NLP and Deep Learning for non_experts
1 like330 views
The document presents an introduction to NLP and deep learning for non-experts, outlining key concepts and methodologies for starting projects in machine learning. It covers aspects such as the bag of words model, deep learning architecture, text classification, and convolutional neural networks, emphasizing the importance of high-quality training data and various optimization techniques. Applications discussed include sentiment analysis, recommendations, and language processing tasks.
NLP and Deep Learning for non_experts
- 1. NLP & Deep Learning for non-experts Sanghamitra Deb Staff Data Scientist Chegg Inc
- 2. How to start projects in machine learning? • Kaggle competitions --- • Make sure to solve the ML problems for concept development before competing
- 3. How to start projects in machine learning? • Kaggle competitions --- • Make sure to solve the ML problems for concept development before competing
- 4. How to start projects in machine learning? • Self guided workshops/projects --- lets say you have data from Zomato • Restaurant recommendation -- user based, content similarity based. • Restaurant tags from reviews. • Sentiment analysis from reviews.
- 5. Outline • What is NLP • Bag of Words model for sentiment analysis using scikit learn • DeepDive into deep learning • Solve the sentiment analysis problem using keras • A short into Convolution Neural Networks (CNN)
- 6. What is Natural Language Processing? • Giving structure to unstructured data • Learn properties of the data that makes decision making simple • Provide concise information to drive intelligence of different systems.
- 7. Why? • Unstructured data cannot be consumed directly • Automate simple and complex functionalities • Inferences from text data becomes queriable. This could help with regular BU reports • Understand customers better and take necessary actions for better experience.
- 8. Applications • Categorization of text • Building domain specific Knowledge Graph • Recommendations • Web --- Search • HR --- people analytics • Medical --- drug discovery, automated diagnosis • ………..
- 9. What are the underlying tasks? • Syntactic Parsing of sentences --- parsing based on structure • Part of Speech Tagging • Semantic Parsing -- mapping text directly into formal query language, e.g. SQL queries for a pre-determined database schema. • Dialogue state tracking --- chatbots • Machine Translation • Language modeling • Text extraction • Classification
- 10. Text Classification Text Pre - processing Collecting Training Data Model Building Offline SME • Reduces noise • Ensures quality • Improves overall performance • Training Data Collection / Examples of classes that we are trying to model • Model performance is directly correlated with quality of training data • Model selection • Architecture • Parameter Tuning User Online Model Evaluation
- 11. Text Data Data Source -- https://archive.ics.uci.edu/ml/datasets/Sentiment+Labelled+Sentences
- 12. Model Building: a simple Bag of words (BOW) model https://realpython.com/python-keras-text-classification/
- 13. Model Building: a simple BOW model https://realpython.com/python-keras-text-classification/
- 14. Deep Learning Deep learning algorithms seek to exploit the unknown structure in the input distribution in order to discover good representations, often at multiple levels, with higher-level learned features defined in terms of lower-level features. --- Yoshua Bengio a kind of learning where the representation you form have several levels of abstraction, rather than a direct input to output --- Peter Norvig When you hear the term deep learning, just think of a large deep neural net. Deep refers to the number of layers typically and so this kind of the popular term that’s been adopted in the press. I think of them as deep neural networks generally. --- Andrew Ng
- 15. Why now? • Explosion in labelled data. • Exponential growth in computation power with cloud computing and availability of GPUs • Improvements in setting initial conditions and activation functions
- 16. Neural Network Simulate the brain and get neurons densely interconnected in a computer such that it can learn things, recognize patterns and take decisions?
- 17. Neural Network Simulate the brain and get neurons densely interconnected in a computer such that it can learn things, recognize patterns and take decisions? What is a neuron?
- 18. Neural Network Simulate the brain and get neurons densely interconnected in a computer such that it can learn things, recognize patterns and take decisions? What is a neuron?
- 19. What is neuron? https://www.slideshare.net/tw_dsconf/ss-62245351 a1 a2 a3
- 20. What is neuron? https://www.slideshare.net/tw_dsconf/ss-62245351 a1 a2 a3
- 22. Neural Network • Each node is a function with input and output vectors • Every network structure is defined by a set of functions
- 23. Output Layer
- 24. • Loss is minimized using Gradient Descent • Find network parameters such that the loss is minimized • This is done by taking derivatives of the loss wrt parameters. • Next the parameters are updated by subtracting learning rate times the derivative
- 25. Commonly used loss functions • Mean Squared Error Loss • Mean Squared Logarithmic Error Loss • Mean Absolute Error Loss Regression Loss Functions • Binary Cross-Entropy • Hinge Loss • Squared Hinge Loss Binary Classification Loss Functions • Multi-Class Cross-Entropy Loss • Sparse Multiclass Cross-Entropy Loss • Kullback Leibler Divergence Loss Multi-Class Classification Loss Functions
- 27. Dropout -- avoid overfitting • Large weights in a neural network are a sign of a more complex network that has overfit the training data. • Probabilistically dropping out nodes in the network is a simple and effective regularization method. • A large network with more training and the use of a weight constraint are suggested when using dropout.
- 29. Adam Optimization • adaptive moment estimation • The method computes individual adaptive learning rates for different parameters from estimates of first and second moments of the gradients. • Calculates an exponential moving average of the gradient and the squared gradient, parameters control the decay rates of these moving averages. https://machinelearningmastery.com/adam-optimization-algorithm-for-deep-learning/
- 30. Activation Functions • Sigmoid/ Softmax • Tanh • Relu • Leaky Relu • Swish
- 31. Activation Functions • Sigmoid/ Softmax • Tanh • Relu • Leaky Relu • Swish
- 32. Activation Functions • Sigmoid/ Softmax • Tanh • Relu • Leaky Relu • Swish Derivative
- 33. Activation Functions • Sigmoid/ Softmax • Tanh • Relu • Leaky Relu • Swish Derivative a = max(0,z)
- 34. Activation Functions • Sigmoid/ Softmax • Tanh • Relu • Leaky Relu • Swish
- 35. Activation Functions • Sigmoid/ Softmax • Tanh • Relu • Leaky Relu • Swish https://arxiv.org/abs/1710.05941v1
- 37. Text Classification using feed forward NN https://realpython.com/python-keras-text-classification/
- 38. Text Classification using feed forward NN
- 39. Fit & measure accuracy! plot_history(history) Clearly overfits the data!
- 40. Can we do better? Word Embeddings • Words are represented as dense vectors • These vectors are • Learned during the training task by the neural network • Pre-trained, learned from Language Models • Encode the semantic meaning of the word.
- 41. Text Pre-processing with Keras PaddingTokenizing
- 42. Start with an Embedding Layer • Embedding Layer of Keras which takes the previously calculated integers and maps them to a dense vector of the embedding. o Parameters Ø input_dim: the size of the vocabulary Ø output_dim: the size of the dense vector Ø input_length: the length of the sequence Hope to see you soon Nice to see you again After training https://stats.stackexchange.com/questions/270546/how-does-keras-embedding-layer-work
- 43. Add a pooling layer • MaxPooling1D/AveragePooling1D or a GlobalMaxPooling1D/GlobalAveragePooling1D layer • way to downsample (a way to reduce the size of) the incoming feature vectors. • Global max/average pooling takes the maximum/average of all features whereas in the other case you have to define the pool size.
- 45. Training Using pre-trained word embeddings will lead to an accuracy of 0.82. This is a case of transfer learning. https://realpython.com/python-keras-text-classification
- 46. Embeddings + Maxpooling -- Benifits • Power of generalization --- embeddings are able to share information across similar features. • Fewer nodes with zero values.
- 47. Convolution Neural Network Detect features ! Downsample.
- 48. What is a CNN? In a traditional feedforward neural network we connect each input neuron to each output neuron in the next layer. That’s also called a fully connected layer, or affine layer. • We use convolutions over the input layer to compute the output. This results in local connections, where each region of the input is connected to a neuron in the output. Each layer applies different filters and combines the result • During the training phase, a CNN automatically learns the values of its filters based on the task you want to perform. Tricky --- dimensions keep changing as we go from one layer to another
- 49. Model definition Embedding_dim = 50 maxlen=10
- 50. Advantages of CNN • Character Based CNN • Has the ability to deal with out of vocabulary words. This makes it particularly suitable for user generated raw text. • Works for multiple languages. • Model size is small since the tokens are limited to the number of characters ~ 70. This makes real life deployments easier and faster. • Networks with convolutional and pooling layers are useful for classification tasks in which we expect to find strong local clues regarding class membership.
- 51. Takeaways! • If you have text data you need to use NLP • Try a simple bag of words model for your data • Having a high level understanding of deep learning will help with better judgement in architecture design and choice of parameters. • Deep Learning has the potential to give high performance, you do need large amount of training data for the benefits.
- 53. Visualization of the architecture 50 10 GlobalMaxPool1D DenseLayer Sigmoid Cov1D
- 55. Appendix
- 57. Character Based CNNs. https://papers.nips.cc/paper/5782-character-level-convolutional-networks-for-text-classification.pdf • Embedding Layer • Six convolutional layers, and 3 convolutional layers followed by a max pooling layer • Two fully connected layer(dense layer in keras), neuron units are 1024. • Output layer(dense layer), neuron units depends on classes. In this task, we set it 4.
- 58. Pre-processing