Open Sentiment Analysis

Photo by Courtney Carmody

Sentiment analysis is fiendishly hard to solve well, but easy to solve to a first approximation. I've been frustrated that there have been no easy free libraries that make the technology available to non-specialists like me. The problem isn't with the code, there are some amazing libraries like NLTK out there, but everyone guards their training sets of word weights jealously. I was pleased to discover that SentiWordNet is now CC-BY-SA, but even better I found that Finn Årup has made a drop-dead simple list of words available under an Open Database License!

With that in hand, I added some basic tokenizing code and was able to implement a new text2sentiment API endpoint for the Data Science Toolkit:

Give it a try, it's as simple as a CURL call from the terminal:

curl -d "I hate this hotel" ""

{"score": -3.0}

I've been having a blast with it, simple-minded as it is, so I hope you do too!

Five short links


A Global Poverty Map Derived from Satellite Data – This is an old paper from 2006, but I love the idea of using how much light that a neighborhood sends into to the night sky to measure how wealthy it is. Richness is highly correlated with wastefulness, apparently.

Open Multi-lingual WordNets – We’re mapping our inner worlds too, these open data sets are incredibly useful information on word meanings for anyone working with computers and human languages.

The Invisible City – A fake Canadian city briefly appeared on OpenStreetMap, complete with an elaborate public transport network. Or was it briefly a real place blinking in and out of existence, with only a lone volunteer mapper spotting it?

The Dark Side of Social Capital – We usually think of community as a good thing, but anybody who grew up in a small town can tell you that the power can be used to exclude outsiders too.

K2C 1N5 – Ervin Ruci is being hounded by the Canadian Postal Service for the crime of making a crowdsourced database of postal codes freely available, and now they’ve decided they own the copyright to the words “postal code” too!

Five short links

Photo by Tasty Goodness

Yoyodyne – How a fictional company was born in the novels of Thomas Pynchon, was adopted by Buckaroo Banzai and Star Trek, and ended up in the GPL.

What will be left of our cities? – The nitty-gritty details of what will happen to our concrete, brick, and steel long after we're dead and gone.

On glitch art, and the fascinating mistakes computers make – I was a terrible VJ with footage, but I had so much fun with live feeds and static. Don't believe technology's mask of perfection, engineers knowwhat a rats' nest every product is under the hood.

Is MS Office the quiet villain of global finance? – Our kids will look back on the last couple of decades as a time when we fell under the spell of cold hard numbers, without really looking at how they were produced.

Search history and accidental class warfare – A variant of the echo chamber effect, and an example of the law of unintended consequences. Recommendation algorithms are becoming our century's version of press barons.

Do we need a slow software movement?

Photo by Tim Regan

When I was an isolated kid in the English countryside my only connections to the computing world were "Public Domain" floppy disks. Mail-order libraries would send me one of the disks in their catalog if I posted them a pound coin taped to a piece of card. I've never forgotten how important those glimpses into a wider world were, and I'll always be grateful to the people who made their demos, games, and utilities freely available. They were a lifeline to me, and I always wanted to give something back in return. My first contribution was a 'desktop palette' of 16 colors I'd selected for an especially pleasing RISC OS background, which didn't exactly set the world on fire.

That set the tone for most of my open source career – when I release a new project, I expect a deafening silence. There are occasional exceptions, but most of them don't make sense to anyone else, at least at first. The majority get quietly ignored by me and everyone else, but a few I keep working on, and they occasionally get picked up by other people too.

The Data Science Toolkit has turned into one of those sleeper projects. Over the last few months I've had a lot of bug reports, which is the best measure of how many people are actually using the code! There have been some nice companion projects too, like this wrapper for Excel or the new API library for Node. It also powers, which also keeps growing like a weed entirely through word of mouth. Hearing about the uses has been fascinating; academics of 19th century American literature mapping the spread of place mentions, reporters analyzing documents to track corruption in developing countries, mobile real estate app startups, university alumni associations.

The common thread for everyone using it is that they're marginal, just like I was growing up. There aren't enough of them and they don't have enough money to tempt commercial developers. Young open source software grows in the cracks between profitable problems, and survives on a starvation diet of spare-time coding. This gives it the time to find its niche, its audience, in a way that a more conventional development approach never could. Slow-growing software has the chance to reach people who'd never be found any other way, so if you're working on an unpopular project that you love, don't give up!

Five short links

Photo by Kurtis Garbutt

Geo-location estimation of Flickr images – The caption, title, and description of a photo is incredibly useful when it comes to guessing where a photo was taken, even using fairly crude language analysis algorithms. This is a great paper that parallels a lot of what we've found using unstructured text for image location at Jetpac.

Create a heatmap in Excel – Excel can be crusty and hard to learn, but I'm constantly surprised by how much you can do with it once you dive into its depths.

Death by a thousand paper cuts – I get asked the same questions over and over again by people I've just met once they detect my accent – "Where are you from?", "Do you like soccer?", "Why did you come here?". I appreciate that they're trying to connect with me, but the sheer repetition and predictability can make it hard to answer them with enthusiasm.

I can only imagine how tough it must be to deal with repetitive comments when people are behaving like jerks, rather than being nice. Julie does a good job explaining why, even when any single incident can seem fairly minor, an unending succession of them becomes impossible to deal with. The programming world tolerates people behaving like jerks in small ways towards anyone who isn't like them, over and over and over again.

Big Data and Conflict Prevention – The world ignored warning signs about famines and wars from small data, and they're doing the same thing with big data.

Helsinki Bus Station Theory – The case for sticking it through an apprenticeship so you can do something truly creative afterwards.

Converting to and from Google map tile coordinates in PostGIS

Google Maps' system of power-of-two tiles has become a defacto standard, widely used by all sorts of web mapping software. I've found it handy to use as a caching scheme for our data, but the PostGIS calls to use it were getting pretty messy, so I wrapped them up in a few functions. The code is up at, and here's a quick rundown:

tile_indices_for_lonlat(lonlat geography, zoom_level int)

Takes a PostGIS latitude/longitude point and a zoom level, and returns a geometry object where the X component is the longitude index of the tile, and the Y component is the latitude index. These values are not rounded, so for a lot of purposes you'll need to FLOOR() them both, eg;

SELECT FLOOR(X(tile_indices_for_lonlat(checkins.lonlat, 4))) AS grid_lon, FLOOR(Y(tile_indices_for_lonlat(checkins.lonlat, 4))) AS grid_lat FROM checkins;

lonlat_for_tile_indices(lat_index float8, lon_index float8, zoom_level int)

Does the inverse of the function above, turning a Google Maps tile index for a given zoom level into a PostGIS geometry point. You may notice that the coordinates are given as separate arguments rather than a single geometry object. That's an artifact of how my data is stored. Here's an example:

SELECT X(lonlat_for_tile_indices(6, 2, 4)::geometry), Y(lonlat_for_tile_indices(6, 2, 4)::geometry);

bounds_for_tile_indices(lat_index float8, lon_index float8, zoom_level int)

This takes latitude and longitude coordinates for a tile, and a zoom level, and returns a geography object containing the bounding box for that tile. I mainly use this for limiting queries on geographic data to a particular tile, eg;

SELECT * FROM checkins WHERE ST_Intersects(lonlat, bounds_for_tile_indices(6, 2, 4);

Five short links

Photo by Alan Levine

Elephant – A beautiful open source project to store data in a way that's "as durable as S3, as portable as JSON, and as queryable as HTTP". Tim O'Reilly has talked about the web operating system, and HTTP, JSON, and REST-like APIs (without the annoyances of full REST) have become the interface layer. I know integration will be do-able whenever I see a project based around them.

Median SF rent for a one-bedroom apartment – I wish Craigslist made their data openly available. It's already public, why not enable more useful services like this?

Everything We Know About What Data Brokers Know About You – The data about us that's used for marketing purposes is essentially unregulated. As someone who works with data about people for a living I'm glad I'm able to innovate, but I'm also depressed by how little the public actually cares about how their information is passed around and used.

The Design-Fiction Slider-bar of Disbelief – A corker of a listicle from Bruce Sterling, covering the continuum from imagination to regulation.

Scrapely – I love pulling data from messy HTML pages, and it's great to see more and more support emerging. Don't give me an API, just give me an open robots.txt.