Multi-class text classification with LSTM in Keras

Question

I'm quite new to Deep Learning and trying to solve the problem of Multi-Class, multi-label text classification using Deep Learning.

https://github.com/fchollet/keras/blob/master/examples/imdb_cnn_lstm.py . I've another dataset. int form of a csv file ("text","classifier"), on which i want to perform text classification task. I've tried a few ways to pass my training text to keras but couldn't so I'm stuck at this point. Can anyone suggest me how should I pass my "train.csv" and "test.csv" file to the X_train, y_train and X_test, y_test?

Typically stuck in this line.

(X_train, y_train), (X_test, y_test) = imdb.load_data(nb_words=max_features)

'train.csv' has this format:

"Job Description:An ideal fitment would apply his/ her advanced analytics expertise at a cutting edge Industrial Analytics specialized Data Science organization; primarily, in any of the following areas- automotive/ energy/ oil & gas/ aerospace/ marine/ chemical. Experience in statistical modeling, predictive modeling, Random forests, Decision tree, Linear Regression, Correlation, Time- series. BE / MS/ PhD in Mechanical/ OR/ IE/ computer science/ EE/ chemical. Mentor/ Lead a small team of data scientist",Business Analytics

'test.csv' has same format that is "job_description","category"

Answer 1

After I read the source code, I find out that keras.datasets.imdb.load_data doesn't actually load the plain text data and convert them into vector, it just loads the vector which has been converted before.

As for your problem, I assume you want to convert your job_description into vector. Maybe you can try sklearn.feature_extraction.text.CountVectorizer.

I'm no expert on NLP, but I've encountered problems (e.g. Two Sigma Connect: Rental Listing Inquiries) which need such technique.

In the meanwhile, there are other word2vec/embedding techniques you may try. Here is a tutorial from keras which shows a detailed example on word embedding.

References

https://github.com/fchollet/keras/blob/master/keras/datasets/imdb.py

http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.CountVectorizer.html

https://www.kaggle.com/sudalairajkumar/two-sigma-connect-rental-listing-inquiries/xgb-starter-in-python

Answer 2

Something like this:

nb_classes = 3 # the number of categories you have
x_train = []
y_train = []

with open('train.csv', 'r') as train_file:
    reader = csv.reader(train_file)
    for row in reader:
       sentence = row[0]
       category = row[1]

       x_train.append(sentence)
       y_train.append(category)

Y_train = np_utils.to_categorical(y_train, nb_classes)
X_train = ? # your choice of tokanization

You should do the same with your test dataset.

You should also change the loss to categorical_crossentropy.