Sentiment Analysis 🙇🏻‍♂️🥴

Graph based Mining was fancy, but on hiatus for now, until I figure out how to import from CSV without that NULL error

How does sentiment analysis work, generally?

TextBlob

TextBlob is a textual data processing library for diving into common Natural Language Processing (NLP) tasks such as part-of-speech tagging, noun phrase extraction, sentiment analysis, classification, translation, and more.

This was funny because PUNKT is a Sentence Tokenizer

1. Pankti in Hindi actually means a sentence (literally ‘line’)!
2. Punk as in ‘Daft Punk’ 🧛‍♂️
3. A friend’s name is Pankti and she was surprised to find out about this

SaaS (Sarcasm as a Sentiment):

“I am so happy that not one thing happened good today”

The polarity ranges from : (-1 to 1)

-1 = Totally Sad

0 = Neutral

1 = Totally Happy

Here, the result is 75% confident that it’s a happy statement. 🤦🏻‍♂️

Part Of Speech (POS) Tagging

Here are some Tag Abbreviations I found from the NLTK documentation:

Basic:

• N Noun
• V Verb
• ADJ Adjective
• ADV Adverb
• P Preposition
• CON Conjunction
• PRO Pronoun
• INT Interjection

Advanced

• CC coordinating conjunction
• RB adverbs
• IN preposition
• NN noun
• JJ adjective
• VBP present tense verb

I made up the sentence above to demonstrate that the same word can be a Verb and a Noun in the same sentence !

✅ Test NLTK Part Of Speech (POS) tagging

Polarity — Positive vs. Negative

Here are some words I tested

✅ Sentiment Analysis using TextBlob

Inference

Positive = Happy

Negative = Sad

The script didn’t execute the next day as in some cases MusixMatch didn’t return the lyrics (I’m assuming they didn’t have it yet). In which case it sends an empty list instead of dictionary. So that was breaking my script (even though it worked fine before, as it had lyrics for all top songs in india of that day’s charts). So I had to put a condition to make sure the script doesn’t fail.

✅ Fixed script breaking when Musixmatch doesn’t have lyrics

I multiplied the scores with 100 to get percentage:

✅ Sentiment polarity to Percentage

There were times when the polarity was 0, meaning ‘Neutral’. I think that information is not helpful… I’m not showing lines and songs that don’t yield a sentiment

✅ Only showing lines with non-zero sentiment

Then I removed the negative sign from sad ones:

✅ Sad percentage value should be absolute, not negative

Language Detection

I was surprised by how accurately I could detect the language (using Google Translate).

The song ‘Putt Jatt Da’ is actually in Punjabi, but I’ll let this one slide for now.

‘Taki Taki’ is Spanish. That’s right.

Believer is in English. So that is correct.

I’m a Gujarati (demonym of Gujarat, India). I know Gujarati. And that up there, is not Gujarati. Proper Patola is a Punjabi song. Google Translate is still dodgy with Punjabi.

This is my favourite example. ‘Kamariya’ is actually a Garba song (traditional dance of Gujarat). But this song, is Hindi film adaptation. So it has both: Gujarati and Hindi. So detecting it as either is good enough; I’m happy it detected it as Gujarati. I’m impressed that it detected Gujarati inside a Hindi song.

✅ Test Language Detection

Comparing with Labels from API:

I recalled that I am getting the language code from the API for each track already.

Hybrid approach for Language Detection:

I’m now verifying the same with the Language Detection and only Analyzing songs that are in English according to both methods.

✅ Restrict mining to english by detecting the language and verifying the detection with labeled data from API

I will now work on finding all the ways to compare environmental factors with lyrics in English language.

Good News: Shout out to Andrew Trask

Andrew Trask, my role model, liked my tweet ! Hurray !

Devansh Trivedi on Twitter

100DaysOfMLCode #100DaysOfCode #100DaysOfX Day 46: ✅ Restrict mining to English by detecting the language and verifying the detection with labeled data from API https://t.co/Io3EQbMUi6 https://t.co/iEU6Dr7xBE https://t.co/1PpmSelZWr #PoweredByMusixMatch

Sir Trask is building OpenMined — a platform that merges cryptographic techniques such as homomorphic encryption and multi-party computation and blockchain technology to create the ability to train ML models with private user data. OpenMined will allow AI companies of the future to develop models, have them trained on user data without compromising user privacy, and incentivise users to train their model.

He is a PhD student at Oxford University where he researches Deep Learning approaches with special emphasis on human language. Get to know him better in this video:

Sentiment Polarity Comparison:

TextBlob contains two sentiment analysis implementations:

1. PatternAnalyzer (based on the pattern library)
2. NaiveBayesAnalyzer (an NLTK classifier trained on a movie reviews corpus)

✅ Naive Bayes Tokenization and comparison with Pattern Analyzer

I would tolerate percentage difference, but in some cases the polarity is totally opposite. I will need to listen these songs, read lyrics and manually verify if the sentiments are consistent with the algorithms (which can be subjective).

Types of combinations:

1. NaiveBayes: +ve & Pattern: +ve
2. NaiveBayes: -ve & Pattern: -ve
3. NaiveBayes: +ve, Pattern: -ve
4. NaiveBayes: -ve, Pattern +ve

Using IBM Watson and BlueMix Console

Alchemy is now Watson Natural Language Understanding

✅ Use @IBM Watson from BlueMix Console for Natural Language Understanding of ‘Entities and Keywords’

I will need more time to go through the documentation for taking full advantage of Watson’s services.

I did manage to get some keywords and entities from Starboy song lyrics using the new API. However, there’s a lot left to learn and integrate with the current script, so IBM code is in a separate script for now.

Code:

• D3V4N5H/StarBoy
• D3V4N5H/StarBoy

More:

D3V4N5H/100DaysOfMLcode

New, Frontiers, Ahead, !

CSV (Cockroach-like Standard Vernacular)

After my struggle to circumvent storage of data into files, the original query requires fetching data line-by-line from a CSV (Comma, Separated, Values) file

I compare it with cockroaches because it existed before me (like they existed before humans).

This is epic. I’m making AI that will handle the data in a way that PRE-DATES the way my computer is made. Like using ANCIENT tools to build something modern.

Come what may, the AI will do whatever it takes !

✅ Exported lyrics into a CSV

To quote my last week’s article,

This is because of the way I’m fundamentally handling the lyrics from API

Fortunately, I’m a Data Scientist. Computer Engineers are wizards. They practice magic for a living:

That’s not quite it. Let me use the other wand…

Wait, did you see those BLANK lines? We need to nitpick the non-empty lines.

✅ CSV export is tip top, no blank lines now

OMG! Those ‘(Ha-ha-ha-ha)’ are still there! We don’t want our baby-AI to be confused. We know it doesn’t mean anything. So I’ll get those out of the way…

✅ Improved Lyrics cleaning using loops

Yep, now that’s what I’m talking about. Good job, Dev (Pun intended)

But I had to manually copy it to the ‘import’ directory of the project’s graph to create nodes

Before Removing Empty Lines

After Removing Empty lines

Wherever you go, there you are !

save_path = '/Users/d3v4n5h/Library/Application Support/Neo4j Desktop/Application/neo4jDatabases/database-25eef966-4d97-4421-9310-903f03421892/installation-3.4.6/import/'

✅ Export Starboy lyrics CSV directly to the path of graph import directory

Using the Devil you know (CSV)

Mind that CSV is awesome, it has never done anything wrong unto anyone (or any thing). I’m just thwarted because I don’t see a way to circumvent it’s NEED by modifying the CYPHER query.

Later, I used this file to Create Nodes and mine Paradigmatic Associations using Jaccard Index.

✅ Mine the lyrics imported from CSV
✅ Store the result in a Python dictionary

Voids in the Valves of the Venture

Found ‘0’ values for words… My guess? 🤔 It’s the first words of some lines.
I checked the CSV file, and it didn’t have any ‘0’ in it !

I will investigate this matter if and when it turns out to be an issue. Don’t get the wrong idea, I’m not being sloppy giving the AI any leeway to give notorious recommendations.

Scaling the Success

Let’s do this for Top Tracks in india !

Approach:

create_Nodes(song_Name) → 
graph.create_Nodes(tx, song_Name) → 
create_Nodes_Query_Fn(song_Name)

Here, tx = Transaction

My code is right, but the Database throws Illegal Character error for songs that have SPACE in their name.

It turns out, Encoding didn’t help either, I just had to strip the SPACE characters out of the song name.

✅ Fix file encoding issue

I removed the last empty line in the CSV manually and checked for other NULL values with Python. But the error persists…

Starboy lyrics had null and still worked fine

And so now I suspect that “the Query is the Culprit” here

Code:

• D3V4N5H/StarBoy

More:

Devansh's #100DaysOfMLcode

The Making of Flubber Weebo for songs!

Engineering Repercussions

Giving Camels some Comfort in the Crowd

I underscored variable names for readability

They were camelCase before

now_They_Are_Both !

Lyrics Gone MAD

Array in a string?

‘[0] =>’ ?

This all looks weird. This isn’t what I was promised when I signed up for the APIs. I’m shocked yet amazed cuz it kinda looks cool!

De-Duplication of Nodes

I implemented query and calls with a different approach to prevent duplicate nodes before Mining

This “different approach” is using negative indexing and slicing to convert the lyrics into two separate list of words; yielding perfect word pairs

Finally it works as intended, can’t wait to fetch and display keywords mined from each song now !

Ghosts of Innovation

Importing weighted keywords from mined graph. The script is taking significant time to execute, which is kinda cool, I can take a walk, drink some water and get fresh air every time I test my script !

But Alas! I can’t “HAVE IT”. Trying different approaches to read Response code from the BoltStatementResult object. I really wish I could just iterate on it 😤

Ok, I got it. But then the ALL of the weights were 1. This means my AI is very unbiased ☯️☯ and doesn’t understand the first thing about English language🤯

This is not Mined, this is MindLess 🤪

Silver Lining: New Top Tracks!

SpiderMan Sunflower by Post Malone and Swae Lee is here! Of course!

Overt Overhaul

I’m DONE. I tried modifying the CYPHER query to fit the data. Now I’ll modify the input for the query to fit the original query. 😑

Forget the whole script. I want to make this happen for Starboy song first. Will loop the whole process when I finish mining this one as expected.

Look at the CSV I manually made at the initial stage of the project:

After several minutes of coding, I got to:

And

This is because of the way I’m fundamentally handling the lyrics from API. I don’t get separate lines, but perhaps I can use ‘\n’ as delimiter to split lines into list of line elements.

Code and Notes:

• D3V4N5H/100DaysOfMLcode
• D3V4N5H/StarBoy
• D3V4N5H/StarBoy

MORE :

Devansh's #100DaysOfMLcode

The Making of Flubber Weebo for songs!

Failure and Support

Unhealthy Boilerplate

I wanted to import the words from the JSON, but the queries I worked on before, used CSV file. Luckily, I had parsed the data into a neat dictionary. But now I needed to pass lyrics from it to the Graph.

At first, this seemed easy. But later I noticed it didn’t work. I even tried to code another query to pass word pairs instead of whole lines. For this, I needed to iterate TWO value in a list at the same time.

So I tried to find a way to do that, but that didn’t do quite as expected:

iterating over two values of a list at a time in python

It had to take the second word again next time!

So I implemented it the old school way:

For the first time, I couldn’t do something, despite my 4 Hours of Hard Work. I felt disappointed and stuck !!

But thanks to Michael Hunger, we got this beautiful graph!

As I was wondering why the word frequency from the API fetched lyrics didn’t match that of the CSV lyrics, it was comforting to recollect that I don’t get FULL lyrics from the free account. Hence the difference.

Despite that, my mining was very accurate if you compare the following to the mining from original FULL lyrics

Bypassing Label limitations using APOC

Currently, we’re only able to see numbers on nodes. I’d like to know

1. What word that node represents,
2. Which song is that cluster of nodes for !

Adding a label with the word as value was easy:

But Song Names often had multiple words, which at the time seems to be a technical limitation of Cypher.

Another Day, Another Angel

Mark Needham, Developer Relations Engineer at Neo4j, suggested using something interesting:

APOC: Awesome Procedures On Cypher

There’s actually a fun story about the acronym. Apoc was the technician and driver in the Matrix movie !

The APOC library consists of about 450 procedures and functions to help with many different tasks in areas like data integration, graph algorithms or data conversion.

And that helped!

So I made a loop to do this for all the Top Tracks in India:

1. Get name of Each Song
2. Get lyrics of that song
3. Clean the lyrics
4. Create Node Pairs
5. Log when all the words of song’s lyrics are merged into nodes of the graph
6. Mine only those nodes in the graph
7. Log that the song was mined
8. Get weighted keywords from the graph
9. Change Label to Song Name using APOC

Labelling songs by their names allowed me to mine each song individually, as opposed to a graph where nodes from words of all songs were pooled.

When AI is drunk, and drives under influence

Even though I added UNIQUE keyword in the query, at least two nodes were created for each word, as the query was run for the next word pair (see 2nd picture in the article for reference).

Interesting note:

Most weighted pairs are the same words!

What I tried but didn’t work:

1. Defining the query AGAIN at every iteration in the loop, concatenating each word pair into it (hard coding it explicitly)
2. Passing lines instead of words with Field Terminator as a single horizontal white space character
3. Using Unwind to access multiple neighbouring values in the line (it seems I can only access and iterate over one element)

Code

D3V4N5H/StarBoy

Making it all work: RECONnaissance

Then I went a step ahead and found the top songs OF top artists. That’s sooo… META!

Getting Release Date METADATA

After parsing the Dates and Formatting:

Breaking news

If you sing Honey Singh songs, your choice is not so bad as a lot of Indian Music Users prefer it too. He’s in the top artists!

Reducing the time of execution

This META loop was taking too long for the script to give output.

I found the data in previous calls

So I removed the function I was using to get Release Date of Each Track of Each Popular Artist, and coded a function to iterate over data from existing calls.

Now I will be able to find out the Trending Tracks FASTER!!!

UPS and DOWNS

(That was not a pun). Using driver is a pain when debugging.

Getting the Lyrics

I fetched lyrics of Top Songs in #India songs to train the AI

Then I fixed it to read clearly and parse it

For AI, I will need to clean this and process further for WHITESPACES like empty lines and “…” and the FREE API notice before I generate keywords.

I DID IT!

Getting lyrics is now officially automated !!

Genius.com resources helped understand parts of the lyrics

The data in my file is analogous to the CSV file that I manually generated at the beginning of the project.

Then I realised that the background vocal music in the brackets is not really contributing in the teaching AI what the lyrics means

So I got rid of it from the lyrics and also removed commas ( , ).

By now I have managed to summarise the song. I can find ‘key’ words from a song, along with their weights.

But understanding the lyrics is not how we choose our songs, do we? Sometimes, we just like the artist; or the genre (like Punjabi Beats), or the topic (like patriotic songs); or its timing (like songs from your childhood).

Lo and Behold the :

“Song Attribute Graph Data Model”

I’m advocating content based recommendations at this stage of the project. I had to guess what parameters can be taken into account to find patterns that match the environment. To keep my project simple, I have chosen Release Date, Featuring Actor, Genre AND Keywords.

Remember, I’m attempting Content-based Recommendation first. That means tracking the user’s preferences over time. This will work better over time. It grows on you, just like Music!

Bring the Data over here!

It took long enough to get ahold of these, but it was worth the wait!

What I CAN do:

I’m planning to use the following to compute relevance:
chart.tracks.get
chart.artists.get
matcher.track.get
track.get
artist.albums.get
matcher.lyrics.get
track.lyrics.get

What I CAN NOT do:

I’d need a commercial plan for the following:

track.search

track.subtitle.get

matcher.subtitle.get

The PlayGround

MusixMatch has an API Playground. Authenticate using a key to test responses.

track.snippet.get didn’t work for starboy track_id 144134659, not even in the playground 😭. “snippet_body” was empty! They don’t have it. 🤬

Word to the Wise

Show me the CODE !!

I know. Almost everyone who reads my articles wants to know what I did. Here’s how I did it:

I’m leveraging local insights. You location plays a part in defining your environment.

Here is the response for top artists in US:

So top songs from one of these artists is a good idea for a US user:

Side note: Look what I found

Subtitle of this Article

Now I will combine both python files : The one that fetches data and the one that fires queries to generate nodes in the graph database to associate data.

All that fancy JSON is not going to understand itself. We need to add this to the ‘Song Attribute Graph’. So I need a function to do what I did: Add the nodes with the data. Like this query:

I wanted to create separate functions for queries to add user, add song, add metadata and link. But because of the limitation of query variables, their scope only exists with the query. So I had to create a common function. If I still wanted to create separate ones, I’d need to MATCH/MERGE them before making associations to a new variable, which would further complicate the code. I will have a workaround in Python to remedy this by creating query templates and loops.

So I did that for a bunch of users with just the Artist relationship.

And then some.

I divvied up the graph above into following upper and lower halves, so you can read the data inside.

…And

No!! You’re not going to believe this!

Today I met a Music Studio Owner 🎤 — 🎧, Sir R. K. Bhanushali. He was surprised to find all my response data had only English songs and none in Hindi.

The algorithm should

1. Find which languages are prominent at user’s location and whether or not user knows / likes songs in that language
2. Find the code for that language to use in the API call
3. Add top tracks in each language to the Graph

Luckily, I don’t need to do all that if I use MusixMatch charts

Now that I have the language agnostic recommendations, I will fetch their lyrics, snippets and metadata. I need more metadata, as I couldn’t get some info about tracks from MusixMatch API.

Links to work:

• johnymontana/nlp-graph-notebooks
• Create UserPreferenceAndSongAttributesGraph.cql · D3V4N5H/StarBoy@b48685d
• Create User Preference and Song Attributes Graph · D3V4N5H/StarBoy@2dd791e
• D3V4N5H/StarBoy
• Day 11: Names of Top Artists in India from JSON · D3V4N5H/StarBoy@61e9663
• D3V4N5H/StarBoy
• Fixed Artists for India and Added Top Tracks · D3V4N5H/StarBoy@2bbfafb

To know what the song is all about, I need to summarize it. In graph-based summarization, I will explore a specific approach: Opinion mining.

Here the yellow node (smiley) is the largest and hence most important because of the number of other nodes that point to it in the graph

TextRank

This is like running PageRank on our Text Adjacency Graph. It has Numpy and Scipy dependencies:

https://github.com/summanlp/textrank

Here’s how it went:

This is bad. Good, but bad…

It worked. But the song is ‘Starboy’ and that’s not in the summary. 🤔

Cool. ‘Starboy’ spotted. Better than before. But I will not settle for this.🤨

My Query made it work !

The chorus had the most weight. Look at the first few lines! 🎉 🎊

Links to work:

• D3V4N5H/StarBoy
• Graph based Summarization and Keyword Extraction · D3V4N5H/StarBoy@97fc56b

Knowing preceding and succeeding words is not much helpful (yet), as it only projected an obvious fact, that the ‘articles’ are common (meaning frequent) in the text adjacency graph. I need a way to mathematically find context from the text corpus.

There are two common types of word associations in natural language processing:

paradigmatic and syntagmatic

Paradigmatic: words A and B are paradigmatically related if they can be substituted for each other. This indicates they belong in the same class, such as “Monday” and “Thursday” or “Cat” and “Dog.

Syntagmatic: words that can be combined with each other, such as “cold” and “weather”.

To compute a measure of paradigmatic relation we take an aggregation of the similarity of relative contexts:

Sim(“Cat”, “Dog”) =

Sim(Left1(“Cat”), Left1(“Dog”))

+

Sim(Right1(“Cat”), Right1(“Dog”))

We can use the Jaccard index as a measure of similarity

A revision to the set theory (very silly, I know. Today #AddTheWordSillyToAnything is trending on Twitter 😜)

0: if no words in common
1: if the sets are equal

A visual Example — Object Detection

The dilemma

The whole point of using the Python driver was the ability to leverage the language. But in this case, the Cypher + Python approach won’t allow identifying the most paradigmatically similar words. For example, I can only ask “what is the paradigmatic similarity between words X, Y”. What we really want to ask is “what are the word pairs that are most paradigmatically similar?” For this I will use a pure Cypher approach to compare word pairs and update the graph with a :RELATED_TO relationship. This relationship will store the similarity score. Once this query runs we can query for word relations using this new :RELATED_TO edge.

The query is 21 lines long. It took 6.581 seconds to fire it. 14762 properties were set and same number of relationships were created.

After all this hard work, my query says ‘starboy’ is not related to anything, even though the song is titled “StarBoy”.

Not only that, nothing is related TO starboy either!

However, the song is all about ‘style’, it seems!

This is kinda okay, because I don’t expect the AI to play this song when someone says “starboy”, as that event is highly unlikely.

Perhaps, the words with most ZEROS can tell us what the song is NOT about. All the words in the phrase “cleaner than your church shoes” had many zeros, here’s their score:

cleaner: 121 (all zeros, no other values)

than: 119
“of than” : 0.2
“out than”: 0.33

church: 119
“league church”: 0.5
“whole church”: 0.5

your: 118
“the your”: 0.1
“of your”: 0.125
“these your”: 0.2

shoes: 114
“me shoes”: 0.25
“too shoes”: 0.3333333333333333
“you shoes”: 0.3333333333333333
“mood shoes”: 0.5
“hefty shoes”: 0.5
“SV shoes”: 0.5
“cry shoes”: 0.5

But we all knew that this song was not about “cleaner church shoes”. Now let’s see if 1.0 score can offer any interesting insights about it; something that can help the AI decide when to play the song, or whether or not it is relevant to the environment.

such pairs with r.paradig == 1.0 are:
“end” “test”
“tease” “hurt”
“cup” “style”
“kill” “take”

This kinda makes sense; the lines containing these pairs were repeated throughout the lyrics, and the pairs are in adjacent lines. Hmm. Good.

Fun for Thoughts:

Enough coding for today. Now I will share funny observations with silly captions…

Daft Punk: +1

Daft Punk: +2

Noted.

AI: +1

My AI is in love with Daft Punk. Maybe it realised it lost by 1 point and is ruthless about it 🤦🏻‍♂️

Ask someone today: Which is cleaner, your shoes or my car?

I’m a Classic Man!

This indicates my AI is a female. “Girls get loose when they hear this song”

Exactly 16.66% desperation is a lot for a computer. Believe me.

What?!! It learned and started Poetry?!

AI flattering me. Stop reading this blog, these are our personal moments ✋🏼 (I blushed a little)

English is not easy. ( for a computer 🤪 )

I’ll let you know when it’s done. Reminds me of quotes from Star Wars

and a similar one in Matrix

Link to work:

https://github.com/D3V4N5H/StarBoy/commit/bf55b72c9c85170b2f14980bc574d8b7dcd79929

The error I was facing while trying to keep a list of word pairs as the keys of the dictionary wpDict, was that there’s no built-in hash function for lists (by design).

TypeError: unhashable type: ‘list’

I’m glad about my brainstorming yesterday, as I just found out that dictionaries are implemented as hash tables! So my implementation is as efficient as it ideally should be.

Then I used the inbuilt tuple() instead of explicit assignment of firstWord and secondWord variables. So they are gone for good.

Paradigmatic Similarity

Consider this example from Neo4j’s official video:

I implemented Right1 and Left1 and will eventually implement Right2 and Left2 when it makes sense to do so.

Left1

This means “up my” and “love my” word pairs are there in the lyrics and hence the graph. Then I figured I could count them !

Then I decided to find out which word had the highest number of distinct left words. Hmmm… Let’s see if you can guess it.

Wait for it, this is second highest

Phrases such as “man I” are in lines like “man I love my baby” and “competition I” in anastrophes like “No competition I don’t really listen”.

Also, I should be able to detect a sentence, when it ends, so adjacency doesn’t have unexpected correlations like “cup I”, where the actual sentence ends at “cup” and another begins with “I”.

Yes, this one’s the winner! 🎉

Word pairs “I’m a”, “Like a”, “in a”, “need a” are very common

Right1

This will make even more sense and as it gave me a valuable insight in how we can observer how our language boils down to the basic rules of grammar!

“The” is an article (in grammar terms).

The first line of the song is: “I’m tryna put you in the worst mood ah” !

All this has started to make sense, as “a” and “the” are articles (grammar). They’re always used with a noun. So it makes sense that they would have the highest right-adjacency word pair frequency.

Links to work:

• D3V4N5H/StarBoy
• D3V4N5H/StarBoy

I considered Data Structures like Hash Table (with multiple key values mapped to a value) and Linked List for storing the data for word pairs and their frequencies. At one point, I thought it was a good idea to develop a new data structure, for which I watched a FreeCodeCamp Full Course for Beginners of Data Structures and Algorithms.

I settled for saving the word pairs as a list in a Python dictionary as keys and their frequency count as values.

But in Python it is not possible to use a list as a key for a dictionary, as the key of a dictionary MUST BE immutable

So I used a tuple instead. For this, in each iteration of the loop, I split the word-pair result from the list in the ‘Record object’ into a tuple of two variables, firstWord and secondWord.

Link to work:

https://github.com/D3V4N5H/StarBoy/commit/05c96ab331569f1831943aa3d168921935e88888