Emotion Analysis

This tutorial is available as an IPython notebook at Malaya/example/emotion.

This module trained on both standard and local (included social media) language structures, so it is save to use for both.

[1]:
%%time
import malaya
CPU times: user 4.85 s, sys: 677 ms, total: 5.53 s
Wall time: 4.67 s
[2]:
anger_text = 'babi la company ni, aku dah la penat datang dari jauh'
fear_text = 'takut doh tengok cerita hantu tadi'
happy_text = 'bestnya dapat tidur harini, tak payah pergi kerja'
love_text = 'aku sayang sgt dia dah doh'
sadness_text = 'kecewa tengok kerajaan baru ni, janji ape pun tak dapat'
surprise_text = 'sakit jantung aku, terkejut dengan cerita hantu tadi'

Get label

[3]:
malaya.emotion.label
[3]:
['anger', 'fear', 'happy', 'love', 'sadness', 'surprise']

All models follow same method as sklearn interface, predict to get batch of labels, predict_proba to get batch of probabilities.

Load multinomial model

All model interface will follow sklearn interface started v3.4,

model.predict(List[str])

model.predict_proba(List[str])
[5]:
model = malaya.emotion.multinomial()
[6]:
model.predict([anger_text])
[6]:
['anger']
[8]:
model.predict(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)
[8]:
['anger', 'fear', 'happy', 'love', 'sadness', 'surprise']
[9]:
model.predict_proba(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)
[9]:
[{'anger': 0.32948272681734814,
  'fear': 0.13959708810717708,
  'happy': 0.14671455153216045,
  'love': 0.12489192355631354,
  'sadness': 0.1285972541671178,
  'surprise': 0.13071645581988448},
 {'anger': 0.11379406005377896,
  'fear': 0.4006934391283133,
  'happy': 0.11389665647702245,
  'love': 0.12481915233837086,
  'sadness': 0.0991261507380643,
  'surprise': 0.14767054126445014},
 {'anger': 0.15051890586527464,
  'fear': 0.13931406415515296,
  'happy': 0.32037710031973415,
  'love': 0.13747954667255546,
  'sadness': 0.11565866743099411,
  'surprise': 0.13665171555628927},
 {'anger': 0.1590563839629243,
  'fear': 0.14687344690114268,
  'happy': 0.1419948160674701,
  'love': 0.279550441361504,
  'sadness': 0.1285927908584157,
  'surprise': 0.14393212084854254},
 {'anger': 0.14268176425895224,
  'fear': 0.12178299725318226,
  'happy': 0.16187751258299898,
  'love': 0.1030494733572262,
  'sadness': 0.34277869755707796,
  'surprise': 0.1278295549905621},
 {'anger': 0.06724850384395685,
  'fear': 0.1283628050361525,
  'happy': 0.05801958643852813,
  'love': 0.06666524240157067,
  'sadness': 0.06537667186293224,
  'surprise': 0.6143271904168589}]

List available Transformer models

[3]:
malaya.emotion.available_transformer()
INFO:root:tested on 20% test set.
[3]:
Size (MB) Quantized Size (MB) Accuracy
bert 425.6 111.00 0.992
tiny-bert 57.4 15.40 0.988
albert 48.6 12.80 0.997
tiny-albert 22.4 5.98 0.981
xlnet 446.5 118.00 0.990
alxlnet 46.8 13.30 0.989

Make sure you can check accuracy chart from here first before select a model, https://malaya.readthedocs.io/en/latest/Accuracy.html#emotion-analysis

You might want to use Tiny-Albert, a very small size, 22.4MB, but the accuracy is still on the top notch.

Load Albert model

All model interface will follow sklearn interface started v3.4,

model.predict(List[str])

model.predict_proba(List[str])
[4]:
model = malaya.emotion.transformer(model = 'albert')
INFO:tensorflow:loading sentence piece model

Load Quantized model

To load 8-bit quantized model, simply pass quantized = True, default is False.

We can expect slightly accuracy drop from quantized model, and not necessary faster than normal 32-bit float model, totally depends on machine.

[3]:
quantized_model = malaya.emotion.transformer(model = 'albert', quantized = True)
WARNING:root:Load quantized model will cause accuracy drop.
WARNING:tensorflow:From /Users/huseinzolkepli/Documents/Malaya/malaya/function/__init__.py:74: The name tf.gfile.GFile is deprecated. Please use tf.io.gfile.GFile instead.

WARNING:tensorflow:From /Users/huseinzolkepli/Documents/Malaya/malaya/function/__init__.py:74: The name tf.gfile.GFile is deprecated. Please use tf.io.gfile.GFile instead.

WARNING:tensorflow:From /Users/huseinzolkepli/Documents/Malaya/malaya/function/__init__.py:76: The name tf.GraphDef is deprecated. Please use tf.compat.v1.GraphDef instead.

WARNING:tensorflow:From /Users/huseinzolkepli/Documents/Malaya/malaya/function/__init__.py:76: The name tf.GraphDef is deprecated. Please use tf.compat.v1.GraphDef instead.

WARNING:tensorflow:From /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/albert/tokenization.py:240: The name tf.logging.info is deprecated. Please use tf.compat.v1.logging.info instead.

WARNING:tensorflow:From /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/albert/tokenization.py:240: The name tf.logging.info is deprecated. Please use tf.compat.v1.logging.info instead.

INFO:tensorflow:loading sentence piece model
INFO:tensorflow:loading sentence piece model
WARNING:tensorflow:From /Users/huseinzolkepli/Documents/Malaya/malaya/function/__init__.py:69: The name tf.InteractiveSession is deprecated. Please use tf.compat.v1.InteractiveSession instead.

WARNING:tensorflow:From /Users/huseinzolkepli/Documents/Malaya/malaya/function/__init__.py:69: The name tf.InteractiveSession is deprecated. Please use tf.compat.v1.InteractiveSession instead.

Predict batch of strings

[16]:
model.predict_proba(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)
[16]:
[{'anger': 0.9998901,
  'fear': 3.2524113e-05,
  'happy': 2.620931e-05,
  'love': 2.2871463e-05,
  'sadness': 9.782951e-06,
  'surprise': 1.8502667e-05},
 {'anger': 1.6941378e-05,
  'fear': 0.9999205,
  'happy': 9.070281e-06,
  'love': 2.044179e-05,
  'sadness': 6.7731107e-06,
  'surprise': 2.6314676e-05},
 {'anger': 0.15370166,
  'fear': 0.0013852724,
  'happy': 0.8268689,
  'love': 0.011433229,
  'sadness': 0.0011807577,
  'surprise': 0.005430276},
 {'anger': 1.2597201e-05,
  'fear': 1.7600481e-05,
  'happy': 9.667115e-06,
  'love': 0.9999331,
  'sadness': 1.3735416e-05,
  'surprise': 1.3399296e-05},
 {'anger': 1.9176923e-05,
  'fear': 1.1163729e-05,
  'happy': 6.353941e-06,
  'love': 7.004002e-06,
  'sadness': 0.99994576,
  'surprise': 1.0511084e-05},
 {'anger': 5.8739704e-05,
  'fear': 1.9771342e-05,
  'happy': 1.8316741e-05,
  'love': 2.2319455e-05,
  'sadness': 3.646786e-05,
  'surprise': 0.9998443}]
[4]:
quantized_model.predict_proba(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)
[4]:
[{'anger': 0.99988353,
  'fear': 3.5938003e-05,
  'happy': 2.7778764e-05,
  'love': 2.3541537e-05,
  'sadness': 9.574292e-06,
  'surprise': 1.9607493e-05},
 {'anger': 1.6855265e-05,
  'fear': 0.9999219,
  'happy': 9.185196e-06,
  'love': 2.0216348e-05,
  'sadness': 6.6679663e-06,
  'surprise': 2.5186611e-05},
 {'anger': 0.22842072,
  'fear': 0.001628682,
  'happy': 0.7477462,
  'love': 0.014303649,
  'sadness': 0.0013838055,
  'surprise': 0.00651699},
 {'anger': 1.28296715e-05,
  'fear': 1.7833345e-05,
  'happy': 9.577061e-06,
  'love': 0.9999324,
  'sadness': 1.3832815e-05,
  'surprise': 1.34745715e-05},
 {'anger': 1.9776813e-05,
  'fear': 1.1116885e-05,
  'happy': 6.3422367e-06,
  'love': 6.905633e-06,
  'sadness': 0.9999455,
  'surprise': 1.0316757e-05},
 {'anger': 5.8218586e-05,
  'fear': 2.07504e-05,
  'happy': 1.8061248e-05,
  'love': 2.1852256e-05,
  'sadness': 3.5944133e-05,
  'surprise': 0.99984515}]

Open emotion visualization dashboard

Default when you call predict_words it will open a browser with visualization dashboard, you can disable by visualization=False.

[ ]:
model.predict_words(sadness_text)
[18]:
from IPython.core.display import Image, display

display(Image('emotion-dashboard.png', width=800))
_images/load-emotion_25_0.png

Vectorize

Let say you want to visualize sentence / word level in lower dimension, you can use model.vectorize,

def vectorize(self, strings: List[str], method: str = 'first'):
    """
    vectorize list of strings.

    Parameters
    ----------
    strings: List[str]
    method : str, optional (default='first')
        Vectorization layer supported. Allowed values:

        * ``'last'`` - vector from last sequence.
        * ``'first'`` - vector from first sequence.
        * ``'mean'`` - average vectors from all sequences.
        * ``'word'`` - average vectors based on tokens.

    Returns
    -------
    result: np.array
    """

Sentence level

[5]:
texts = [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
r = quantized_model.vectorize(texts, method = 'first')
[6]:
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt

tsne = TSNE().fit_transform(r)
tsne.shape
[6]:
(6, 2)
[8]:
plt.figure(figsize = (7, 7))
plt.scatter(tsne[:, 0], tsne[:, 1])
labels = texts
for label, x, y in zip(
    labels, tsne[:, 0], tsne[:, 1]
):
    label = (
        '%s, %.3f' % (label[0], label[1])
        if isinstance(label, list)
        else label
    )
    plt.annotate(
        label,
        xy = (x, y),
        xytext = (0, 0),
        textcoords = 'offset points',
    )
_images/load-emotion_30_0.png

Word level

[9]:
r = quantized_model.vectorize(texts, method = 'word')
[10]:
x, y = [], []
for row in r:
    x.extend([i[0] for i in row])
    y.extend([i[1] for i in row])
[11]:
tsne = TSNE().fit_transform(y)
tsne.shape
[11]:
(49, 2)
[12]:
plt.figure(figsize = (7, 7))
plt.scatter(tsne[:, 0], tsne[:, 1])
labels = x
for label, x, y in zip(
    labels, tsne[:, 0], tsne[:, 1]
):
    label = (
        '%s, %.3f' % (label[0], label[1])
        if isinstance(label, list)
        else label
    )
    plt.annotate(
        label,
        xy = (x, y),
        xytext = (0, 0),
        textcoords = 'offset points',
    )
_images/load-emotion_35_0.png

Pretty good, the model able to know cluster top right as surprise emotion.

Stacking models

More information, you can read at https://malaya.readthedocs.io/en/latest/Stack.html

[4]:
multinomial = malaya.emotion.multinomial()
[6]:
malaya.stack.predict_stack([multinomial, model], [anger_text])
[6]:
[{'anger': 0.5739743139312979,
  'fear': 0.002130791264743306,
  'happy': 0.0019609404077070573,
  'love': 0.0016901068202818533,
  'sadness': 0.001121633002361737,
  'surprise': 0.0015551851123993595}]
[7]:
malaya.stack.predict_stack([multinomial, model], [anger_text, sadness_text])
[7]:
[{'anger': 0.5739743139312979,
  'fear': 0.002130791264743306,
  'happy': 0.0019609404077070573,
  'love': 0.0016901068202818533,
  'sadness': 0.001121633002361737,
  'surprise': 0.0015551858768478731},
 {'anger': 0.0016541454680912208,
  'fear': 0.0011659984542562358,
  'happy': 0.001014179551389293,
  'love': 0.0008495638318424924,
  'sadness': 0.5854571761989077,
  'surprise': 0.001159149836587787}]
[ ]: