Write up plan for how speakerbox + cuequeue-lite and current CDP data integrate for larger paper

[evamaxfield]

• sentiment and topics for council vs sentiment and topics for public commenters
• Sentiments and topics per council member
• lead to sentiment and topics in social media in that area
• Which council members are most "aligned" with social media

[evamaxfield]

Work, Tools, and How it Connects

Speakerbox

Allows For

• Segmentation of dataset into each known and unknown speaker (i.e. "give me all the sentences from a specific councilmember")
• At a bare minimum, can be used to measure who is speaking in which meetings and how much.
• Needed for "Justice, Interrupted" replication ("Policy, Interrupted: how does gender, seniority, and ideology affect council deliberation")

Current Progress

• :heavy_check_mark: Working prototype of model training and application

Upcoming Work

• :arrow_forward: Need more data for training and evaluation
• :arrow_forward: Need documentation and general workflow for productionization (undergrads?)
• :arrow_forward: :arrow_forward: Apply to event gather pipeline for all instances

Interjection Classifier

Allows For

• Automated labeling of interruptions between individuals
• Needed for "Justice, Interrupted" replication ("Policy, Interrupted: how does gender, seniority, and ideology affect council deliberation")

Current Progress

• :x: Not started
• Minimal ideation in: https://github.com/BITS-Research/CDP-PolicyInterrupted

Upcoming Work

• :arrow_forward: Need dataset of interjections (labeled both as interjection but reason for interjection too -- "clarification", "talking over", etc.)
• :arrow_forward: Train model to classify sentences as interjections
• :arrow_forward: Model should work for all instances, no need for instance specific, apply during pipeline

CueQueue Lite

This is a good project for undergrads

Allows For

• Automated segmentation of public comment portion of meeting from the rest of the meeting
• Can be used to understand topics and sentiment from commenters vs council
• Municipal Latency

Current Progress

• :x: Not started

Upcoming Work

• :arrow_forward: Need dataset of start and end points of public comment
• :arrow_forward: Train classifier from dataset to label "start", "end", and "other" -- like CueQueue, these can be sentence windows instead of just single sentences.
• :arrow_forward: This model should be general enough to apply to all instances -- apply in pipeline

CueQueue

Allows For

• Complete segmentation of meeting by minutes item portions
• Understand sentiment difference between public commenters on a specific topic / minutes item and discussion from council
• Better data linking between matter / minutes item and transcript
• Future models related to matters (using discussion data for matter models)

Current Progress

• :x: Out of date prototype
• Paper proposal written and detailed

Upcoming Work

• See proposal paper

Municipal Latency

Unstructured

Develop a method which looks at the ngrams used in social media for a municipal area and measure the lag time between discussion about a topic on social media to discussion in public comment and then finally discussion by council.

Social Media -> Public Comment -> Council

We can technically start this without the social media aspect.

In combination with Speakerbox, we could also see which councilmembers have roughly the same sentiment as their constiuents AND are quick to follow their constiuents priorities.