wzheng-94/macs30200proj

individual course project on predicting news popularity

macs30200proj's Issues

Poster Evaluation

Design:

I loved the design of your poster! In particular the way it allowed you to guide me through it verbally step-by-step. It flows in a very logical fashion.
Suggestion: Include only the top non-zero predictors. It is a slightly overwhelming number currently included, in my opinion.

Viz:

Beautiful graphs - very eye catching. Also, very clear exposition of the results through the visualization.
Suggestion: Lining up the two blue graphs on the left gives the impression that they are connected, when I believe that they are not.

Content:

Highly innovative use of buzzy websites (Kaggle, Reddit). You're knowledge of the domains is really strong, and you have crafted some interesting insights.
Suggestion: Explain more clearly the importance of the non-zero predictors for reaching your conclusion.

Proposal feedback

• Are you looking at specific articles or a news organization?
• I see the potential for selection bias - not all news sources are posted to Reddit. Reddit skews young, more educated, white, and liberal, which effects the potential sources of news submitted to the subreddit and the number of up and downvotes
• I also don't see a clear operationalization of concepts in your proposed model. How will you measure agency and culture?

Poster evaluation - Hyun Ki Kim

Hi Weiwei, this is Hyun Ki Kim, your dear friend, writing a peer evaluation for your poster. I think your poster was great overall. It had most of the necessary parts for the good poster that we learned from class (title, name, affiliation, introduction, data, methods, results). However, I feel like your poster focused too much on the variables, considering different background knowledge audiences had in yesterday’s poster session. Also, the size of the title and your name are too small, and logos are too big. I think title is the most important of your poster, so I would recommend having bigger title. I also think the font size of the texts in the body are too small. It could be better if you focus on fewer, but more important variables instead of presenting all the variables. Also, I think where you put github repository address and putting “Email:” in front of your email are weird. However, I like your title because it showed what your research question is and how you are trying to solve clearly. Introduction part is also written well. Design looks nice with unevenly divided poster into 3 columns with a clear white background. Visualization looks nice too. However, I have feeling that most students had put too much detail into their poster when the audience’s background knowledge differ. Overall, I think it your poster was great.

poster comment

Hi Weiwei,

Glad that I got a chance to hear your presentation even though it was at the last minute. Your project was quite methodologically advanced and theoretically sound. Here are my comments on your poster:

In terms of design and layout: the poster was tidy and informative. You included all necessary sections from theory grounding though data and method to conclusion. I also liked the division and the university logos as well as your incorporation of the github repo address. That said, I think you can use more visualizations instead of texts – this way you can catch people’s eyes easily without overwhelming them. See next section for more details.

In terms of visualization: the six graphs effectively compared your different models and showed us the importance of network structure. However, it would have been better if you also visualize some other textual parts. For instance, when you presented and compared the coefficients of non-zero predictors, you could probably try to visualize this using, for instance, diverging dot plot?

In terms of content: it was really impressive that you incorporated so many different analysis and methods, including textual (both lexical and sentiment!) and network analysis. One suggestion: when you present the relationship between upvotes and network density/comment karma/lexical diversity, it was not very obvious that network matters much more than texts, especially when all of them did not present a linear relationship – you may want to fit a curvilinear regression (possibly quadratic) to show the relationships in a better way.

Overall, I can see all the efforts you have put into your project and I really liked it!
Yinxian

Poster evaluation

Hi Eric, this is an interesting topic! Here are some of my thoughts on your poster.

Design/layout
The design is great and clear. It is well-balanced on graphic content and textual content. The research method is on the top middle part while the result is on the top middle part, so it is easy to know what you are doing in a short time.

Visualization
The visualization is good. You showed how network density, Hub's karma, and lexical diversity are related to the number of upvotes. Also,
Suggestions: I wish you could add legends to your network graphs and label y-axis and x-axis. It is confusing when just looking at your poster.

Content
The content is good. However, you used R-square as a metric on your test set. This is not what people commonly do (correct me if I am wrong). See https://stackoverflow.com/questions/25691127/r-squared-on-test-data. Also, you are trying to build a prediction model but this seems like an explanatory model. There is little information about how your model predicts something. Finally, I think there is a typo in your method section on the equation of ridge regression.