COMP9414: Artificial Intelligence Assignment 2: Topic Classification
Due Date: Week 9, Friday, July 30, 11:59 p.m.
Value: 25%
This assignment is inspired by a typical real-life scenario. Imagine you have been hired as a Data Scientist by a major news organization. Your job is to analyse the news feed to determine the topic of incoming news articles so they can be organized and distributed to your readers.
For this assignment, you will be given a collection of BBC news articles and also summaries of the same articles. The articles have been manually labelled as one of five topics: business, entertainment, politics, sport and tech. Important: Do not distribute these news articles on the Internet, as this breaches BBC copyright.
You are expected to assess various supervised machine learning methods using a variety of fea- tures and settings to determine what methods work best for topic classification in this domain. The assignment has two components: programming to produce a collection of models for topic classification, and a report to evaluate the effectiveness of the models. The programming part involves development of Python code for data preprocessing of articles and experimentation of methods using NLP and machine learning toolkits. The report involves evaluating and comparing the models using various metrics.
You will use the NLTK toolkit for basic language preprocessing, and scikit-learn for feature con- struction and evaluating the machine learning models. You will be given an example of how to use NLTK and scikit-learn to define the machine learning methods (example.py), and an example of how to plot metrics in a graph (plot.py).
Data and Methods
A training dataset is a .tsv (tab separated values) file containing a number of articles, with one article per line, and linebreaks within articles removed. Each line of the .tsv file has three fields: instance number, text and topic (business, entertainment, politics, sport, tech).
A test dataset is a .tsv file in the same format as the training dataset except that your code should ignore the topic field. Training and test datasets can be drawn from supplied files articles.tsv or summaries.tsv (see below).
For all models, consider an article to be a collection of words, where a word is a string of at least two letters, numbers or the symbols #, @, , $ or %, delimited by a space, after removing all other characters (two characters is the default minimum word length for CountVectorizer in scikit-learn). URLs should be treated as a space, so delimit words. Note that deleting “junk” characters may create longer words that were previously separated by those characters.
Use the supervised learning methods discussed in the lectures: Decision Trees (DT), Bernoulli Naive Bayes (BNB) and Multinomial Naive Bayes (MNB). Do not code these methods: instead use the implementations from scikit-learn. Read the scikit-learn documentation on Decision Trees1 and Naive Bayes,2 and the linked pages describing the parameters of the methods.
1 https://scikit-learn.org/stable/modules/tree.html 2https://scikit-learn.org/stable/modules/naive bayes.html
Look at example.py to see how to use CountVectorizer and train and test the machine learning algorithms, including how to generate metrics for the models developed, and plot.py to see how to plot these metrics on a graph for inclusion in your report.
The programming part of the assignment is to produce DT, BNB and MNB models and your own model for topic classification in Python programs that can be called from the command line to train and classify articles read from correctly formatted .tsv files. The report part of the assignment is to analyse these models using a variety of parameters, preprocessing tools and scenarios.
Programming
You will submit four Python programs: (i) DT classifier.py, (ii) BNB classifier.py, (iii) MNB classifier.py and (iv) my classifier.py. The first three of these are standard models as defined below. The last is a model that you develop following experimentation with the data. Use the given datasets (articles.tsv and summaries.tsv) containing 1000 labelled articles and their summaries to develop and test the models, as described below.
These programs, when called from the command line with two file names as arguments, the first a training dataset and the second a test dataset (i.e. not hard-coded as training.tsv and test.tsv), should print (to standard output, not a hard-coded file output.txt), the in- stance number and topic produced by the classifier of each article in the test set when trained on the training set (one per line with a space between the instance number and topic) – each topic being the string “business”, “entertainment”, “politics”, “sport” or “tech”. For example:
python3 DT classifier.py training.tsv test.tsv > output.txt
should write to the file output.txt the instance number and topic of each article in test.tsv, as
determined by the Decision Tree classifier trained on training.tsv.
When reading in training and test datasets, make sure your code reads all the instances (some
Python readers use “excel” format, which uses double quotes as separators).
Standard Models
You will develop three standard models. For all models, make sure that scikit-learn does not convert the text to lower case. For Decision Trees, use scikit-learn’s Decision Tree method with criterion set to ’entropy’ and with random state=0. Scikit-learn’s Decision Tree method does not implement pruning, rather you should ensure that Decision Tree construction stops when a node covers fewer than 1% of the training set. Decision Trees are prone to fragmentation, so to avoid overfitting and reduce computation time, for the Decision Tree models use as features only the 1000 most frequent words from the vocabulary, after preprocessing to remove “junk” characters as described above. Write code to train and test a Decision Tree model in DT classifier.py.
For both BNB and MNB, use scikit-learn’s implementations, but use all of the words in the vocabulary as features. Write two Pythons programs for training and testing Naive Bayes models, one a BNB model and one an MNB model, in BNB classifier.py and MNB classifier.py.
Your Model
Develop your best model for topic classification by either varying the number and type of input features for the learners, the parameters of the learners, and the training/test set split, or by using another method from scikit-learn. Submit one program, my classifier.py, that trains and tests a model in the same way as for the standard models. Conduct new experiments to analyse your model and present results that justify your choice of this model in the report.
Report
In the report, you will first evaluate the standard models, then present your own model. For questions 1–4 below, consider two scenarios:
(1) with the full articles in articles.tsv for training and testing, and
(2) with the summaries in summaries.tsv for training and testing.
For evaluating all models, report the results of training on the first 800 instances in the dataset (the “training set”) and testing on the remaining 200 instances (the “test set”), rather than using the full datasets of 1000 instances for training – so the 1% stopping rule for Decision Trees is when nodes cover less than 8 instances rather than 10.
Use the metrics (micro- and macro-accuracy, precision, recall and F1) and classification reports from scikit-learn. Show the results in Python plots (do not take screenshots of sklearn classification reports), and write a short response to each question below. The answer to each question should be self contained. Your report should be at most 10 pages. Do not include appendices.
1. (3 marks) Develop Decision Tree models for training and testing: (a) with the 1% stopping criterion (the standard model), and (b) without the 1% stopping criterion.
(i) Show all metrics on the test set for scenario 1 comparing the two models (a) and (b), and explain any similarities and differences.
(ii) Show all metrics on the test set for scenario 2 comparing the two models (a) and (b), and explain any similarities and differences.
(iii) Explain any differences in the results between scenarios 1 and 2.
2. (3 marks) Develop BNB and MNB models from the training set using: (a) the whole vocabulary (standard models), and (b) the most frequent 1000 words from the vocabulary, as defined using sklearn’s CountVectorizer, after preprocessing by removing “junk” characters.
(i) Show all metrics on the test set for scenario 1 comparing the corresponding models (a) and (b), and explain any similarities and differences.
(ii) Show all metrics on the test set for scenario 2 comparing the corresponding models (a) and (b), and explain any similarities and differences.
(iii) Explain any differences in the results between scenarios 1 and 2.
3. (3 marks) Evaluate the effect of preprocessing for the three standard models by comparing models developed with: (a) only the preprocessing described above (standard models), and (b) applying, in addition, Porter stemming using NLTK then English stop word removal using sklearn’s CountVectorizer.
(i) Show all metrics on the test set for scenario 1 comparing the corresponding models (a) and (b), and explain any similarities and differences.
(ii) Show all metrics on the test set for scenario 2 comparing the corresponding models (a) and (b), and explain any similarities and differences.
(iii) Explain any differences in the results between scenarios 1 and 2.
4. (3 marks) Evaluate the effect of converting all letters to lower case for the three standard models by comparing models with: (a) no conversion to lower case, and (b) all input text converted to lower case.
(i) Show all metrics on the test set for scenario 1 comparing the corresponding models (a) and (b), and explain any similarities and differences.
(ii) Show all metrics on the test set for scenario 2 comparing the corresponding models (a) and (b), and explain any similarities and differences.
(iii) Explain any differences in the results between scenarios 1 and 2.
5. (5 marks) Describe your chosen “best” method for topic classification. Give new experimental results for your method trained on the training sets of 800 articles/summaries and tested on the test sets of 200 articles/summaries. Explain how this experimental evaluation justifies your choice of model, including settings and parameters, against a range of alternatives. Provide new experiments and justifications: do not just refer to previous answers.
Submission
• Make sure your name and zid appears on each page of the report
• Submit all your files using a command such as (this includes Python code and report):
give cs9414 ass2 DT*.py BNB*.py MNB*.py my classifier.py report.pdf
• Your submission should include:
– Your .py files for the specified models and your model, plus any .py “helper” files
– A .pdf file containing your report
• When your files are submitted, a test will be done to ensure that one of your Python files
runs on the CSE machine (take note of any error messages printed out)
• When running your code on CSE machines:
– Set SKLEARN SITE JOBLIB=TRUE to avoid warning messages
– Do not download NLTK in your code: CSE machines have NLTK installed
• Check that your submission has been received using the command: 9414 classrun -check ass2
Assessment
Marks for this assignment are allocated as follows: • Programming (auto-marked): 8 marks
• Report: 17 marks
Late penalty: 5 marks per day or part-day late off the mark obtainable for up to 3 (calendar) days after the due date.
Assessment Criteria
• Correctness: Assessed on standard input tests, using calls such as:
python3 DT classifier.py training.tsv test.tsv > output.txt
Each such test will give two files, a training dataset and a test dataset, which contain any number of articles (one on each line) in the correct format. The training and test datasets can have any names, not just training.tsv and test.tsv, so read the file names from sys.argv. The output should be a sequence of lines (one line for each article) giving the instance number and classified topic, separated by a space and with no extra spaces on each line or extra newline characters following any newline character after the last classification. There are 2 marks allocated for correctness of each of the three standard models.
For your own method, 2 marks are allocated for correctness of your methods on test sets of articles that include unseen examples.
• Report: Assessed on correctness and thoroughness of experimental analysis, clarity and succinctness of explanations, and presentation quality.
There are 12 marks allocated to items 1–4 as above, and 5 marks for item 5. Of these 5 marks, 1 mark is for the description of your model, 2 marks are for new experimental analysis of your model, and 2 marks are for the justification of your model using new analysis. In general, if the presentation is of poor quality, at most 50% of the marks can be obtained.
Plagiarism
Remember that ALL work submitted for this assignment must be your own work and no code sharing or copying is allowed. You may use code from the Internet only with suitable attribution of the source in your program. Do not use public code repositories on sites such as github – make sure your code repository, if you use one, is private. All submitted assignments will be run through plagiarism detection software to detect similarities to other submissions, including from past years. You should carefully read the UNSW policy on academic integrity and plagiarism (linked from the course web page), noting, in particular, that collusion (working together on an assignment, or sharing parts of assignment solutions) is a form of plagiarism.
DO NOT USE ANY CODE FROM CONTRACT CHEATING “ACADEMIES” OR ONLINE “TUTORING” SERVICES. THIS COUNTS AS SERIOUS MISCONDUCT WITH A HEAVY PENALTY UP TO AUTOMATIC FAILURE OF THE COURSE WITH 0 MARKS.