Community Insights

What does the community want or need? A free tool to discover community needs by analyzing public discourse​

The pulse of the community

Community voice and insights are critical to developing viable solutions for impact, and there are many widely used tools and approaches to listening to and capturing what communities want and need. A few examples include: interviews, surveys (including reviewing existing survey data), Appreciative Inquiry summits, and community roundtables.

Depending on who you are able to connect with, it’s possible the resulting data may not reflect the current pulse of the community or may only represent a narrow segment. You may be missing emerging trends or persistent issues that remain unidentified in surveys.

To expand methods for listening to community needs, NewImpact is in the early stages of developing Community Insights, a free and open research tool that aims to increase the range of community voices in the tri-sector innovation process and has broader application beyond tri-sector innovation.

Community Insights is a cloud-based tool designed to help uncover needs and issues emerging or persisting in the community conversation. A different perspective is gained by exploring organic conversation especially when considered over time. The tool’s backend uses Artificial Intelligence (AI) and Machine Learning to scrape and organize discourse from multiple sources. The user interface allows users to explore trends and read the actual discourse.

Background

Community Insights is a prototype tool ​designed and built through a partnership between NewImpact and the Urbanalytics Group at the University of Washington iSchool led by Bill Howe. The project was funded by grants from the Washington Research Foundation and The Civic Commons.​

For this initial prototyping stage of the product :

  • Data sources were curated by the NewImpact team to help focus on community needs rather than the wide range of discussion on social media

  • Machine learning models to label the data by impact area, sentiment, and type of discourse were developed by the UW team and trained on public data tagged by hand by the NewImpact team ​

  • While these models achieve high accuracy in evaluation, they may in some cases produce unusual results due to the inherent ambiguity of the task and various sources of uncertainty​

  • Code is available on GitHub

How it works

The Community Insights backend scrapes discourse data from various sources, then machine learning models analyze and label each piece of discourse with the following information:

  • Impact area (SDGs, SPIs, Civic Commons Scorecard for Shared Prosperity, and NewImpact roll-up of SDG/SPI/SSP)

  • Sentiment (positive, neutral, negative)

  • Type of discourse (comment, question, response, idea, proposal)

  • Source (Twitter, Reddit, public meeting comments, poll, etc.)

  • Geography (currently only tagging for King County, WA)

  • Date (date of discourse item)

  • Source URL (link back to the original post)

  • Related discourse (id’s of other discourse items in the thread, if exists)

This process provides structure to the data that supports uncovering trends and details in the community conversation.

No personally identifiable information is stored, but the tool does store and share links back to the original post on the original platform to enable understanding of context.

Currently the tool scrapes organic discourse from social media platforms of Twitter and Reddit. As the tool is refined it will expand to support other social media platforms, public data (comments, surveys, 311, etc.), and support informal polls using mobile short-codes and hashtag campaigns.

The Community Insights user interface features two views:

  1. A ‘Trends’ view allows users to explore trends in the data using interactive data visualizations. Visualizations display sentiment trends over time, frequent terms for selected date range, and impact areas represented in the discourse. 

Screenshot of top half of Community Insights Trends View. Visualizations display most frequent

terms in selected time period, as well as the sentiment trends for those terms. Click image for expanded view

Screenshot of bottom half of Community Insights Trends View. Visualizations display breakdown

of impact areas represented in the discourse. Click image for expanded view.

2. A ‘Verbatim’ view displays actual contents of discourse and supports searching/filtering by date, source, impact area, sentiment, and type.

Screenshot of Community Insights Verbatim View filtered for any discourse containing the string ‘vaccin’.

Content of discourse is displayed along with attributes. Click image for expanded view.

  • Hosted on AWS built with Amplify, Javascript, GraphQL, mysql, ML models, and scraping apis. Code available on GitHub

  • Current:

    • Limited to King County.

    • Limited data sources, three social media platforms, select accounts, manually uploaded public sector data and surveys.

    • No support for polls or prompts for community feedback.

    • Term frequency limits.

    • Text analysis only.

    • Limited to English.

    • Discourse classification and sentiment analysis.

    • Research tool with non-production quality code, limited to NewImpact use, partially built in Amplify.

    Future:

    • Expand geographies supported including wider range of admin units; display data on map.

    • Identify and integrate data sources for under-represented communities; add non-organic sources such as surveys, public meeting comments, expand social media platforms; dynamic discovery of new sources.

    • Interface and tools to support SMS polls and prompts for feedback.

    • Ability to identify topics and emerging trends; self training models to improve labeling and tagging quality.

    • Image analysis and insights included with text to capture increasing trend of discourse occurring in images.

    • Support multiple languages.

    • ML that more robustly labels discourse and sentiment.

    • Production quality code that performs well supporting significant increase in data and users.

  • AI/ML:​

    • Refine classification and labeling​.

    • Refine sentiment analysis​.

    • Analyze discourse data to identify topics and insights​.

    • Image analysis to gain insights from images/memes​.

    Database:​

    • Assess current architecture and make recommendations​.

    • Query optimization​.

    Data Science​

    • Assess discourse data elements and make recommendations​.

    • Assess visualizations and make recommendations​.

    • Help implementing recommendations.

    Data Sourcing​

    • Expand discourse data sources and types; self-training discovery​.

    • Help identifying discourse data for under-represented communities​.

    Data Aggregation ​& Scraping

    • Stabilize​

    • Add monitoring​.

    • Expand scraping to include more source types and sources.​

    Design​

TLDR

  • Community Insights is a cloud-based research tool in early stages of development that aims to capture voice of the community

  • Helps uncover needs/issues emerging or persisting in community conversation​

  • System gathers, processes, and visualizes public discourse from a range of sources including social media, public sector data, custom polls​

  • Discourse is analyzed and organized by impact area, sentiment, source, geography, date, type, etc.​

  • Supports exploring trends and individual posts/ discourse

  • Discourse links to original source

Needs Addressed by Community Insights

Orgs from all sectors:​

  • How can we get more representative input from the community we serve and work in?

  • What needs to be fixed?​

  • What’s working great and should be replicated or bolstered?​

  • How to hear individual and collective voice?​

  • Where are there bottlenecks or unmet needs?​

Community Needs:​

  • How can I see what my fellow residents care about?​

  • How can I be heard when I can’t make it to meetings, etc.?