Skip to main content

So how much difference do stopwords make in a Naive Bayes text classifier?

POPFile contains a list of stopwords (words that are completely ignored for the purpose of classifying messages) that I put together by hand years ago (if you are a POPFile user they are on the Advanced tab and called Ignored Words). The basic idea is that stopwords are useless from the perspective of classification and should be ignored; they are just too common to provide much information.

My commercial machine learning tools did not, until recently, have a stopwords facility. This was based on my belief that stopwords didn't make much difference: if they are common words they'll appear everywhere and probabilities will be equal for each category of classification. I had a 'why bother' attitude.

Finally, I got around to testing this assumption. And I can give some actual numbers. Taking the standard 20 Newsgroups test and using a (complement) Naive Bayes text classifier as described here I can give you some numbers.

The bottom line is that stopwords did improve classification accuracy for 20 Newsgroups (and for other test datasets that I have), but the improvement was by 0.3%. The stopword list used was the top 100 most frequently occurring English words.

So, I was wrong... they do make a difference, but not by much.

Comments

Justin Mason said…
Where'd you get the list of stopwords?

What I did for the SpamAssassin classifier was to train on a "sample" spam/ham corpus, then select the tokens between 0.4 and 0.6 -- the wishy-washy tokens that spamassassin would ignore anyway (we ignore tokens that are not "strong" enough for classification). so in other words, the stopword list was derived from training, rather than a "500 common english words" list.

Popular posts from this blog

Your last name contains invalid characters

My last name is "Graham-Cumming". But here's a typical form response when I enter it: Does the web site have any idea how rude it is to claim that my last name contains invalid characters? Clearly not. What they actually meant is: our web site will not accept that hyphen in your last name. But do they say that? No, of course not. They decide to shove in my face the claim that there's something wrong with my name. There's nothing wrong with my name, just as there's nothing wrong with someone whose first name is Jean-Marie, or someone whose last name is O'Reilly. What is wrong is that way this is being handled. If the system can't cope with non-letters and spaces it needs to say that. How about the following error message: Our system is unable to process last names that contain non-letters, please replace them with spaces. Don't blame me for having a last name that your system doesn't like, whose fault is that? Saying "Your

All the symmetrical watch faces (and code to generate them)

If you ever look at pictures of clocks and watches in advertising they are set to roughly 10:10 which is meant to be the most attractive (smiling!) position for the hands . They are actually set to 10:09.14 if the hands are truly symmetrical. CC BY 2.0 image by Shinji I wanted to know what all the possible symmetrical watch faces are and so I wrote some code using Processing. Here's the output (there's one watch face missing, 00:00 or 12:00, because it's very boring): The key to writing this is to figure out the relationship between the hour and minute hands when the watch face is symmetrical. In an hour the minute hand moves through 360° and the hour hand moves through 30° (12 hours are shown on the watch face and 360/12 = 30). The core loop inside the program is this:   for (int h = 0; h <= 12; h++) {     float m = (360-30*float(h))*2/13;     int s = round(60*(m-floor(m)));     int col = h%6;     int row = floor(h/6);     draw_clock((r+f)*(2*col+1), (r+f)*(row*2+1),

The Elevator Button Problem

User interface design is hard. It's hard because people perceive apparently simple things very differently. For example, take a look at this interface to an elevator: From flickr Now imagine the following situation. You are on the third floor of this building and you wish to go to the tenth. The elevator is on the fifth floor and there's an indicator that tells you where it is. Which button do you press? Most people probably say: "press up" since they want to go up. Not long ago I watched someone do the opposite and questioned them about their behavior. They said: "well the elevator is on the fifth floor and I am on the third, so I want it to come down to me". Much can be learnt about the design of user interfaces by considering this, apparently, simple interface. If you think about the elevator button problem you'll find that something so simple has hidden depths. How do people learn about elevator calling? What's the right amount of