Text Mining with Machine Learning and Python

This video highlights the types of data the text mining data scientists might come into contact with, and where he can get his hands on some text data to get working on.

•         Explain how data sources differ
•         Explain the uses for each type
•         Point to some sources for each type

Under the motto garbage in = garbage out, we look at some often-used pre-processing steps.

•         Explain why we need to clean text
•         Explain what aspect can be cleaned
•         Explain the importance of order to follow

A key aspect of transforming text to numeric is correctly building your corpus of words and word features. The appropriate techniques are explained in this video.

•         Find out what tokenization is
•         Learn what POS tagging is
•         Find out what lemmatization is

To further correctly refine our corpus, n-grams often play an important part. This video explains how to approach this.

•         Tackle the problem in a non-intelligent way
•         Use chunking to be more precise
•         Use statistical approaches to handle the problem

This video gives a quick intro to the information extraction problem. First up is a very important difference between an ML approach and a text search approach.

•         Find out what information extraction is
•         Find out what information extraction is
•         Differentiate between search-based and machine learning based

A deeper dive into the world of named entity recognition, the machine learning approach to information extraction. It is important to know how this approach works.

•         Explain what Named Entity Recognition is
•         Explain the types of approaches and models
•         Explain how to choose the correct approach

For a lot of so-called entities, pre-trained models exist which one can use of the shelf. But when, which entities and which models often remains unclear. This is explained in this video.

•         Lay out the various popular entities
•         Lay out the popular models to do use
•         Demonstrate how to use them and when they fall short

Often, pre-trained entity types are just not enough. This video will demonstrate what to do in such a scenario.

•         Discuss why you sometimes need to train your own model
•         Train using spaCy
•         Train using PyCRF

State of the art results can be obtained using more advanced deep learning models. This video gently explores that option, along with the opportunities and pitfalls.

•         Explore the next step in NER – deep learning
•         Find out when does it (not) make sense
•         Demonstrate and urging the viewer to further dive into this

This video gives an introduction to the problem of text classification, along with the first step in the process – representing your text in a mathematical vector for use in a learning algorithm.

•         Find out what text classification is
•         Explain why mathematical feature representation is needed
•         Differentiate between various representation techniques

There are many algorithms and techniques out there to tackle the problem. This video aims to guide the student to selecting the right one for his or her problem.

•         Explain how to divide your problem
•         Lay out which algorithms work well and which don’t
•         Explain model parameters you need to take into account

A dive into a coded example, starting from a popular dataset, to show what we have learned so far in each step, and have a full working example.

•         Demonstrate feature representation and data preparation
•         Demonstrate model setup
•         Demonstrate evaluation

In our previous example, there are a lot of choices to be made and hyperparameters to be tuned, this video gives an overview of this.

•         Explain why and how to do hyperparameter tuning
•         Explain other choices
•         Provide neat tips and tricks

This step is often omitted in other courses but since this course aims to provide a real world view on text machine learning, some attention is spent on putting classification into production.

•         Explain why it matters to think about this
•         Explain what you need to think about
•         Provide other tips and tricks

State of the art results can be obtained using more advanced deep learning models. This video gently explores that option, along with the opportunities and pitfalls.

•         Lay out the general thinking approach
•         Explore CNNs
•         Explore other options

This video gives an introduction to the concept of word embeddings – the history and the use cases.

•         Introduce word embeddings
•         Explain how they became important
•         Explain the use cases

Moving deeper into Word2Vec (with skip-grams and CBOW) and Glove to better clarify and structure the main techniques used.

•         Explain the difference between glove and Word2Vec and FastText
•         Explain the difference between skip-grams and CBOW
•         Point to pre-trained word embeddings

We now know what it is, but not yet how to use it. The first step is training a word embedding model.

•         Introduce the gensim package to help us
•         Demonstrate how to train a Word2Vec model with it
•         Demonstrate how to train a FastText model with it

To demonstrate some of the powerful aspects of word embeddings, we will try to visualize one.

•         Explain the problem, dimensionality reduction
•         Introduce and use T-sne
•         Demonstrate some key word embedding strengths

The power of word embeddings is being applied more and more often in other domains as well, as well on other things than just words. This video tries to tease some of these possibilities.

•         Explain the transferability of word vectors
•         Explain expansion one – from word to other aspects
•         Explain expansion two – embeddings in other domains

We saw a lot of concepts introduced in the previous sections. To consolidate, we will try to stitch everything together in one overview.

•         Introduce an example to map everything to
•         Map it in terms of the various sections
•         Place introduced tools on overview

A method we haven’t seen yet is topic modelling. Though it can be a useful one for some topics. Based on what we already know (tools and methods), the introduction is easily made.

•         Illustrate what topic modelling is
•         Illustrate the use cases for topic modelling
•         Illustrate the tools you can use

From here on, the more exotic topics will be addressed, starting with text generation. Using neural network architectures, models can be trained to predict sequences in the style of the training corpora.

•         Introduce the use case of text generation
•         Quickly browse the neural network types used for this
•         Illustrate with a fun example

One of the large areas where neural networks have revolutionized NLP is machine translation. This video quickly touches upon this topic.

•         Introduce SMT and NMT
•         Touch upon encoder-decoder networks
•         Point to further reading and examples

From here on forth, there are a number of areas one can use to get further acquainted. It can be difficult to find these right paths on your own, so this video will give some nice pointers to get started.

•         Point to recommended books
•         Point to recommended blog repos
•         Point to Kaggle

Again, we saw a large number of different topics, both in this section and the previous ones. In this final video, we try to structure everything in one large overview.

•         Place the basic NLP topics on the overview
•         Place the advanced NLP/ML topics on the overview
•         Place the exotic NLP/ML topics on the overview