AFRY NLP Competence Group Meetings

. In this repository I'll try to summarize & keep the data from the different meetings. Some meetings are found in the "jvm"-folder but most are found in "python" as Python is the "de facto" langauge for Machine Learning in the industry.

The different meetings:

1. Text Classification IMDB - 20190204
2. (WIP) Text Classification Quora Insincere Questions - 20190223
3. Quora pt 2 - 20190402
4. Neural Network 101 - 20190521
5. Text Generation pt 1 - 20190626
6. Text Generation pt 2 - 20191017
7. Data Mining & Social Media Perception - 20191127
8. SparkNLP (& how it compares to other approaches) - 20200212
9. [NER]

Text Classification IMDB [20190204, #1]

Goal: To classify IMDB review as positive or negative
How: We started out with explaining basic concepts, then we implemented classic Machine Learning & ended up looking into Word Embeddings
Why: Fun task where we learned a lot of basics in Machine Learning and Natural Language Processing
Keywords: TFIDF, Bag of Words, Preprocessing, Machine Learning, Word2Vec, Word Embeddings, scikit-learn, pandas,
Where: Local: python/20190204 - #1 [TextClassification]. Google Colab (browser)

(WIP) Text Classification Quora Insincere Questions [20190223, #2]

Currently a mess really. We did this locally & had a lot of dependencies... I'm gonna find time sometime to convert this into .ipynb-files to make it easier to work with.
Goal: To classify Quora posts as insincere or sincere (Kaggle competition). A very hard task as not even humans agree.
How: Implemented a baseline using spaCy which is a widely used tool in the industry. Reached a decent scoring.
Why: This meeting we decided what approaches to use in the future (more practical than theoretical)

Text Classification Quora Insincere Questions pt 2 [20190402, #3]

Goal: To classify Quora posts as insincere or sincere (Kaggle competition). A very hard task as not even humans agree.
How: Two implementations that we can tweak directly on Kaggle provided by me.
Why: To see how Kaggle works, test more complex solutions and have some fun by joining a competition

Neural Networks Basics [20190521, #4]

Goal: To learn the basics of Neural Networks such as the building stones, the flow & some feel of the math.
How: A great Python Notebook that shows theory and then lets you implement a Recurrent Neural Network that performs Machine Translation (Eng -> Swe)
Why: Learn basics in Neural Network so that new layers won't they're recognizable, such as Dropout layer etc.

Text Generation [20190626, #5]

Goal: To get a greater understanding of Language Models & see how the "old-school" approaches might even beat Neural Networks with much simpler & easier to grasp concepts + some clever tricks.
How: A Python Notebook as usual which was mainly code to code yourself (!). We generated words on a word-to-word and char-to-char basis.
Why: Understand the core to later expand with more advanced techniques such as Neural Networks

Text Generation pt 2 [20191017, #6]

Goal: To finally try out Neural Networks & generate some Shakespeare.
How: A Python Notebook as usual which was mainly code to code yourself (!). We generated words on a char-to-char basis.
Why: Work with Neural Networks & learn tensorflow

Social Media Perception / Data Mining [20191127, #7]

Goal: Learn some techniques for Data Understanding and ways to visualize data to learn more about it
How: A Python Notebook as usual, during this workshop you'll first learn how to get data from Twitter, parse it and then show it in a Word Cloud. In the end a classifier for which one who's tweeting a certain tweet will be done, comparing J Sjöstedt to J Åkesson.
Why: Learn how to fetch Twitter data in bulk, how to make word-cloud and have some fun.

SparkNLP [20200212, #8]

Goal: Learn what Spark, SparkML and SparkNLP is. What we can do with it and so on. Also learn when not to use it (as it brings overhead).
How: Using a JVM project (hidden in jvm-folder) we had tooling to try out a Scala Spark approach and a Kotlin "Local" approach. Why: For fun, and to understand that all tools have their place.

NER (Named Entity Recognition) [20200212, #9]

Goal: understand what NER is, what applications exists & how to create a basic NER classifier.
How: Jupyter Notebook.
Why: Very interesting challenge that could create the base of...