# FOLT Gedächtnisprotokoll
## Aufgabe 1 Terminology (15 pkt)
Define and provide example per term
a)
Tokenization
b) BIO Tagging
c) Word Embeddings
d) Named Entities
e) Constituent Tree
f) WordNet
g) BERT
h) What is the Difference between POS Tagging and Dependency Parsing
I) Naives Bayes Assumption
j) Provide an example where context is important in NER
k) What is Cohens Kappa
l) WordSim353 (.....wtf)
## Aufgabe 2 Python (8 pkt)
a) Explain python code and turn into list comprehension:
```python=
output_list = []
for word in input_list:
if 'a' in word:
output_list.append('a')
else:
output_list.append('b')
```
b) What does this code (string manipulation, string indexing) print
```python=
v = 'Masked '
print(v)
v = v + 'E.'
print(v)
v = v[:-1] + 'lon'
print(v)
```
```python=
def f(arg1, arg2):
if True:
arg1[0]=True
for i in range(99):
arg2 = i
l = [1,1,1]
v = 50
f(l, v)
print(l,' - ',v)
```
c) Describe 3 Regex And provide an Example
Another one was: Count how often characters appear in a list of words & which of the characters appear in every word. Don't use nltk classes, just use python dict.
## Aufgabe 3 Classifikation (11 pkt)
b) 2 ppl want to to write detection model for FakeNews spread by bots on twitter. Person A collects 10 samples by one bot and gives them to their friend to label these. What is wrong with this approach and how could you improve it (min. 2 things)?
d) Write down confusion matrix and calculate precision accuracy recall f1
## Aufgabe 4 Deep Learning (10 pkt)
a)
How many layers does this MLP have?
b)
What function does the MLP calculate?
c)
How many nodes in inner layer?
d)
??
e)
Matrix size of weights after input
Sketch Recurrent Neural Network with output and something with a rationale
f)
What does a RNN do better than MLP for Natural language processing tasks?