7 ms·

I intend to announce an alpha test here on HN. Some people at Battelle, etc. worked in information retrieval decades ago. A friend of my father's was involved

I intend to announce an alpha test here on HN. Some people at Battelle, etc. worked in information retrieval decades ago. A friend of my father's was involved. Those people concluded that key words could cover only part of search. My rough guess is 1/3rd. I'm going for the other 2/3rds. The computing is pretty simple and fast. Given enough documentation, the code was easy to write. The crucial core of the work is some applied math I derived based on some advanced pure math prerequisites. I got most of the prerequisites in my applied math Ph.D. So, to me, the project is mostly applied math with some cute data manipulations and a simple Web site user interface. Maybe for a computer science audience, my work should do well giving people content with the meaning they have in mind. I know; I know: Writing code for much of anything having to do with meaning has been a challenge. Well, I derived some math. Tonight the work is not very exciting: I'm writing some Rexx code making crucial use of some cute Rexx functions to correct the time-date stamps on the directories written by Microsoft's XCOPY. The time-date stamps written are usually not those of the source directories and, instead, are just the time, date when XCOPY created the copy. The files XCOPY writes DO retain the time-date stamps of the source. For the correction, my idea is to set the time-date stamps of each directory to the newest time-date stamp of the files/directories in that directory. Important is how the Windows NTFS file system does time-date inheritance: If create or change a file or create a directory, then the time-date stamp on the parent directory will within a few seconds be changed to time-date stamp on the new or changed file or the new directory. If the new directory has a file created in it, then, sure, that directory will have it's time-date stamp changed, BUT, surprising or not, the parent we mentioned will not have its time-date changed. I'm typing quickly; if this is unclear, complain and I'll be more clear. This information is important: If the time-date inheritance kept feeding up the directory tree to the root, then the corrections I want would have to work essentially from the leaves of the directory tree up and in a 'breath first' way, with some tricky accumulation at each 'level' (distance from the roof) of some maximum time-date values, to the root. But with the way the inheritance actually does work, I can correct the time-date stamps as I mentioned on the directories in any order. Not very exciting but has to be done.

anonytrary 8y ago
It sounds pretty clear to me, albeit your explanation is long winded. You're saying that, in the FS you're using, the time-stamp changes are "local" in the sense that they don't bubble up the tree after N = 1 levels. In elevator-pitch terms, you're working on a search engine that takes into account user intent more accurately? Is that more or less correct?
- graycat 8y ago
  Thinking a little more, if the time-stamp changes did bubble up, then just start making changes at the leaves of the tree; at each leaf, work back to the root on the unique path to the root; and at each directory make a change in the time-stamp of that directory if and only if the new time-date stamp would be more recent than the old one; and, if don't make a change at that directory, then are DONE on that path. Revision, to make this work, first set all the directory time-date stamps to, say, year 1776. This solves the problem that the directories closer to the root may be on the path of several leaves and get changed several times. To identify the leaves, that is, directories with no subdirectories, get the directory tree names and just sort them into ascending order. Then a leaf directory and all the directories on the path from that leaf back to the root will sort together with the leaf directory the last one of those that sort together. That is, a directory is a leaf if and only if it is not on the path to the root (not an initial substring, using whole directory names) of the next directory in the sort. In particular, the last directory in the sort is necessarily a leaf. Ah, an algorithm! I don't want now to further characterize or change the wording on what the startup is doing. My best word is the one I used, "meaning".
  - anonytrary 8y ago
    How exactly will you know a user's "meaning" when they are asking for content? Do you intend to read their minds, somehow? Or perhaps you might hook them up to a heart rate monitor, track their eye movements and use ML to obtain user sentiment without requiring them to provide click-based feedback (like likes/dislikes).
    - 8y ago

anonytrary 8y ago

It sounds pretty clear to me, albeit your explanation is long winded. You're saying that, in the FS you're using, the time-stamp changes are "local" in the sense that they don't bubble up the tree after N = 1 levels. In elevator-pitch terms, you're working on a search engine that takes into account user intent more accurately? Is that more or less correct?

graycat 8y ago
Thinking a little more, if the time-stamp changes did bubble up, then just start making changes at the leaves of the tree; at each leaf, work back to the root on the unique path to the root; and at each directory make a change in the time-stamp of that directory if and only if the new time-date stamp would be more recent than the old one; and, if don't make a change at that directory, then are DONE on that path. Revision, to make this work, first set all the directory time-date stamps to, say, year 1776. This solves the problem that the directories closer to the root may be on the path of several leaves and get changed several times. To identify the leaves, that is, directories with no subdirectories, get the directory tree names and just sort them into ascending order. Then a leaf directory and all the directories on the path from that leaf back to the root will sort together with the leaf directory the last one of those that sort together. That is, a directory is a leaf if and only if it is not on the path to the root (not an initial substring, using whole directory names) of the next directory in the sort. In particular, the last directory in the sort is necessarily a leaf. Ah, an algorithm! I don't want now to further characterize or change the wording on what the startup is doing. My best word is the one I used, "meaning".
- anonytrary 8y ago
  How exactly will you know a user's "meaning" when they are asking for content? Do you intend to read their minds, somehow? Or perhaps you might hook them up to a heart rate monitor, track their eye movements and use ML to obtain user sentiment without requiring them to provide click-based feedback (like likes/dislikes).
  - 8y ago