Ask HN: How to monetize an algorithm? | Hacker News Reader

Ask HN: How to monetize an algorithm?

7 pointsSingAlong13 comments

How to monetize an algorithm? Some suggest licensing it to different companies. But will I have to show them the math behind the algo? Is it possible to monetize it in a way in which the math is not revealed?

(--edited later after feeback in comments--): I have two algos. One algo is an addon for search engines. The data generate by the algo improves usability of the search engine. Its not a ranking system. Just a taxonomy addon on top of othe ranking system that groups terms as required. Example: searching for 'gates', would generate the topics related to gates. Results will be precise as:

'gates' as in bill gates and microsoft

'gates' as in house

The other algo is just still a crappy 'basic stuff' stage. This algo analyse notes in the input sound and classifies it as music and noise. And this I still have a lot to complete.

View on HN

Do a literature search (I'll try to find mine), whether you have a effective /original named entity recognition algo, how scalable and domain-specific it is. This is also referred to, or covered in related research, as anaphora or antecedent resolution and co-reference / record linkage / deduplication. sounds kinda like a gazzetteer (sp?)

To quote one paper (ginormous project, 4 PI's at UIUC):

http://serrano.ai.uiuc.edu/doi/

==============

research on methods for Named Entity Recognition (NER) is voluminous but has tended to focus on the problem in widely used languages such as English, otherWestern European languages, Arabic, and Asian languages such as Chinese, Japanese and Korean.

I'm guessing that youre doing some kinda work in NLP, so you know where the literature is, but for folks who aren't, the first stop is

http://aclweb.org/anthology-new/

i have in my notes that NER was a conference task for 2003

http://www.cnts.ua.ac.be/conll2003/ner/

Of course you can monetize it without revealing the algorithm, just expose an API.

The questions you need to ask yourself is what are the advantages of your algorithm in comparison to known algorithms for tackling the same type of tasks, which companies use algorithms to accomplish these tasks and what is the monetary worth of the improvement in their results your algorithm generates.

Google built an application out of theirs, and monetized it. Their search engine could also just have been a proof-of-concept, used as a demo in selling PageRank to another firm.

You should investigate patents (preferably with a patent attorney), and figure out just how easy it would be to reverse-engineer your algorithm.

Keep in mind that many programmers find the idea of secret algorithms distasteful. Hiding these processes is often a necessary evil in order to make profit, but do not do it if you don't have to (recent example: reddit).

Not sure where to begin here without knowing anything about what you're "algorithm" is about, what it would be used for, etc.

If you've created something useful, why not build a business on top of it yourself?

If you don't want to do that, I suppose you could patent your algorithm and then grant licenses to use it. If what you're licensing is literally just "the algorithm" as opposed to an implementation of the algorithm that you write, than I don't see how you can get around telling them "the math behind the algo".

Yup! I just updated thread with more info. I had created a thread with a generalized title thinking others with different algorithms would find replies useful too. :)

And still more info: The algorithm doesn't generate tags but instead generates precise topic results as shown in the thread title/description. But I am still tweaking it.

Question is what do you really want out of life.

Is it good enough to be the basis of a company? Are you daring enough to start a company? These are some questions you have to answer.

I think the best way is to demo this by starting a free service. If its good enough and meets people's needs you people will go to you with offers.

If you REALLY believe in the value of this algorithm, I would think the best way to monetize it would be to put it to commercial use. The usual reason to license is because implementation requires too much capital. Algorithm commercialization is cheap.

If you really can do natural language parsing that well, the most upside is in starting your own search engine.

Hmm... Its not so good right now. Right now its working with a few sample data sets I chose: python, gates and sun

I am currently downloading the wikipedia database which is about 5gb(on my slow internet connection should take another 3-4 days). The reliability of the algorithm will then be known.

Create a proof of concept and make it publicly available.

Hmm... I am trying to get some bucks for the attorney fee and the patent fee. And also getting my friend's help. I currently have a webapp under dev too(Also taking part in a contest to save some money for the patent fee which is expensive in my country)

So will post the demo here soon. :)

SingAlong, I think both commenters are right. To put it in more explicit terms: You can't sell the algorithm, you can only sell its output. I.e. Google has an algorithm to determine the relevance of a web page to a search result, but what they are selling you is the output, i.e. the relevance for a given combination of search term and page. In order to do that, you will have to host the software that implements the algorithm on a server yourself and your customers will have to feed the input to it and pay for the output in a subscription-like model or similar. Email me at jerry5@p2pbroking.net if you are interested in an exchange of ideas.

Reply on news.ycombinator.com