What is Text Mining, NLP Analysis and the N-gram model?

TLDR

This article will discover the number of words that appear in “Sonnet,” and by analyzing high-frequency words, know Shakespeare’s intentions and preferences. Also, I will use part-of-speech tagging and n-gram models to improve our understanding of the article.

Source: Giphy

Outline

• Intro

• Preliminary preparation

• Text data preprocessing

• Split words

• Count the number of words in the article

• Stop words

• Calculate the frequency of the top 20 words

• Conclusion or Final thoughts

• Part of Speech Tagging

• Training The Model

• Closing Thought

Intro

Text mining is a large part of artificial intelligence. It helps computers understand, interpret, and manipulate human language. For example, how to analyze customers’ reviews for a product. How do chatbots mine for users' emotional information? How to extract the information people want from an article. This article analyzes Shakespeare’s “The Sonnets,” which can provide digital help for researchers of literature and history.

Source: Giphy

Text Data Preprocessing

Since there are many abbreviations and punctuation marks in the “Sonnets,” I need to restore these abbreviations and remove useless punctuation marks. The operation steps are as follows:

Regular Expression is a sequence of characters that form a search pattern. RegEx can be used to check if a string contains the specified search pattern.

Find specific patterns in string

Pattern substitution

• As shown in line 12, first load regular expression packages.
• As shown in line 13, 's is following the pronouns, such as it's.
• As shown in line 14, 's is following the noun, such as world's.
• As shown in line 15, ' is following the words ending by s, such as others', cars'.
• As shown in line 16, 't, such as doesn't.
• As shown in 17, 'd such as I'd.
• As shown in line 18, 'll such as he'll.
• As shown in line 19, 'm such as I'm.
• As shown in line 20, 're such as they're. As shown in line 21, 've such as I've.
• After that, there are some abbreviations in the source file.
• As shown in line 23, th'&Th' means "the","The".
• As shown in line 24, in medieval English, verbs followed by second person are usually followed by 'st', which has no special meaning. So it is directly removed when restored.
• As shown in line 27, 's following the pronoun is replaced by is.
• As shown in line 29, 30, 's indicates the possessive form, which is removed when restoring, but the verb prototype is retained.
• As shown in line 31, 't is replaced by "not".
• As shown in line 32, 'd is turned into "would".
• As shown in line 33, 'll is replaced by "will".
• As shown in line 34, 'm is turned into "am".
• As shown in line 35, 're is replaced by "are".
• As shown in line 36, 've is turned into have.
• As shown in line 37, 'th is replaced by the.
• As shown in line 39, 't is turned into 'it'.
• As shown in line 40, 'Tis' is replaced by "This is''.
• As shown in line 41, 'er' is turned into 'ver'. As shown in line 42, remove 'st'.

I don't consider lowercase and uppercase of words, uniformly converted to lowercase. Remove non-word characters.

Lowercase and remove non-English words

Result of Lowercase and remove non-English words

Split Words

Split the paragraph into words.

Result of split words

Count the number of words in the poetry

Circulate each word in “words, which is word after split.” if it is a word, count+1

Count the number of words in the poetry

Count the number of words after split, get the result:

The total number of words

Stop words

Stopwords are English words which don’t add much meaning to a sentence. They can safely be ignored without sacrificing the meaning of the sentence. For example, the words like the, he, have, etc. So I put these words in the txt file and tell Python to remove these stop words.

Read stopwords and get the new words.

Calculate the frequency of the top 20 words

Frequency of the top 20 words

Result of top 20 words

Here are the top 20 words in the poem “The Sonnets.”

Conclusion

In "The Sonnets,” Shakespeare repeats the word "love" 194 times, which shows that he likes to use this word very much, and it’s the main theme of the poem. Shakespeare depicts a multi-faceted image of love. Love is so powerful that it can defeat all obstacles. He believes true love and real love, like the stars can always be seen and never changing.

Source: Giphy

In "The Sonnets,” Shakespeare repeats the word "Time" 70 times, the second most common of all keywords. Shakespeare often personifies time. Shakespeare describes time as a tyrant and consumption of love and beauty. Shakespeare describes time as a “bloody tyrant” (Sonnet 16), “devouring” and “swift-footed” (Sonnet 19), “injurious hand” and “age' s cruel knife” (Sonnet 63). Time comes in many different forms, most of which are described as negative.

Source: Giphy

In "The Sonnets'', Shakespeare repeats the word "beauty" 70 times, the second most common of all keywords. Beauty, irrefutably, is a common theme throughout the Shakespearean sonnets. Beauty in Shakespeare’s Sonnets is represented in two dimensions: the physical beauty and the spiritual beauty. For example, he used the word "beauty" many times to describe the beauty of humans and the beauty of the heart.

Source: Giphy

Part of Speech Tagging

Parts of speech tagging simply refers to assigning parts of speech to individual words in a sentence. The parts of speech are nouns (NN), verbs (VB), adjectives (JJ), etc. For details, please refer to the part-of-speech tagging table.

I word_tokenize all the words after preprocessing named “files” and then put these into dict. And then pos tag these dict and name them as tag.

Part of speech tagging

Result as follows:

Result of tagging

The meaning of part of speech: The parts of speech are important because they show us how the words relate to each other. Now, they are just words, they don't really tell us something. But, as soon as we assign each word a role (a part of speech), and put them into a sentence, we actually get something meaningful: "touch my breast."

Train the Model

The Brown Corpus was the first million-word electronic corpus of English, created in 1961 at Brown University. Please note all the words in the Sonnets are put into the corpus as a whole data set, and we choose 90% of the whole data as a training set to train the model.

In this example, we are using DefaultTagger as the backoff tagger. Whenever the UnigramTagger is unable to tag a word, backoff tagger, i.e. DefaultTagger, in our case, will tag it with ‘NN’. DefaultTagger is most useful when it gets to work with the most common part-of-speech tag. That's why a noun tag is recommended.

Train the model

Unigram: Unigram tagger is a tagger that only uses a single word as its context for determining the POS (Part-of-Speech) tag.

Bigram: Unigram tagger is a tagger that only uses two words as its context for determining the POS (Part-of-Speech) tag.

Trigram: Unigram tagger is a tagger that only uses three words as its context for determining the POS (Part-of-Speech) tag.

If we didn't find the default tagger, then setting backoff. If one tagger doesn’t know how to tag a word, the word would be passed to the next tagger and so on until there are no backoff taggers left to check.

Train the n-gram model

Result of n-gram model

Unigram just considers the conditional frequency of tags and predicts the most frequent tag for every given token. Bigram is more accurate than the other two, However, the Unigram tagger has better coverage.

Closing Thoughts:

Since this is poetry, there is no need for prediction, we just need to know how to do lexical tagging. If we want to design automatic text filling, or speech prediction, then we can proceed to prediction. I’ll update my knowledge about the N-gram model and visualization of part-of-speech tagging in the next article, please stay tuned.