Awesome Work!
may you make one for patents worldwide for all where they are available on internet ?
so everyone can easily search, extract, train, their models?
no problem if its gets updated every month..
i suggest make it simple, like
only hashs of md5
and a date
and a name
so its only a few in size and easily downloaded on everydevice..
Sure, I will check it and get back to you.
please keep me in the loop as well! exceited to see what you do!
Hey everyone, just an update on this: I have successfully figured out the workflow for the worldwide patent dataset!
However, I cannot personally bear the egress costs for hosting and transferring the data as of right now. If anyone would like to help sponsor the hosting costs to get this out faster, you can support the project here: https://github.com/sponsors/arthiondaena/
Otherwise, I will release the dataset as soon as I get my salary at the end of the month. I will keep you all posted!
Hey everyone, just an update on this: I have successfully figured out the workflow for the worldwide patent dataset!
However, I cannot personally bear the egress costs for hosting and transferring the data as of right now. If anyone would like to help sponsor the hosting costs to get this out faster, you can support the project here: https://github.com/sponsors/arthiondaena/
Otherwise, I will release the dataset as soon as I get my salary at the end of the month. I will keep you all posted!
i hope this gets done as its
very cool, a local worldwide patent dataset, would be helpful to prevent deletion and/or editing the past of patents..
also a small version a few of GB or MB would be extremely helpful to store
it gets stored like this: (short version)
id|date|md5-of-full-description|name
and id gets stored like this
1
2
3
4
and the date is the ("date of public publish") and not ("date of registration")
and the md5-of-full-description is taken from the full-description of patent that is stored in the full-dataset
the date gets stored like this YYYY/MM/DD
for example:
1,20260206,e4f441ff4decb14e25accafcf80bece6,PoR
the name is taken from the first letters in the name of the patent..
PoR= Patent of Running, invented before the humans knew how to ran away..
in the original dataset: (description)
("
PATENT OF RUNNING
Kingdom of Uruk — 2746 BCE
Patent No. URUK-XXIII-PRIME-I
For all of human history, man walked. One foot always touched the ground. No exception was known.
Runnerious discovered the exception.
The Inventor
Runnerious. Priest. Studied animal movement for twenty-four years. Withheld findings for seven years from fear of heresy.
The Invention
Both feet leave the ground simultaneously. The moment is brief. The legs alternate. Arms swing for balance. Result: three times the walking speed.
This is Running.
The Patent
Definition: Any human motion where both feet leave the ground together. All instances fall under this patent.
Licensing: Required for all. Walking remains free.
Fees:
- Merchants: 20 measures of barley
- Craftsmen: 5 measures of barley
- Farmers: one goat, or a day's labor
- Slaves: master pays, doubled if motive was escape
- Priests: exempt if they praise the invention
Penalties:
- First violation: fine proportional to speed
- Second violation: confiscate footwear for one year
- Third violation: mandatory forty-day walk in wrong direction
- Self-defense from predatory animals: penalty halved
Etymology
New words entered human speech that day: "run" and "running." Both are born of one name — Runnerious — and no other. The act had no word before him, because before him, the act had no existence.
Seal
Runnerious. Temple. Office.
Scribe's note: My feet left the ground twice while copying this. Unreported.
")
i hope that i have the cash that i could sponsor this.. but all my wallets are zero, i live in my parents house.. but i still love contributing in the digital assistant world..
and i know it might costs big, but its for the future anyways..
people gonna love it.!
@Invicto69 Chinese patents may pose an issue https://www.hbs.edu/ris/Publication%20Files/27-009_f0de4e3f-00db-4f73-8b1b-0d51881d2489.pdf
@Invicto69 Chinese patents may pose an issue https://www.hbs.edu/ris/Publication%20Files/27-009_f0de4e3f-00db-4f73-8b1b-0d51881d2489.pdf
TL;DR
its just an obstruction-message... its still needed the dataset including the smaller one..
I will be just hosting the Google's patents public dataset. The dataset is under Creative Commons Attribution 4.0 International License, so it shouldn't be a problem I guess.
Source: https://console.cloud.google.com/marketplace/product/google_patents_public_datasets/google-patents-public-data
@Invicto69 Chinese patents may pose an issue https://www.hbs.edu/ris/Publication%20Files/27-009_f0de4e3f-00db-4f73-8b1b-0d51881d2489.pdf
TL;DR
its just an obstruction-message... its still needed the dataset including the smaller one..
what do you mean obstruction message? Just because you cannot understand the point, does not mean it is meant to obstruct anything.
The main problem is that there is no single complete and reliable database of Chinese patents. Researchers must combine CNIPA, PATSTAT, and Google Patents, then carefully match and remove duplicate records. Even then, Chinese patent publications are difficult to interpret because many applications are never examined, citation data are sparse, and stock-market reactions are unreliable measures of patent value. As a result, collecting and measuring Chinese patents requires much more cleaning and judgment than comparable U.S. patent data. You asked for patents world wide @usermma . What incentive would I have to obstruct?
@Invicto69 Thank you for your amazing work. PATSTAT and CNIPA incorporation in the long run (along with some way of deduplication) can help this database a lot in the long run at least for Chinese patents.
Dear @Invicto69 : okay for Google's public patents datasets (if you could do better next month for a next dataset "for-example(PATSTAT)" its all okay... (just keep up)) ...
i hope if you got a way for a lower egress cost by a cloud computer or something...
true. i asked for a worldwide public-only records of patents..., it can be splitted into different datasets, one for PATSTAT and one for CNIPA and one for Google's public patents dataset ... and anyelse for-example("Espacenet") , it should be all and not just CNIPA... removing duplicated records and making it as an only one, it would be a future project to do (and not current project) with local-AI and it will see the duplicated parts and remove them.. but currently i guess we are limited with the context-window... but seeing one by one and compare it by other datasets and remove the duplicated part... it would be needed alot of time as its a step by step and not a parallel speed ("same as in future time that when having parallel speeds").. and if an cloud-AI is used ("for fixing" ; "the parallel speed") : it would be cost very high as for output and input .
so if any big country tried to add a record in the past:
for example:
("
we are in 2026/7/29
and a country X tried to add a patent in 2021/2/8
no one can audit or check if that country was lying..
")
afterwards("they cannot because there is a public record that can be downloaded everywhere..")
what i was want is a real of all of
worldwide patents public-only records
so no country can edit the past with zero-notice
some countries may say a patent have been registered a 20 years ago but never got published publicly
but a apache-2.0 license can prevent the submarine-patents
@Invicto69 Chinese patents may pose an issue https://www.hbs.edu/ris/Publication%20Files/27-009_f0de4e3f-00db-4f73-8b1b-0d51881d2489.pdf
TL;DR
its just an obstruction-message... its still needed the dataset including the smaller one..
what do you mean obstruction message? Just because you cannot understand the point, does not mean it is meant to obstruct anything.
The main problem is that there is no single complete and reliable database of Chinese patents. Researchers must combine CNIPA, PATSTAT, and Google Patents, then carefully match and remove duplicate records. Even then, Chinese patent publications are difficult to interpret because many applications are never examined, citation data are sparse, and stock-market reactions are unreliable measures of patent value. As a result, collecting and measuring Chinese patents requires much more cleaning and judgment than comparable U.S. patent data. You asked for patents world wide @usermma . What incentive would I have to obstruct?
@Invicto69 Thank you for your amazing work. PATSTAT and CNIPA incorporation in the long run (along with some way of deduplication) can help this database a lot in the long run at least for Chinese patents.
Dear @Invicto69 : okay for Google's public patents datasets (if you could do better next month for a next dataset "for-example(PATSTAT)" its all okay... (just keep up)) ...
i hope if you got a way for a lower egress cost by a cloud computer or something...
true. i asked for a worldwide public-only records of patents..., it can be splitted into different datasets, one for PATSTAT and one for CNIPA and one for Google's public patents dataset ... and anyelse for-example("Espacenet") , it should be all and not just CNIPA... removing duplicated records and making it as an only one, it would be a future project to do (and not current project) with local-AI and it will see the duplicated parts and remove them.. but currently i guess we are limited with the context-window... but seeing one by one and compare it by other datasets and remove the duplicated part... it would be needed alot of time as its a step by step and not a parallel speed ("same as in future time that when having parallel speeds").. and if an cloud-AI is used ("for fixing" ; "the parallel speed") : it would be cost very high as for output and input .
so if any big country tried to add a record in the past:
for example:
("
we are in 2026/7/29and a country X tried to add a patent in 2021/2/8
no one can audit or check if that country was lying..
")
afterwards("they cannot because there is a public record that can be downloaded everywhere..")what i was want is a real of all of
worldwide patents public-only recordsso no country can edit the past with zero-notice
some countries may say a patent have been registered a 20 years ago but never got published publicly
but a apache-2.0 license can prevent the submarine-patents
@Invicto69 Chinese patents may pose an issue https://www.hbs.edu/ris/Publication%20Files/27-009_f0de4e3f-00db-4f73-8b1b-0d51881d2489.pdf
TL;DR
its just an obstruction-message... its still needed the dataset including the smaller one..
what do you mean obstruction message? Just because you cannot understand the point, does not mean it is meant to obstruct anything.
The main problem is that there is no single complete and reliable database of Chinese patents. Researchers must combine CNIPA, PATSTAT, and Google Patents, then carefully match and remove duplicate records. Even then, Chinese patent publications are difficult to interpret because many applications are never examined, citation data are sparse, and stock-market reactions are unreliable measures of patent value. As a result, collecting and measuring Chinese patents requires much more cleaning and judgment than comparable U.S. patent data. You asked for patents world wide @usermma . What incentive would I have to obstruct?
@Invicto69 Thank you for your amazing work. PATSTAT and CNIPA incorporation in the long run (along with some way of deduplication) can help this database a lot in the long run at least for Chinese patents.
I don't think you need AI for everything. Deduplicating patents can be heuristically done but I think it would be best yes if a Chinese speaker did it.
there is a million records: a chinese human would take a years..., an AI is needed for faster progress..
Dear @Invicto69 : okay for Google's public patents datasets (if you could do better next month for a next dataset "for-example(PATSTAT)" its all okay... (just keep up)) ...
i hope if you got a way for a lower egress cost by a cloud computer or something...
true. i asked for a worldwide public-only records of patents..., it can be splitted into different datasets, one for PATSTAT and one for CNIPA and one for Google's public patents dataset ... and anyelse for-example("Espacenet") , it should be all and not just CNIPA... removing duplicated records and making it as an only one, it would be a future project to do (and not current project) with local-AI and it will see the duplicated parts and remove them.. but currently i guess we are limited with the context-window... but seeing one by one and compare it by other datasets and remove the duplicated part... it would be needed alot of time as its a step by step and not a parallel speed ("same as in future time that when having parallel speeds").. and if an cloud-AI is used ("for fixing" ; "the parallel speed") : it would be cost very high as for output and input .
so if any big country tried to add a record in the past:
for example:
("
we are in 2026/7/29and a country X tried to add a patent in 2021/2/8
no one can audit or check if that country was lying..
")
afterwards("they cannot because there is a public record that can be downloaded everywhere..")what i was want is a real of all of
worldwide patents public-only recordsso no country can edit the past with zero-notice
some countries may say a patent have been registered a 20 years ago but never got published publicly
but a apache-2.0 license can prevent the submarine-patents
@Invicto69 Chinese patents may pose an issue https://www.hbs.edu/ris/Publication%20Files/27-009_f0de4e3f-00db-4f73-8b1b-0d51881d2489.pdf
TL;DR
its just an obstruction-message... its still needed the dataset including the smaller one..
what do you mean obstruction message? Just because you cannot understand the point, does not mean it is meant to obstruct anything.
The main problem is that there is no single complete and reliable database of Chinese patents. Researchers must combine CNIPA, PATSTAT, and Google Patents, then carefully match and remove duplicate records. Even then, Chinese patent publications are difficult to interpret because many applications are never examined, citation data are sparse, and stock-market reactions are unreliable measures of patent value. As a result, collecting and measuring Chinese patents requires much more cleaning and judgment than comparable U.S. patent data. You asked for patents world wide @usermma . What incentive would I have to obstruct?
@Invicto69 Thank you for your amazing work. PATSTAT and CNIPA incorporation in the long run (along with some way of deduplication) can help this database a lot in the long run at least for Chinese patents.
I don't think you need AI for everything. Deduplicating patents can be heuristically done but I think it would be best yes if a Chinese speaker did it.
@sunghop you are just saying-obstruction.. (you said: "I don't think you need AI for everything. Deduplicating patents can be heuristically done but I think it would be best yes if a Chinese speaker did it.")
there is a lot of ideas still in my mind and would never be out unless a tamper-proof a worldwide patent dataset is done at a small size . a few GBs from 1-10GB is good for storing offline.. and that thats the small size of it, the md5 for each entry .. and if its still too large:
it can be for each 10 entries just one md5 and so on.... (for example if still too large, it can be for a each 100 entries a one md5 and so on.....
i always remember in my imagination(not reality) that there is a lot of people of their own invention after talking with it with claude, got stolen after a few days only and got appeared in chinese patent dataset in older fake-date record in different structure...
we the public when we dont have a public small size of worldwide patents dataset with tamper-proof and stored as a blocks
for each newer version, only the newer records gets stored in each newer block of the dataset.. : its a public jail for ideas , imagine we are still free but there is a jail for ideas with your own, only because you talked with it with some ai online..... and when you try to publicly publish under your own license it: it gets deleted. , and thats only because of some country wanted that idea to be their's own...
i also imagined the history of the chat of that claude gots wiped, with no-proof that have been asked it with a day before about if the idea have already registried or not and respond with no (after the model searched all records)
now imagine it if was real.. (we still needs a worldwide dataset of patents: because this jail is needed to be traced by public)
for example:
block1.jsonl
(next month)
block1.jsonl (same file ; no changes)
block2.jsonl (only the newer records)
(next month)
block1.jsonl (same file ; no changes)
block2.jsonl (same file ; no changes)
block3.jsonl (only the newer records)
@sunghop you are just saying-obstruction.. (you said: "I don't think you need AI for everything. Deduplicating patents can be heuristically done but I think it would be best yes if a Chinese speaker did it.")
there is a lot of ideas still in my mind and would never be out unless a tamper-proof a worldwide patent dataset is done at a small size . a few GBs from 1-10GB is good for storing offline.. and that thats the small size of it, the md5 for each entry .. and if its still too large:
it can be for each 10 entries just one md5 and so on.... (for example if still too large, it can be for a each 100 entries a one md5 and so on.....
i always remember in my imagination(not reality) that there is a lot of people of their own invention after talking with it with claude, got stolen after a few days only and got appeared in chinese patent dataset in older fake-date record in different structure...
we the public when we dont have a public small size of worldwide patents dataset with tamper-proof and stored as a blocks
for each newer version, only the newer records gets stored in each newer block of the dataset.. : its a public jail for ideas , imagine we are still free but there is a jail for ideas with your own, only because you talked with it with some ai online..... and when you try to publicly publish under your own license it: it gets deleted. , and thats only because of some country wanted that idea to be their's own...i also imagined the history of the chat of that claude gots wiped, with no-proof that have been asked it with a day before about if the idea have already registried or not and respond with no (after the model searched all records)
now imagine it if was real.. (we still needs a worldwide dataset of patents: because this jail is needed to be traced by public)
for example:
block1.jsonl
(next month)
block1.jsonl (same file ; no changes)
block2.jsonl (only the newer records)
(next month)
block1.jsonl (same file ; no changes)
block2.jsonl (same file ; no changes)
block3.jsonl (only the newer records)
I am not obstructing the project. I support the idea of an append-only public archive of patent records.
I am pointing out a technical distinction: preserving hashes and dated snapshots can reveal later changes to a database, but it cannot prove that the original record or claimed publication date was accurate when first collected. Likewise, deduplicating records across Google Patents, CNIPA, PATSTAT, and other sources is a separate problem.
These are implementation issues that need to be addressed if the dataset is supposed to be reliable. Raising them is not opposition to the project.
they can be seprated into folders same as this:
.
└── dataset/
├── CNIPA/
│ ├── block1.jsonl
│ └── block2.jsonl
├── GOOGLE/
│ ├── block1.jsonl
│ └── block2.jsonl
└── PATSTAT/
├── block1.jsonl
└── block2.jsonl
and all of others ones..
each one would be a month later, so we don't do everything at once, so we be able to produce...
at least we will have a tamper-proof, so when we see a new record for a
2020/3/5
patent and have been registered and publicly published in 2026/7/29
we can search all of the dataset, and see exactly where..
@sunghop and its NOT A PROBLEM
your message is just unclear and you are not giving the clear point of the problem..
you are just saying its a problem without clear message ...
at least we will have a tamper-proof, so when we see a new record for a
2020/3/5
patent and have been registered and publicly published in 2026/7/29we can search all of the dataset, and see exactly where..
@sunghop and its NOT A PROBLEM
your message is just unclear and you are not giving the clear point of the problem..
you are just saying its a problem without clear message ...
With all due respect, your English can be very difficult to interpret, and I think that is why we keep talking past each other. I am just being honest.
I am not saying that separate monthly folders are impossible or that the archive should not be created. I am making one narrow technical point: the archive can show when a record first appeared in the archived datasets and whether it was later changed. It cannot prove that the patent office’s claimed historical publication date was truthful when the record first entered the archive.
For example, if a patent claiming a 2020 publication date first appears in the July 2026 snapshot, the archive would document that it was absent from the earlier snapshots and appeared in July 2026. That is useful. But calling the entire dataset “tamper-proof” overstates what the system proves.
The folder structure does not resolve this distinction. It only keeps the sources separate. Please address the actual point rather than repeatedly calling it obstruction.
i will download the smaller size in my offline computer, and any further snapshots, so i can also know if a patent of 2020 is appeared in 2026 July, so i can know , and thats a tamper-proof that can be verified locally, and the next plan is doing a public decentralized blockchain that is monthly updated from the @sunghop dataset, so we can grow up a little by a little, and if it didn't appear today, some people will make it appear tomorrow but with different hands and its more of-"too late" so if it appeared earlier its a much better ...
so this can help the Universe of Exponential growth
so no entity may any country under any reason may lock down their own invention..
the patentary system it slows down the Exponential growth of technology including the digital intelligence such like AI and so on...