search

herospanda / Thesis_SijevanderVeen.

0 stars 0 forks source link

Delivery documents go or no go #6

Open herospanda opened 5 years ago

herospanda commented 5 years ago

Dear Maarten @maartenmarx ,

I uploaded the documents which where are ask in previous issue.

Report.pdf contains the report for so far. The introduction and related work is finished. Also two sub-questions are finished and added to the results section.

The form for the sub-questions is added in subquestions.pdf. The subquestions are also explained in the introduction of the report.

The notebooks for the two sub-questions are added (subquestion1.ipynb and subquestion2.ipynb).

Met vriendelijke groet,

Sije van der Veen

maartenmarx commented 5 years ago

Beste @sije91 ,

Het spijt me je mede te moeten delen dat ik geen go kan geven voor je scriptie. Hieronder is mijn review en korte motivatie voor dit oordeel. Dit betekent dat je niet verder kan met dit project en voor je scriptie een nieuw project zal moeten starten.

met vriendelijke groeten Maarten Marx

Review

Introduction and RQs

The thesis introduces the work and motivates why labelling tweets is important for RNW. All language technology has disappeared from the research questions and now they are only about LDA topic modelling.

My main concern is that questions 1 and 2 ask for accuracy and optimal number of LDA topics and measure that by two properties of the learned probability distributions: perplexity and coherence. It is not explained how or why these two rather technical measures can give us any insight in the usefulness of LDA for RNW. Also the questions RQ1 and RQ2 are not answered.

The graphics are hard to interpret as the dependent variable is put on the x-axis.

Related work

The related work is just a background section on LDA. Not on th eresearch topic of this thesis: trying to automatically learn labels for Tweets in an African language by using clustering.

Notebooks

The notebooks seem fine, although they cannot be reproduced as the data in not in the repo.

Final decision

We believe it is not possible to succesfully defend the Master thesis with the current research problem and provided data. Therefore we do not give a go for this thesis.