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.

%%time
import malaya

CPU times: user 6 s, sys: 1.23 s, total: 7.23 s
Wall time: 8.33 s

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

malaya.emotion.label

['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

def multinomial(**kwargs):
    """
    Load multinomial emotion model.

    Returns
    -------
    result : malaya.model.ml.Bayes class
    """

model = malaya.emotion.multinomial()

Predict batch of strings

def predict(self, strings: List[str]):
    """
    classify list of strings.

    Parameters
    ----------
    strings: List[str]

    Returns
    -------
    result: List[str]
    """

model.predict([anger_text])

['anger']

model.predict(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)

['anger', 'fear', 'happy', 'love', 'sadness', 'surprise']

Predict batch of strings with probability

def predict_proba(self, strings: List[str]):
    """
    classify list of strings and return probability.

    Parameters
    ----------
    strings: List[str]

    Returns
    -------
    result: List[dict[str, float]]
    """

model.predict_proba(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)

[{'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': 0.14667998117610198,
  'fear': 0.1422732633232615,
  'happy': 0.29984520430807293,
  'love': 0.1409005078277281,
  'sadness': 0.13374705318404811,
  'surprise': 0.13655399018078768},
 {'anger': 0.1590563839629243,
  'fear': 0.14687344690114268,
  'happy': 0.1419948160674701,
  'love': 0.279550441361504,
  'sadness': 0.1285927908584157,
  'surprise': 0.14393212084854254},
 {'anger': 0.13425914937312508,
  'fear': 0.12053328146716755,
  'happy': 0.14923350911233682,
  'love': 0.10289492749919464,
  'sadness': 0.36961334597699913,
  'surprise': 0.12346578657117815},
 {'anger': 0.06724850384395685,
  'fear': 0.1283628050361525,
  'happy': 0.05801958643852813,
  'love': 0.06666524240157067,
  'sadness': 0.06537667186293224,
  'surprise': 0.6143271904168589}]

List available Transformer models

malaya.emotion.available_transformer()

INFO:root:tested on 20% test set.

	Size (MB)	Quantized Size (MB)	macro precision	macro recall	macro f1-score
bert	425.6	111.00	0.99786	0.99773	0.99779
tiny-bert	57.4	15.40	0.99692	0.99696	0.99694
albert	48.6	12.80	0.99740	0.99773	0.99757
tiny-albert	22.4	5.98	0.99325	0.99378	0.99351
xlnet	446.5	118.00	0.99773	0.99775	0.99774
alxlnet	46.8	13.30	0.99663	0.99697	0.99680
fastformer	446.0	113.00	0.99197	0.99194	0.99195
tiny-fastformer	77.2	19.60	0.98926	0.98783	0.98853

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 Transformer model

def transformer(model: str = 'xlnet', quantized: bool = False, **kwargs):
    """
    Load Transformer emotion model.

    Parameters
    ----------
    model : str, optional (default='bert')
        Model architecture supported. Allowed values:

        * ``'bert'`` - Google BERT BASE parameters.
        * ``'tiny-bert'`` - Google BERT TINY parameters.
        * ``'albert'`` - Google ALBERT BASE parameters.
        * ``'tiny-albert'`` - Google ALBERT TINY parameters.
        * ``'xlnet'`` - Google XLNET BASE parameters.
        * ``'alxlnet'`` - Malaya ALXLNET BASE parameters.
        * ``'fastformer'`` - FastFormer BASE parameters.
        * ``'tiny-fastformer'`` - FastFormer TINY parameters.

    quantized : bool, optional (default=False)
        if True, will load 8-bit quantized model.
        Quantized model not necessary faster, totally depends on the machine.

    Returns
    -------
    result: model
        List of model classes:

        * if `bert` in model, will return `malaya.model.bert.MulticlassBERT`.
        * if `xlnet` in model, will return `malaya.model.xlnet.MulticlassXLNET`.
        * if `fastformer` in model, will return `malaya.model.fastformer.MulticlassFastFormer`.
    """

model = malaya.emotion.transformer(model = 'albert')

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.

quantized_model = malaya.emotion.transformer(model = 'albert', quantized = True)

WARNING:root:Load quantized model will cause accuracy drop.

INFO:tensorflow:loading sentence piece model

INFO:tensorflow:loading sentence piece model

Predict batch of strings

def predict(self, strings: List[str]):
    """
    classify list of strings.

    Parameters
    ----------
    strings: List[str]

    Returns
    -------
    result: List[str]
    """

model.predict(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)

['anger', 'fear', 'anger', 'love', 'sadness', 'surprise']

quantized_model.predict(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)

['anger', 'fear', 'happy', 'love', 'sadness', 'surprise']

Predict batch of strings with probability

def predict_proba(self, strings: List[str]):
    """
    classify list of strings and return probability.

    Parameters
    ----------
    strings : List[str]

    Returns
    -------
    result: List[dict[str, float]]
    """

model.predict_proba(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)

[{'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}]

quantized_model.predict_proba(
    [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
)

[{'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

def predict_words(
    self, string: str, method: str = 'last', visualization: bool = True
):
    """
    classify words.

    Parameters
    ----------
    string : str
    method : str, optional (default='last')
        Attention layer supported. Allowed values:

        * ``'last'`` - attention from last layer.
        * ``'first'`` - attention from first layer.
        * ``'mean'`` - average attentions from all layers.
    visualization: bool, optional (default=True)
        If True, it will open the visualization dashboard.

    Returns
    -------
    result: dict
    """

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)

from IPython.core.display import Image, display

display(Image('emotion-dashboard.png', width=800))

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

texts = [anger_text, fear_text, happy_text, love_text, sadness_text, surprise_text]
r = quantized_model.vectorize(texts, method = 'first')

from sklearn.manifold import TSNE
import matplotlib.pyplot as plt

tsne = TSNE().fit_transform(r)
tsne.shape

(6, 2)

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',
    )

Word level

r = quantized_model.vectorize(texts, method = 'word')

x, y = [], []
for row in r:
    x.extend([i[0] for i in row])
    y.extend([i[1] for i in row])

tsne = TSNE().fit_transform(y)
tsne.shape

(49, 2)

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',
    )

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

multinomial = malaya.emotion.multinomial()

malaya.stack.predict_stack([multinomial, model], [anger_text])

[{'anger': 0.5739743139312979,
  'fear': 0.002130791264743306,
  'happy': 0.0019609404077070573,
  'love': 0.0016901068202818533,
  'sadness': 0.001121633002361737,
  'surprise': 0.0015551851123993595}]

malaya.stack.predict_stack([multinomial, model], [anger_text, sadness_text])

[{'anger': 0.5739743139312979,
  'fear': 0.002130791264743306,
  'happy': 0.0019609404077070573,
  'love': 0.0016901068202818533,
  'sadness': 0.001121633002361737,
  'surprise': 0.0015551858768478731},
 {'anger': 0.001604580129233267,
  'fear': 0.0011600003908574707,
  'happy': 0.0009737663531537643,
  'love': 0.0008489265368074127,
  'sadness': 0.6079418541812244,
  'surprise': 0.001139192858067602}]