9,501 reads

Document-Term Matrix in NLP: Count and TF-IDF Scores Explained

by Bala Priya CJanuary 17th, 2022

Too Long; Didn't Read

In NLP, Document-Term Matrix (DTM) is a matrix representation of the text corpus. The TF-IDF score is widely used to populate the DTM. Read this tutorial to understand more about the Document-Term Matrix, and the TF-IDF score. And use scikit-learn to generate the Document-Term Matrix for your favorite piece of text!

featured image - Document-Term Matrix in NLP: Count and TF-IDF Scores Explained

Natural Language Processing (NLP) is a subfield of Linguistics and Computer Science that deals with processing and understanding huge volumes of unstructured text data.

To leverage NLP to analyze textual data, you’ll have to come up with a numeric representation of the text corpus, or the collection of documents. This tutorial will teach you all about Document-Term Matrix (DTM)—a matrix representation of the text corpus.

By the end of this tutorial, you'll have learned enough to understand the Document-Term Matrix representation, and the significance of the TF-IDF score. And you'll also be able to generate Document-Term matrices for any text corpus in scikit-learn.

What is the Document-Term Matrix?
What is the TF-IDF Score?
TF-IDF Score Equation
What is the Significance of the TF-IDF Score?
How to Generate the Document-Term Matrix in scikit-learn

What is the Document-Term Matrix?

In the context of NLP, a text corpus is simply a collection of documents. And a document can be a sentence, a group of sentences , or even a phrase—depending upon the use case.

Here’s a simple example of a corpus with only 3 documents

D1

D2

and

D3

D1: NLP is super cool to learn and apply.

D2: I am currently learning NLP, you can too!

D3: CS224n at Stanford is the best NLP class you can ever take!

Let's suppose the text corpus contains M documents, and N terms in all. The general structure of the Document-Term Matrix is shown below:

Document-Term Matrix (Image by the author)

Let's parse the matrix representation:

```
D1
```
,
```
D2
```
,...,
```
DM
```
are the
```
M
```
documents
```
T1
```
,
```
T2
```
,...,
```
TN
```
are the
```
N
```
terms.

One of the simplest ways of populating the Document-Term Matrix is using the number of occurrences of the

terms across all the

documents.

The entry
```
w11
```
denotes the number of times the term
```
T1
```
occurs in the document
```
D1
```
,
```
w12
```
denotes the number of times the term
```
T2
```
occurs in the document
```
D1
```
, and so on.
In general,
```
wij
```
denotes the number of times the term
```
Tj
```
occurs in the document
```
Di
```
.

This way, a term that occurs frequently in the text corpus will have a much higher score, and a term that occurs rarely in the text corpus will have a much lower score. The ordering and semantic relationship between the terms is lost.

Populating the Document-Term Matrix as shown above—with the number of times a term occurs in the documents is both simple and intuitive. However, are there any caveats you should be aware of?

Head over to the next section to find out.

What is the TF-IDF Score?

In populating the Document-Term Matrix with the number of occurrences, frequently occurring terms are assigned a higher score than the rarely occurring terms.

Now, take a step back, and ask yourself the question: "Is a frequently occurring term necessarily important?"

Well, this may not always be the case.

For example, when you're reading a piece of text on astronomy—you'll come across words like 'sky' and 'stars' very often. But, would you not like to see beyond these words to understand what the text is talking about?

This is exactly the motivation behind TF-IDF score—another useful metric that you can use to populate the Document-Term Matrix.

TF-IDF score is a combination of two metrics: the Term Frequency (TF) and the Inverse Document Frequency (IDF).

TF-IDF Score Equation

The TF-IDF score is given by the following equation:

where,

```
TF_ij
```
is the number of times the term
```
Tj
```
occurs in the document
```
Di
```
```
dfj
```
is the number of documents containing the term
```
Tj
```

Let's parse the above expression:

The first term,
```
TF_ij
```
is called the Term Frequency (TF). It is identical to the count we had in the previous section: the number of times the term
```
Tj
```
occurs in the document
```
Di
```
. So, the more frequently a term occurs in a particular document, the higher its TF score.

The second term is the Inverse Document Frequency (IDF).

Here,

•

is the total number of documents, and

•

df_j

is the number of documents containing the term

TF_ij

What is the Significance of the TF-IDF Score?

Let's first understand the significance of the IDF score.

If the term

Tj

occurs in all the documents, the value of

df_j = M

. This would leave you with

log(M/M) = log(1)

Recall from your school math: log(1) = 0 😄✅

On the other hand, if the term

Tj

occurs in only one of the

documents, you'll have

log(M/1) = log(M)

Notice how this is the maximum value that the IDF score can take.

Putting it all together:

A term that occurs frequently in a particular document has a higher TF score.
A term that occurs rarely across the entire corpus has a higher IDF score.
The TF-IDF score is the product of the TF and the IDF scores.

In simple terms, the TF-IDF score says, "Hey! I believe terms that occur frequently in a particular document are likely more important than terms that occur frequently in all documents in the corpus!"

Though the TF-IDF score fails to capture contextual relationships between terms in the corpus, it's still considered an effective metric and is preferred over counting only the number of occurrences.

Now that you’ve learned the basics of Document-Term Matrix and the TF-IDF scores, let’s now see how you can generate the DTM and populate it with the TF-IDF score in scikit-learn.

How to Generate the Document-Term Matrix in scikit-learn

1️⃣ Let's first get our text corpus

The following code cell contains a piece of text on computer programming—and this will be our corpus of interest.

text=["Computer programming is the process of designing and building an executable computer program to accomplish a specific computing result or to perform a specific task.",
      "Programming involves tasks such as: analysis, generating algorithms, profiling algorithms' accuracy and resource consumption, and the implementation of algorithms in a chosen programming language (commonly referred to as coding).",
      "The source code of a program is written in one or more languages that are intelligible to programmers, rather than machine code, which is directly executed by the central processing unit.",
      "The purpose of programming is to find a sequence of instructions that will automate the performance of a task (which can be as complex as an operating system) on a computer, often for solving a given problem.",
      "Proficient programming thus often requires expertise in several different subjects, including knowledge of the application domain, specialized algorithms, and formal logic.",
      "Tasks accompanying and related to programming include: testing, debugging, source code maintenance, implementation of build systems, and management of derived artifacts, such as the machine code of computer programs.",
      "These might be considered part of the programming process, but often the term software development is used for this larger process with the term programming, implementation, or coding reserved for the actual writing of code.",
      "Software engineering combines engineering techniques with software development practices.",
      "Reverse engineering is a related process used by designers, analysts and programmers to understand and re-create/re-implement"]

2️⃣ The next step is to import the

 TfidfVectorizer

class from scikit-learn's feature extraction module for text data. This

TfidfVectorizer

helps generate the Document-Term Matrix populated with the TF-IDF score.

from sklearn.feature_extraction.text import TfidfVectorizer

3️⃣ Let's instantiate a

Tfidfvectorizer

object. Let's call it

vectorizer

 
vectorizer = TfidfVectorizer(stop_words='english',smooth_idf=True)

▶️ Now, call the

fit_transform()

method on the

vectorizer

with

text

as the argument.

input_matrix = vectorizer.fit_transform(text)

print(input_matrix)

 # [truncated view]
 # sparse representation of the DTM
 # non-zero indices and TF-IDF score  
  (0, 83)	0.21527315443847012
  (0, 55)	0.25487697925239533
  (0, 73)	0.25487697925239533
  (0, 20)	0.25487697925239533
  (0, 80)	0.5097539585047907
  (0, 1)	0.25487697925239533
  (0, 63)	0.21527315443847012
  (0, 33)	0.25487697925239533
  (0, 11)	0.25487697925239533
  (0, 27)	0.25487697925239533

As you can see, the

input_matrix

has the scores, and the indices corresponding to the non-zero entries only—and not the entire matrix.

To cast it into a matrix of dimensions

M x N

, you can use the

to_dense()

method, as shown in the code snippet below:

input_matrix = vectorizer.fit_transform(text).todense()

# Truncated view of the entire matrix

[[0.         0.25487698 0.         0.         0.         0.
  0.         0.         0.         0.         0.         0.25487698
  0.         0.         0.         0.         0.         0.
  0.         0.37434759 0.25487698 0.         0.         0.
  0.         0.         0.         0.25487698 0.         0.
  0.         0.         0.         0.25487698 0.         0.
  0.         0.         0.         0.         0.         0.
  0.         0.         0.         0.         0.         0.
  0.         0.         0.         0.         0.         0.
  0.         0.25487698 0.         0.         0.         0.18717379
  0.         0.         0.         0.21527315 0.         0.13251329
  0.         0.         0.         0.         0.         0.
  0.         0.25487698 0.         0.         0.         0.
  0.         0.         0.50975396 0.         0.         0.21527315
  0.         0.         0.         0.         0.         0.
  0.         0.         0.        ]
 [0.         0.         0.22103745 0.         0.56007525 0.22103745
  0.         0.         0.         0.         0.         0.
  0.         0.22103745 0.         0.18669175 0.         0.22103745
  0.         0.         0.         0.         0.22103745 0.
  0.         0.         0.         0.         0.         0.
  0.         0.         0.         0.         0.         0.
  0.         0.22103745 0.         0.         0.16232309 0.
  0.         0.         0.         0.22103745 0.         0.22103745
  0.         0.         0.         0.         0.         0.
  0.         0.         0.         0.         0.         0.
  0.         0.         0.22103745 0.         0.         0.22983952
  0.         0.         0.22103745 0.         0.         0.
  0.22103745 0.         0.         0.         0.         0.
  0.         0.         0.         0.         0.         0.
  0.18669175 0.         0.         0.         0.         0.
  0.         0.         0.        ]