Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

M.S.

Spring Semester Final Exam 2020


CSE 537: Text Mining
Department of Computer Science and Engineering
University of Dhaka
Answer any 3 questions.
1. a. What are the minimum and maximum possible values of the conditional 2+2
entropy H(X|Y)? Under what situations do they occur?
b. I. Explain with example how basic vector space model can be 4+2
used for Paradigmatic relation mining. +6
II. What are the shortcomings of this method of mining
Paradigmatic relations?
III. What techniques can be adopted to make the mining method
robust against these shortcomings?
c. Among H(X|Y), H(Z|Y) and H(X|Z), which are comparable and which 4
are not. Explain why.

2. a. Given a training dataset 1+2+


𝑥 ,𝑦 , 𝑥 ,𝑦 ,⋯ 𝑥 ,𝑦 7
where, 𝑥 ∈ 𝑅 , 𝑦 ∈ 0,1 and 𝜃 ∈ 𝑅 is the parameter vector,
answer the following question:
I. If we want to use Logistic regression, then what is the range
of values for ℎ 𝑥 ? What does it actually measure (write
using probability notation)?
II. If we define ℎ 𝑥 𝑔 𝜃 𝑥 , then draw the graph of
𝑔 𝜃 𝑥 .
III. Show step by step derivation of the objective function of the
aforementioned model. Also define the intuition behind this
objective function using graphs.

b. Say in a text categorization problem your algorithm will categorize N 3+ 7


documents into k categories. Answer the following questions:
I. Why Classification Accuracy is not a good measure to
evaluate performance of your algorithm?
II. How can you measure the average performance of your
algorithm over all categories and over all documents?
Describe with dummy examples.
3. a. Give a training dataset 5+1
𝑥 ,𝑦 , 𝑥 ,𝑦 ,⋯ 𝑥 ,𝑦
𝑤ℎ𝑒𝑟𝑒, 𝑥 ∈ 𝑅 𝑖𝑠𝑎𝑠𝑒𝑛𝑡𝑒𝑛𝑐𝑒 ∧ 𝑦 ∈
1,2, ⋯ , 𝑘 𝑖𝑠𝑡ℎ𝑒𝑟𝑎𝑡𝑖𝑛𝑔𝑜𝑓𝑡ℎ𝑒𝑠𝑒𝑛𝑡𝑒𝑛𝑐𝑒,
answer the following question:
I. Design a multiple logistic regression classifier for rating
prediction of any sentence.
II. How many parameters are there in your model?

b. What are the problems in using a single term as a topic in topic mining? 3
Define with example.
c. I. If we want to train a generative probabilistic topic model on 3+1
our corpus of 𝑁 documents with 𝑉 be the set of vocabulary +1
and we want to consider 𝑘 topics, then what shall be the
parameters of our model and what would they measure?
Define with small example.
II. How many parameters are there in total?
III. What constraints they must follow?
d. In the text categorization problem, when labeled data is available three 6
categories of classifiers can be used to solve the problem. Write their
names with short description on how they work.

4. a. Describe the unigram language model for discovering one topic 10


b. Explain the multilevel logistic regression algorithm for sentiment 10
analysis

5. a. How can a term be represented as term vector based on the words in the 10
context? How does it help to discover paradigmatic relations?
b. Mention the advantages of a neural language model. Describe the skip- 10
gram neural language model in detail.

You might also like