Hacker News Reader

jisaacsoon How We Derive Context from Big Data

DBpedia is a great, structured representation of Wikipedia. Agreed, there are many challenges, NLP being one. I think you'll like our other blog posts http://blog.urx.com/urx-blog/2015/7/28/named-entity-recognit...

jisaacsoon How We Derive Context from Big Data

Great example! This is the exact problem URX is solving: empowering content providers to present their products within meaningful contexts.

jisaacsoon Automatic Keyword Extraction from Text

Opentable gave a great presentation on this not too long ago, check it out http://www.slideshare.net/SudeepDasPhD/recsys-2015-making-me...

jisaacsoon Automatic Keyword Extraction from Text

Great suggestion. We've used LDA for topic modeling in the past. I'm a big fan of running word2vec and clustering.

jisaacsoon Automatic Keyword Extraction from Text

Totally agree! Our goal was to work with Insight Data Science to build a lightweight, simple keyword extractor in 3 weeks.

jisaacsoon The Science of Crawl: Improving on PageRank

Author here,

The article discusses applying PageRank to help decide how a web crawler should discover new content. While this article does not touch on how URX evaluates search quality, I can promise it's a metric we care deeply about. One of the slogans on our home page "Page content + user context = relevance".

Happy to answer any questions.

jisaacsoon Atomwise (YC W15) Discovers Drugs for Diseases That Don’t Even Exist Yet

Hi Abraham, could you describe some of the differences between Atomwise's approach and D.E. Shaw research's approach to computational identification of novel pharmaceuticals? Thanks!

jisaacsoon The Science of Crawl, Part 1: Deduplication of Web Content

Thanks for the reference. It looks like SequenceMatch is "cubic time in the worst case and quadratic time in the expected case". Did you notice any performance issues as kouio scaled?

jisaacsoon Mapcode – A short address for any location on Earth

There exists a hash of lat, long globally unique for each level of precision, a geohash [1]. One common application utilizes a space filling Hilbert curve to define an arbitrary precision bounding box around a set of coordinates. You can see an example hash [2].

[1] http://en.wikipedia.org/wiki/Geohash [2] http://geohash.org/9q8yyvfekk8h

jisaacsoon Ask HN: Who is hiring? (May 2014)

URX (YC S13) - Core software engineer, Android engineer, Data scientist - San Francisco, CA

URX is a fast growing, engineering-focused team with backgrounds in machine learning and e-commerce. We look for mission-driven people who have large ambitions and are curious to learn more about rapidly developing markets and technology.

Our engineering team is constructing the fabric that is connecting the web of mobile apps. We are building a platform to handle billions of data points and enabling thousands of apps to benefit from our intelligence. Build with our engineering team.

Check out our guiding tenants. http://blog.urx.com/post/70336738684/urxs-guiding-tenets

If this sounds like you apply online at: http://urx.com/careers ^^please mention "HNMay14"

jisaacsoon Genetic mugshot recreates faces from nothing but DNA

Check out https://www.facebase.org/

Facial prediction is still in its infancy but is an exciting area of research.

jisaacsoon Differential Equations in Data Science

Good catches. One point,

"because they can be used to describe how quantities change over (continuous) time"

I would argue that time series problems are common throughout data science. Uber's prediction of arrival time, Kaggle flight time prediction, etc.

jisaacsoon Differential Equations in Data Science

That's awesome to hear! Keep me posted as you work through some projects. I'm curious to hear how ODEs can help in environmental science.

jisaacsoon Differential Equations in Data Science

I would say it's under-appreciated in modern data science. It's well appreciated in applied mathematics.

jisaacsoon My deadly disease was just a 23andme bug

This will likely be downvoted to hell, but personally I'm a fan of the service 23andme provides. It has aggregated scientific knowledge across a large number of domains to provide an end-to-end pipeline for ancestry prediction, kinship analysis and phenotype prediction.

It's a service built on learning algorithms to correlate features (DNA SNPs) with diseases. As with any learning algorithm, results should be interpreted with care. That doesn't lessen the fact that predictions it provides can be helpful.

As an aside, 23andme has a solid API. It's opened up (with consent) a huge DNA dataset for developers to mine. I'm excited for the science that can be learned from this data.

jisaacsoon My deadly disease was just a 23andme bug

To take it one step further, I don't necessarily trust studies linking genomic variants with disease significance. A subset derive correlations using oversimplified models:

http://www.ncbi.nlm.nih.gov/pubmed/23770567

jisaacsoon My deadly disease was just a 23andme bug

You can potentially predict a male's surname from a DNA sample. The accuracy is roughly proportional to the frequency of the name.

http://www.sciencemag.org/content/339/6117/321.abstract