Structure Analysis
So, we conducted an experiment to investigate our first goal, and generated 20,000 songs by each model. We made a statistical analysis of the results in aspect of structure, and here they are:
As we can see in the Average Words Per Song graph , Trigram-Model creates songs with 15% less words per song than Billboard, whereas Char-Model actually increases the number of words per song by 6%.
No surprise here, since the models stop generating a song only when reaching the upper limit that we set, or when generating a sequence of words\chars which appeared at the end of a song in the learned dataset. The odds to generate a sequence of 11 characters which ended a song are much lower than generating a sequence of 2 words.
Similar conclusions can be drawn regrding to the Average Characters Per Song graph.
We can see that both models create the same number of stanzas per song in average,
one less stanza compare to the original Billboard database.
To increase the number of stanzas in a song, the models need to generate '\n' twice in a row.
We can see from the top right graph that the Chars-Model generates 2 words more per line in avarage, resulting in a smaller number of lines per song as you can see in the top left graph.
Distrubuition of Songs by Words
The next graph represents how many songs there are by number of words
We can see that both models are relativliy stable when generating its songs (compare to Billboard songs database), except of one anomalous pick by each model.
Billboard graph gets its pick in 248 words, with 80 songs composed of this number of words. Chars-Model gets its pick in 405 words, generates 770 songs of this number.
However Ngram-model, gets its pick in 602 words, with 252 songs.
In other words, Char-Model creates its most songs with a number of words closest to the learned Billboard database.
Common Words Analysis
Before we present the best song made by each model in terms of structure,
Let us show the most 200 common words of Billboard songs lyrics.
That is to examine later how many of them appear in each of those best songs.
Interesting fact:
We also found the 200 common words of each model, and compared them to the words compose the above words cloud, here are the results:
-
Trigram-Model - gets a match of 196 words out of 200.
The four missing are : [used, young, door, sing]
-
Chars-Model - gets a match of 195 words out of 200.
The four missing are :[god, came, ha, door, sing]
Sentiment Analysis
We made a quick sentiment analysis to the songs labeling each as positive, negative or neutral
Interesting fact:
Char-Model generated 86% 'Positive' songs, and leads the race to generate a new positive hit! Whereas Trigram-Model generated 77% 'Positive' songs.
Notice that the Billboard songs compose of only 70% 'Positive' songs.
The Greatest Hit By Each Model
Out of 20,000 songs generated by each model, we selected the best one in terms of structure.
In other words, we chose the song which is closest to the average Billboard statistics.
But to support the hypothesis that this song is a great hit, we picked a positive one, with good enough number of common words.
Trigram - Model Best Hit
paint the sky
dream on, dream on
tryin' to say...
life will never die
hollerin' for mercy
say you'll never shine if you want to know this, check it out my cypher keeps moving
just say no
i wonder if i could never lose, what you like
she's got electric boots, a mohair suit
knock three times on the front seat of my car
i can make me feel good
sha, sha-ba-dah, sha-ba-dah, feel good
showin' much flex when it's you run
no regrets, just love the way you do
ooh, ooh, the joy
you got a drinkin' problem
yeah, boo, i dun told you to know the game
imagination
you are
it's just another day, another breath
she paid it no more of that ring, get rid of those two polar opposite spectrums instead of kisses, we get it, skrrt
i'm down to the, move to the front
hair weave off your neck ties to purple labels
upgrade you, i promise i'll be crazy
well, they sweatin' when they move their lips
drench yourself in jesus' hands.
grandma's hands played a lot more love make you put yours, and i am finished with you
but what about us?
we turn this room says not hardly
odd-ly enough, i guess i was rolling one up in my zone
no no oh no, babe
i ain't nobody's fool, bump you like
baby
if or if we fall
i keep in contact with certain exes
do you remember, when i pick you up i got one like me
i got to do this no beer
hes always around like 4l-dub, i'm calling mo' niggas
this mirror's telling no lie
real hasta la nube
pa pa l'americano
i'll be movin'
it's time to say
i can't stop the war, now
i'm living in san antone and santa fe
or are we waiting on?
i asked her why we're who we are, we are diamonds taking shape
shape it up cause girl, i gotta be fresh as hell
chains of love and i ride his dick through his mane
- the words of scarface, again
Songs Statistics:
Stanzas: 5
Lines: 50
Words: 351
Characters: 1787
Lines Per Stanza: 10
Words Per Line: 7
-
The song contains 39 common words out of 200
ain't
another
around
baby
cause
crazy
dream
fall
time
told
turn
two
want
way
will
oh
one
put
real
ride
run
say
stop
lie
life
love
make
move
much
never
now
feel
game
good
got
gotta
keep
know
la
-
By sentiment analysis, song is POSITIVE
Chars - Model Best Hit
let's do it all for my baby
don't you give me your love baby
and step down, they don't see my ones ain't no problems!
no problems
when's daddy ever gonna change
you know i don't
cause you're a little rubber later on this special love last forever
nobody know how to run things
livin' in the streets, so many chances
searching for you
you turn a corner in the night
i know, i know that we'll be singin'
i love rock n' roll
so come and make a night you won't see me blush, but i can make this my last chance
we're gonna die young
we're gonna love her as she looking good to celebrate
holiday ooohooohooohooo
well, it reminded me of the game, they run through the days and nights we don't really care about us
mmm babe
i can't stay
don't have to try
lay down tonight, girl
and i'ma make it right
baby, if you've got to boogie
on the disc
thelonius, my description
and if it feels good
i feel it coming together
we will stand on the wall on the ding-a-ling
won't you call my name
come on y'all come with fries
love it when i see the way i treat her like a man
she waan a man weh mek she fly like kites never get away, get away from me
i'm giving you, giving you ever seen the moon disappear
you're still a part of your crowd?
i never had that feeling
streetlights
no tomorrow
and ooooohhhh, they call their own
they watchin' us while we dancing in the middle of the night and i got it goin' on with the radio
and sing with that hip still real mother for my chance
don't stop now
you can't out run the race is not, for the sunday shoes
please, baby, please don't play yourself
move on back two squares
send an instant
i knew the plan
and it didn't stop
they just sayin' if you got it move it boy
yeah
Songs Statistics:
Stanzas: 5
Lines: 50
Words: 340
Characters: 1699
Lines Per Stanza: 10
Words Per Line: 7
-
The song contains 55 common words out of 200
ain't
away
baby
back
boy
call
care
cause
change
come
run
see
sing
stand
stay
still
stop
together
try
turn
night
nobody
now
play
please
real
really
right
rock
roll
know
last
little
looking
love
make
man
move
name
never
coming
feel
feeling
gorever
girl
give
gonna
good
fot
knew
two
way
will
yeah
young
-
By sentiment analysis, song is POSITIVE