Description
This is a GROUP assignment. This document contains two project descriptions and a fine print section. Your group will select and complete ONE of these projects – with the end and graded result being a project report (to be clear, we will not be reviewing your code or data – only the final report document).
Project 1: COVID in Canada
Overview
Your team has been hired by a not-for-profit that wants to understand the discussions currently happening around COVID in Canadian social media. They have indicated that they are especially concerned with vaccine hesitancy. Specifically, they want to know:
- The salient topics discussed around COVID and what each topic primarily concerns
- Relative engagement with those topics
- How positive/negative the response to the pandemic/vaccination has been.
You will conduct this analysis and submit a report discussing your findings.
Analysis Details
Your analysis will draw on Twitter posts (tweets). To inform your analysis, you should collect 1,000 tweets within a 3 day window. You should set filters such that all 1,000 posts mention either COVID, vaccination, or a name-brand COVID vaccine AND all are in English (ensure that the language field is set to “en”… this isn’t exact, but it gets close). You can filter by hashtags or words when collecting Twitter data. You can choose the exact words, as long as they are related to the context that we mentioned before.
Each post in your collection should be unique – meaning that you shouldn’t include an identical tweet or retweet twice.
To develop your topics, conduct an open coding on 200 tweets (approach the exercise requiring each tweet to belong to exactly one topic). You should aim for between 3-8 topics in total.
Once your topics have been designed, manually annotate the rest of the 1,000 tweets in your dataset. During this annotation, also code each post for positive/neutral/negative sentiment. While double annotation would usually be used, for this project (given time constraints), use single annotation. While double annotation would usually be used, for this project (given time constraints), use single annotation.
Characterize your topics by computing the 10 words in each category with the highest tf-idf scores (to compute inverse document frequency, use all 1,000 posts that you originally collected.
Project 2: Movie Release
Overview
Your team has been hired by a media company that wants to understand the discussions currently happening around the film “(insert a recently-released movie that your team selected here)”. They have indicated that they are especially concerned with the favorability of the audience response. Specifically, they want to know:
- The salient topics discussed around their film and what each topic primarily concerns
- Relative engagement with those topics
- How positive/negative the response to the movie has been
You will conduct this analysis and submit a report discussing your findings.
Analysis Details
Your analysis will draw on Twitter posts (tweets). To inform your analysis, you should collect 1,000 tweets within a 3 day window. You should set filters such that all 1,000 posts have a very high likelihood of being related to the movie AND all are in English (ensure that the language field is set to “en”… this isn’t exact, but it gets close). You can filter by hashtags or words when collecting Twitter data. You can choose the exact words, as long as they are related to the context that we mentioned before.
Each tweet in your collection should be unique – meaning that you shouldn’t include an identical tweet or retweet twice.
To develop your topics, conduct an open coding on 200 tweets (approach the exercise requiring each tweet to belong to exactly one topic). You should aim for between 3-8 topics in total.
Once your topics have been designed, manually annotate the entire set of the 1,000 tweets in your dataset. During this annotation, also code them for positive/neutral/negative sentiment. While double annotation would usually be used, for this project (given time constraints), use single annotation.
Characterize your topics by computing the 10 words in each category with the highest tf-idf scores (to compute inverse document frequency, use all 1,000 posts that you originally collected.
Final Report Details
Your report should be written using the AAAI format found here (you may use either Word or Latex). The template formatting (e.g., font, font size, spacing, citation style) should be followed strictly. The report structure should consist of the following sections (the lengths are suggestions):
- Introduction (0.5 page) – general overview and key findings
- Data (0.5 page) – describe your dataset. This should include statistics relevant to the project – the number of tweets you originally started with, the keywords used to
collect tweets, and any design decisions you had to make around the filtering of this content.
- Methods (0.5 page) – explanation and justification for what you did. Focus on the design decisions you made NOT listed in this document that impacted your results.
- Results (1 pages) – share all your findings including the topics selected (and their definitions), topic characterization, and topic engagement.
- Discussion (1 pages) – interpret your results in terms of what they reveal about the way each candidate was being discussed and perceived. Make extensive use of your results to justify your interpretations.
- Group Member contributions (0.25 page) – a description of the contributions each group member made to this project.
- References (< 1 page) – this is an optional section should you reference other works in your report.
The report must be between 5 and 7 pages in length, not including references. Figures are encouraged – but should be used to maximum effect (fluffy or otherwise unnecessary images that do not make strong contributions to the report will lead to point deductions).
Fine Print
- Each group will submit one report which will receive one grade that all members of the group will share. The one exception to this is in the case of strong evidence of delinquent group members. In this case, each member’s grades may be adjusted up or down as appropriate.
- While there are no rules about how work should be divided up, good team participation and fair sharing of the workload are absolutely expected in this project.
FAQ
- You can use https://www.overleaf.com/ if you decide to work with LaTeX
- There are multiple libraries for fetching Tweets from Twitter API. You can use anything you like, in any language you like. TAs will support you with Python specific tools.
