something about Jim Carrey before?
Then you could fall on a non-return situation where you can't get off.

Google uses your recent searches and what it knows about you for offering content to you but also uses social context and searches from people near you or with similar not-nominal features, also it trains itself with user interaction - it means if a high percentile of users who searched the same (or similar) got the result on the website A, it will offer you this website A because it's probable that you find what you want-, moreover it score websites with other metrics that leads you to a better overall user experience.

I think they did a good work polishing the search engine and what you want to achieve is provided by a conversation with a "virtual assistant" that has been trained with NLP (natural language processing) that is a branch of machine learning, these are two different tools and two different concepts that solves two different needs.

Also the big G is working on that to implement it along Google Assistant, which is a better place to use it, and always you need a way to stop the current conversation and start again from zero for not falling into the death spiral mentioned above πŸ˜‚

Let's unravel the two situations for further details:

NLP Assistant vs Search Engine

NLP Assistant

Imagine you are bored at home and you linked your Google Assistant to your TV (it must be a requirement huh?)

You ask your Google Assistant something like "How old is Jim Carrey?"
It will tell you "58 years"
then you remember the movie and ask "I want to watch Mask"

with that data it should be able to point you to the movie:

  1. because you asked to watch something.
  2. because you both were talking about Jim Carrey on THIS conversation.

the workaround is something similar to this simplified example:
It first gets the training:

Then when using it (also training at this point, in fact it's always training):

  1. keep current conversation data.
  2. figure out if the current question or statement has something to be with the latest one, if not search for the one before and so on.
  3. Use the training results it already has to figure out which answer to your question/statement is more "true" (actually it's a value between 0 and 1, - rarely it's 1- so the movie will be scored with something like 0.9 true)
  4. offer the answer to the user.
  5. If user asks again almost the same keep this information for training and recalc the conversation to give a more accurate answer.

Search Engine

On the other hand, when you search on Google Search you must act like a good writer: adding the necessary details for it to figure out what do you mean but not too much for it to being verbose (boring).

If you search "Mask" obviously it will show you masks, because by probability it's more "true" that you want a mask than anything else, because percentiles never lie and it may be a 99% of people that searched for masks to protect against covid or for a peeling mask to deal with the irritation produced by the first ones πŸ˜†

If you search "The Mask" it's highly probable that you will be pointed to the movie (because actually this is the name of the movie, you can't expect searching for "Lambs" and crying because you wanted the movie "The Silence of the Lambs").

If it's not the case, of course you can add "movie" to your search and that's all.

Take a look at the official user documentation if you are sometimes in trouble for getting your answer. If you feel this is a bit dumb, take a look at the "Expert search tips" you'll find at the bottom that will help you for sure!

Hope it helps to understand the workaround and the difference of this two technologies :D

Collapse
sologub profile image
Sologub Author

Very detailed analysis. But what if you try to look towards this tool here: wolframalpha.com/input/?i=Marvin+G...

Collapse
joelbonetr profile image
JoelBonetR • Edited on

Didn't know about, just tried to search "mask" on it. Still needs to clarify that it's a movie, otherwise you'll be fulfilled with information about what "mask" means and historical data.
Also useless if you want to buy a mask.

I assume this is a tool with pre-defined queries over an elastic search or similar where your clarifications add a string into it, some way of categorization. It could be useful when searching for concepts but that's all, have you find any utility to it? Please tell us what it solves and in which situations it could be useful vs other tools

Collapse
ender_minyard profile image
ender minyard

I tried creating an answer engine (as opposed to search engine) based on this idea, but I wasn't knowledgeable about web scraping. I hope someone does make an answer engine that solves this problem. I really like answer.js but the programmer stopped working on it.

Side note: when I tried to make my own search engine, I finally understood that Google's biggest strength is its infrastructure, not the code that runs on its infrastructure.

Collapse
sologub profile image
Sologub Author

Can you show what you tried to create?

Collapse
ender_minyard profile image
ender minyard

I really love the app Peach, which detects keywords as you type. My idea was to use natural language processing (and probably a recurrent neural network) to detect whenever someone is asking a question while typing, and then show a button to them that answers that question in one paragraph with one single answer.

Of course when it comes to your example, like "mask," I would be more scared to code for that because I'm not sure if the user means the movie or the object. I also had the idea of using sentiment analysis to aggregate responses to create that single answer, but web crawling proved difficult.

search

Thread Thread
sologub profile image
Sologub Author

That's the fun: you don't need to scan the entire Internet. You only need to crawl sites with relevant information. See how Wolfram (wolframalpha.com/input/?i=Marvin+G...) is doing on his search network.