You would like to build a road on the moon made out of hemp oil.
"How to build a road, on the moon; made out of hemp oil" you type into a search engine.
The parser searches and finds that verbs 'build' and 'made out of' determine the request, And that 'road on the moon' and 'hemp oil' are each a desired end of the desired search.
'Build' toggles an algorithm that searches on key words such as 'construct', 'mix', 'blend', 'hammer', etc, basically words or concepts that A ( … An) transmogrifies into B( … Bn) via construction method.
'Hemp oil' is recursively searched and all the chemical derivations are returned, in a dependency chain, including manufacturing methods. At failure or a preset iteration limit, the search returns.
'How to build a road' is recursively searched as well, and all chemical and constructive derivations of roads are delivered, including tar-seal roads and wicker roads, but excluding metal roads. If aggregate is required, all forms of aggregate are then searched.
At intersection (depending on the speed of searching, the intersection can be increased until more valid results return) thus shews a map more or less of what is required to get from A to B.
All items in the found linked dependency are thence parsed via a multidimensional A* algorithm that weights manufacturing methods and material costs, and a list of results are then shown, complete with a fully linked 'process' one has to follow in order to get from A to B.
The algorithm that understands what begets what would be evolving and of course not perfect but even simple keywords can reveal these answers (i.e. 'extracted from'). The ultimate joining of the double-ended search would of course perfect the answer and discard the most useless results.
Results are then shown, with high, cheap matches first, and more complex matches second (i.e. a gravel path strewn with hemp rushes (which contain oil) would be #1, a hemp-oil bitumen based road with glycol and xantham gum additives would be #2, etc)
The A* algorithm can then be run again on the results, to return all the things, and all the costs and manufacturing processes involved, to reveal other possible business opportunities, such as selling castoff chemicals, converting hemp fibers into shirts, and selling hemp beer and shirts on the side of the road to bolster your hip and cool status amongst the dirty moon youth.
The benefit of double-ended searching is to cast the net wider for more search results, and to more evenly weight the two desires better (searching from just one end will favor that end). The benefit of A* is simply to quickly filter the most direct route. The benefit of recursion is to automate the searching process and to aid in dependency tracking, and as well to cast a wide net for incidental and creative solutions (i.e. as at the beginning how all roads are built will be searched, not just roads built with oil). There is no benefit to multidimensional, it just comes with the package, although it would make for wonderful graphs.-- mylodon, Jun 07 2008 (?) The top Google result! Multidimensional_2c...20Solution_20SearchShould've seen it coming. [MikeD, Jun 07 2008] http://www.powerset.com/ Trying to understand questions and the searched documents. [jutta, Jun 08 2008] This is just a hypothetical road, right ?-- 8th of 7, Jun 07 2008 I typed //How to build a road, on the moon; made out of hemp oil// into a Google search ... link.-- MikeD, Jun 07 2008 Hemp oil bricks are yellow, right?-- lurch, Jun 07 2008 This is also called "semantic search" - searching based on the meaning of things, not their specific wording.
The basic technology it requires goes by the name of "natural language understanding" and is one of the holy grails of Artificial Intelligence. It's harder than you make it out to be; you don't get very far just by replacing phrases with other phrases, because words have so many meanings in different contexts, and the relationships between A() and B() are complicated.
For example, when you've really made something out of yourself, or you made it out of the crowded concert hall in time to catch the last train - those are different "made out of"s. Statistics can help here, but errors also add up, until your search engine is more funny than helpful.
While companies like powerset are still working on that, did you know that prefix-tilde (~) turns on synonym matching in google?-- jutta, Jun 08 2008 Ask Jeeves about this idea.-- WcW, Jun 08 2008 The primary difference between this and something like ask jeeves or powerset is that the result, i.e. what you get back, is not a single webpage.
The pathfinding aspect of this returns the linked values, so the result is extracted contents of many web pages, reformatted, and with added value (such as prices, locations, stock quotes,e tc). The assumption is that the relationship from A to B has never been established by anyone anywhere on a single page.
Returning a list of closest 'results' is not what we want; we want lists of an A-to-B graph of connections. Of course these will be branching and multidimensional.-- mylodon, Jun 08 2008 I think you are describing an ontological natural language mediator for information retrieval that then also mediates the retrieved various kinds of data into an uncannily useful structured report.
Any part of that working properly would be an amazing IR breakthrough. I know people are working on it.
I am for it [+], even though it sounds WIBNI.-- Laughs Last, Jun 08 2008 Not WIBNI at all. All these algorithms exist, its a question of fine tuning. Jutta brought a great site, it worked for many questions I asked! I once was assigned to check technology of a company called Intelligate. We aquired the company following that, and I know that it still exists (perhaps with a new name). This technology creates a discussion with you, out of free text, asking you what you meant until it is "sure" it has all parameters needed to give you an intelligent answer. So it would ask: "Are you looking for engineering or chemical details"? And would intelligently parse your answer and add it to the pool of knowledge, filtering and fine tuning the request.
The kind of extra discussion could be a feature of this halfbaked idea.
In any case mylodon, please give it a name, for the rest of the discussion. How about DeepSearch, or AnswerSearch?-- pashute, Jun 10 2008 The discussion idea seems cool but... I don't think natural language actually is required for this idea, at least not for interface. It would be nice but a formal query would work as well. The primary point of this idea is the pathfinding and intersection.
This originated when I cracked a bad joke to a transportation engineer about George Washington making roads out of hemp, because it was the best material for the job. Nowadays of course big oil is keeping it down.
But could you actually make a road out of hemp? I started by searching first for what roads were made out of, and then for what hemp could be made into. Eventually, I discovered that yes, you could use hemp for bitumen, but it would really only work for roads that didn't fall much below freezing.
In order to find that out I needed to 'pathfind' across about six different webpages (linked by topic, in a cloud of a dozen or more not necessarily linked webpages) until I reached an intersection where 'what hemp could be made into' matched 'what roads were made of', qualified (somewhat) by 'on the moon'.
And it was branching -- where hemp oil can be made into bitumen, hemp oil can also be processed to glycol (if I'm remembering this right) which is an antifreezing additive -- thus while not required, it would fulfill any requirements for (somewhat) colder climates. This was a bonus find! I found this coming from the 'made of hemp oil' side, whereas the qualities and use of vegetable oil bitumen came from 'what are roads made from'.
It was a pretty dead simple search, I just looked for keywords and ignored most context, in the manner someone would use to quickly bolster a political argument with a chain of research, linking a politician with an unsolved string of dead hookers, or connecting royalty to UFO phenomena.
Maybe I should have called it the 'Skeletons In Your Closet Search'. Be popular with the press!-- mylodon, Jun 10 2008 I am opposed to the "moon road" but it appears I will rather than fight to the death for your right to speak publicly I will be casually assassinated while remaining amiable about things for this right-- beanangel, Jun 10 2008 //did you know that prefix-tilde (~) turns on synonym matching in google? // I didn't. I've always wondered if google search terms could include symbolic items. Are there any other such tricks? Is there a place where they're documented?-- MaxwellBuchanan, Jun 10 2008 I would love to see the iterations:
"build a road on the moon made out of hemp oil" "build a road" "on the moon" "made out of" "hemp oil" "construct a path" "in space" "using" "hemp" "create a journey" "to the stars" "using" "marijuana"-- reap, Jun 11 2008 The first time reading through I had missed the importance of recursing after identifying keywords and what Amazon.com calls "Statistically Improbable Phrases".
I enjoy doing that kind of research. Stop trying to automate my hobbies.-- Laughs Last, Jun 11 2008 random, halfbakery