The Tao of Text Normalization

Why bother?

Texts documents are noisy. You will realize it brutally when you switch from tutorial datasets into real world data. Cleaning things like misspellings, various abbreviations, emoticons, etc. will consume most of your time. But feature processing step is crucial to provide to quality samples for later analysis.

This article will provide you a gentle introduction to some techniques of normalizing text documents.


We will discuss a couple of techniques that can be immediately used. The plan for the following sections is as follows:

  1. Basic processing
  2. Stemming
  3. Lemmatization
  4. Non-standard words mapping
  5. Stopwords

For experimentation purposes, an environment with Python 3.5 and NLTK module is used. If you have never used in before check this first.

All examples assume that basic modules are loaded, and there is a helper function capable of presenting the whole text from tokens.

As a playground, a review of new Apple Watch 2 is used (from Engadget).

Basic processing

Feature processing will start with bringing all characters into lowercase, and tokenize them using.RegexpTokenizer In this case, the regexp - \w+ will extract only word characters.

For all the consecutive examples we assume that the text is processed this way.


Stemming is the process of reducing inflected (or sometimes derived) words to their word stem, base or root form—generally a written word form. The stem need not be identical to the morphological root of the word; it is usually sufficient that related words map to the same stem, even if this stem is not in itself a valid root. ~ Wikipedia

NLTK introduces several stemmers i.e. Snowball, Porter, Lancaster. You should try them on your own and see which one will be best for the use case.


Lemmatisation is the algorithmic process of determining the lemma for a given word. ~ Wikipedia

In NLTK you can use built-in WordNet lemmatizer. It will try to match each word to an instance within a WordNet. Mind that this process returns a word in its initial form if it cannot be found, and is much slower than standard stemming.

Non-standard words mapping

Another normalization task should be to distinguish non-standard words - for example, numbers, dates etc. Each such word should me mapped to a common value, for example:

  • Mr, Mrs, Dr, ... → ABR
  • 12/05/2015, 22/01/2016, ... → DATE
  • 0, 12, 45.0 → NUM
  • ...

This process allows to further easily summarize a text document and to derive new features (for example: count how many times a number appears).

Stop words removal

Stop words usually refer to the most common words in a language. They do not provide any informative value and should be removed. Notice however that when you are generating features with bigrams, stop words might still provide some useful insights.

There are built-in lists for many languages that you can use (or extend).

Let's see how a lemmatized version with removed stop words looks like:


Pre-processing text data makes it more specific. It's get cleaned from things human consider important but does not provide any value for machines. Very often a positive byproduct of normalization is the reduction of potential features used in a later analysis, which makes all computation significantly faster ("curse of dimensionality"). You should also keep in mind that some of the data is irreversibly lost.

So you are eating healthy... Oh, really?


I bet you all heard that in order to stay fit you should consider eating 5 meals per day. This roughly means eating every 3 hours!

Inspired by a talk given by Tim Ferris, I decided to conduct a conscious experiment to track each meal I was consuming. Just for fun.

It all took about 2 months to complete, but the outcomes are very thought to provoke. I got acquired with the brutal truth about myself.

What's more interesting - the experiment is fully repeatable. At the end of the post, I will give you some Python scripts that will be helpful to replicate the whole process and obtain your own personalized results.

Let's begin.

Collecting data

First and foremost you need your some data. I have used a DietSnaps app. It's purpose is to take a photo of each consumed eaten meal. You can get it from the AppStore.

Even though the app provides an option to export all data (i.e. CSV file), I decided to take a manual approach. Each dish was labeled using the following categories:

  • was it healthy (rather yes / rather not)
  • alcohol (yes / no)
  • meat included (yes / no)
  • vegetables included (yes / no)
  • fruits included (yes / no)
  • fast-food (yes / no)
  • full-meal (yes / no)
  • form (mostly raw / mostly processed)

and put into Google Docs Spreadsheet - link.


The first thing I wanted to know is how much beer I drink each day. Let's try to visualize it with the following plot - average number of meals and beer bottles consumed per each weekday.

Oh, that interesting. I would take a bet that most beers are drunk on the weekend - but ... it's Wednesday (exhausting middle of the week). On the other hand - my training days are Mondays and Thursdays (less consumption). Hopefully, I was eating more these days.

Let's proceed with another question.

The whole experiment took nearly 8 weeks. The fact of taking photos of each meal has obviously made me more conscious about the quality of food. I should be eating better with each meal, right?

Everything was going well until week 3. After that time fast-food consumption was continuously growing. The overall number of meals with fruits is also very depressing.

"What gets measured gets managed" ~ Peter Drucker

Maybe there were little fruits and vegetables but the dishes were overall quite healthy. I can calculate some proportions for each day (green color means super-healthy eating, red - mega-unhealthy).

Mondays tend to be healthier than other days (new week begins with extra powers). Tuesdays and Thursdays are also quite ok (due to workouts). There are also some bad periods - see last three days of the fourth week. Awww.

Finally, let's try to answer how often do I eat? Am I following a rule of "meal every 3 hours"? To visualize we will use a great concept of a time-map (you can read more about it here).

Time-map is very good for recognizing how do events relate each other in time (are they occurring fast or rather slow). Each event is plotted on XY plot, where axes show time before previous, and after next event.

And this is where the drama starts.

All of the plots are fully interactive. If you zoom-in to the purple rectangle you will see how many meals were eaten in a healthy fashion.

It turns out that I have eaten roughly FIVE meals that were introduced and followed with about 3h break. It's exactly 1.86% of all meals. How the hell I was supposed to build muscle if only 1.86% of all meals during 8 weeks were consumed properly?

Try it

You might be thinking that you're living good. But these beliefs should also be verified from time to time.

Painful truth: numbers don't lie.

Plotting the results is useless if it is not followed by understanding the data and coming with some action to make a change.

If you are curious about your own performance feel free to use this Jupyter Notebook. It will generate all of the plots presented above for you.

Stay strong.

Hypothesis Testing for Gangsters

Okay. Okay. OKAY. Look. I know you have a problem. You've been screwed by someone and now want your money back. Totally agree.

But first take a big breath and relax - you don't want to get into bigger trouble. Let's do it another way. I want to help to go one step further and do it like a PRO. And believe this makes a huge difference.

So go, grab your drink, and read this 5 tips.

How to do this

Read each step carefully. After the end, you will find what should you have after accomplishing it.

  1. Formulate hypothesis you want to validate
    A null hypothesis (H_{0}) is a statement we want to validate. Unless we will find sufficient evidence, there will be no reasons to reject it.

    A drug dealer states that cocaine is pure in 90%.

    A null hypothesis is (H_{0}\colon\ p = 0.9)An alternative hypothesis (H_{1}) is a statement that automatically becomes "true" (not rejected) if null hypothesis gets discarded.

    A customer doubts drug's purity. He states that it contains more than 10% additives. The alternative hypothesis can be (H_{1}\colon\ p < 0.9)

    After this step, you should have formulated (H_{0}) and (H_{1})

  2. Choose test statisticsOur overall aim is to validate the null hypothesis. We have to assure that it is true and then look for arguments to demolish it. Yeah.In more scientific speech we have to come up with probabilistic distribution ensuring that null hypothesis is correct.

    A customer bought 15 decks of a drug. After hosting a big party he realized that ONLY 11 decks were meeting the norm guaranteed by the dealer (test statistics). Remembering the wise words of a dealer, his test distribution can be (X \sim B(15; 0.9)). Someone will have a problem.

    After this step, you should have figured the test statistics (based on the experience) and the test distribution

  3. Choose a critical region (one-tail or two-tail test)Right now we have our probability distribution of test statistics, but still need to choose which values the null hypothesis get rejected (critical region) and for which accepted (acceptance region).We use a term of significance level (\alpha) which is a parameter describing certain probability, that for an event the likelihood of it's occurrence is small enough to agree that the null hypothesis gets rejected.

    A customer have chosen a value of significance level (\alpha = 5\%) meaning that the critical region (when we reject the null hypothesis) can be described as: (P(X < c) < 0.05)

    Depending on the form of (H_{1}) we can also specify whether the critical region is one-tailed or two-tailed.

    One-tail critical region occurs when the alternative hypothesis is expressed with inequities. For example if (H_{1}\colon p <\ c ) we should use left one-tailed critical region, and for (H_{1}\colon p >\ c right one-tailed.

    When the (H_{1}) is expressed with the (\neq) sign we are dealing with two-tail critical region. In this case, the critical region is placed in both tails of the distribution, where each side corresponds to the (\frac{\alpha}{2} ) probability.

    Because the alternative hypothesis is (H_{1}\colon\ p < 0.9 ) the scammed customer is dealing with one-tailed critical region.

    After this step, you should specify the significance level (\alpha ) and know whether the critical region is one-tailed or two-tailed.

  4. Calculate the probability (p-value)P-value is a probability of getting the same (or worse) results from the perspective of a null hypothesis.It's value depend on two things:
    • form of alternative hypothesis (H_{1}) (one or two tails),
    • a value of test statistics (based on the test distribution)

    In the case of our customer the test statistics is 11 (doses of pure drugs) and the critical region is located in left tail. The formula for p-value is ( P(X < 11) ). Taking into consideration (X \sim B(15; 0.9) ) it's value is (P(X < 11) = 0.55). To calculate this he used this snippet.

    After this step, you should obtain p-value

  5. Make a decisionIn this last step, we are finally deciding if the null hypothesis gets rejected or not (i.e. dealer was right or not).The null hypothesis will get rejected if the p-value will get into critical region.For example if the critical region is in the left tail the (H_{0}) will get rejected if ( \alpha < P_{value}).

    Customer has to reject his hypothesis (H_{1} ). In this case the P-value (( P_{value} = 0.055)) is greater than the significance level ( \alpha = 0.05 ), which means that the drug dealer was right ( H_{0}) is true). DAMN.

    After this step you finally know if there are reasons to reject (H_{0})


Question: What value of significance level should I choose?

Answer: It all depends on how sure you want to be that you are making no mistake when rejecting a null hypothesis. For example, choosing ( \alpha = 1\% ) gives you more certainty that your decision about rejecting ( H_{0}) was correct than ( \alpha = 5\%).


I have to admit it. I'm a bit scared. You have received a powerful tool. Tool that help to prove you that you're RIGHT in many cases.

But please, remember about other that still might need some help. Share it with them, and make them your debtors.