You will implement an end-to-end document classi?cation system that predicts which category pages belong to, using the classi?cation scheme.
Your system will use the averaged perceptron machine learning algorithm, which you will implement. You will test your implementation of the learning algorithm on a pre-computed dataset, so that you can see whether your learner performs as expected.
Once you have your learner, you will apply it to the article classi?cation task, using features you design and extract yourself. You will evaluate your classi?er using the n-fold cross validation technique, which you will also implement.
Finally, you will describe your experiments in a three-page report, which you will submit alongside a tarball/zip?le including your code and instructions to run your system.
You are free to use a programming language of your choice to implement the assignment.
The assessment of this assignment is not about the quality of your code. Rather, it is about how well you can set up, evaluate and analyse a typical statistical natural language processing experiment.
However, the correctness of your code will prove critical in producing intelligible results: if you do not implement your learner, extractor and evaluator correctly, you will produce results that are impossible to explain.
An important part of this assignment is learning to identify and describe relevant details. There are an almost limitless combination ofmeasures you can use or experiments you can do to analyse how your system performs. However, space is limited, so you must be selective. Once you have a correct implementation, asking the right questions and using statistics that answer them concisely is the key to good marks.
You will be assessed on a 3-page report (not including tables and/or diagrams) that describes and analyses your results. You are not required to describe your implementation in the report.
The analysis of the results of the ?rst machine learning problem should be brief. This experiment is to help you verify the correctness of your implementation.
Most of your report should describe your article classi?cation experiment. Describe which features you included, and identify which types of features were most important for your classi?er's accuracy. Characterise the kinds of errors the system made, using some combination of qualitative and quantitative analysis.
Although in general the choice of how to present your results is up to you, you must include micro-averaged Precision, Recall and F-Measure statistics using 10-fold cross validation for the article classi?cation task. You are encouraged to evaluate a baseline con?guration using only the most obvious features (such as bag- of-words), and analyse the contribution of more innovative features individually.
Download:- statical natural language processing.zip