In the individual assignment we will use the same design brief we used in the tutorial. Your goal is to gain insights from analyzing the student’s answers regarding their feeling during the COVID-era. Following the same steps, we saw in the tutorial try to answer the following questions. Hint: To answer the questions you will need to change some of the parameters we used, understand their effect and importance, and reflect on the results.
Experiment with changing some of the variables from “True” to “False”. Try to answer the following questions:
- Why do we remove capitalization? When it can be important to leave capitalization be?
- What is tokenization and why do we use it?
- Why do we remove punctuation and numbers? What will happen if we don’t remove it? Are there cases in which it is better not to remove them?
- What are stop words? What happens if we do not remove them?
- What is stemming and why do we use it?
- Set all the arguments to true, add extra_stopword “stress”, and generate a wordcloud. The word “stress“ is still there. Why do you think this is the case? Try to fix the situation.
The main goal of this task is to learn how to work with Wordcloud visualization. For that purpose try to follow the following steps:
- Draw a wordcloud on preprocessed data
- Draw a wordcloud on data with different preprocessing protocol
- (a) Get back to the preprocessing
- (b) Remove stemming (argument stem=False) from preprocessing
- (c) Draw a new wordcloud and save it under a different name
- What’s different?
- Try increasing the number of shown words. Are additional words give you meaningful information about text collection?
The tricky part about sentiment analysis is to understand its limitations and the origin of sentiment score (i.e. which words affect the score the most). To practice working with sentiment analysis, try to do the following steps:
- Calculate the sentiment scores for your text collection
- Find top-3 most positive documents
- Find top-3 most negative documents
- What’s different about these documents? Why do you think the algorithms finds some sentences negative? What words give away the sentiment? What about the positive sentences?
- How sentiment analysis of this collection can help decision-makers improve the life of the students?
- Set a min_len to different thresholds: 3, 10, 15. How does it affect the output of your model? Repeat steps 1-3 to see how text length affects your results.
The main task here is to experiment with different parameters and explore how they affect the outcome of the LDA model. Ideally, you will generate several models during this step.
- Run the default model and generate csv “examples of topics”.
- Try to interpret at least 5 topics (you can split different topics among teammates).
Now, repeat steps 1-2 and try changing these parameters:
- How does the outcome of the model change when you change the number of topics? How do models on 10-30-50-100 topics differ?
- Now, leave the number of topics the same and change min_len. How do models on 0-5-10-20 min_len differ?
- Play again with the parameters of the preprocessing function and see how this affects the results of the LDA model.
- For instance, what happens would the topics make sense if you had not removed the stopwords?
- You can also add new stopwords inside the extra_stopwords list if you feel that some words that appear in your topics do not provide much information.
- What happens if you don’t stem words?
- What if you remove numbers? Punctuation?