Monitoring Online Discourse During An Election: A 5-Part Series
How to monitor social media with AI-based tools during an election campaign
Traditional election monitoring is a formalized process in democratic countries, set out in the mandate of the national Electoral Management Body (EMB). As social and digital cultures change, however, EMBs are finding it useful to expand their monitoring capacities to include social media.
Given the media coverage of interference with the 2016 US and UK elections, and the fallout from the Cambridge Analytica debacle, politicians and the public are wary of the impact social media manipulation can have on electoral processes.
As well, automated tools like bots, created locally or outside the country, disrupt the existing system. They amplify political messaging, yet are not currently covered by political financing regulations; and they can disseminate disinformation.
Tracking social media allows an EMB to stay on top of operational issues during an election period (see Part III: Managing Operational Issues) and also to detect and track disinformation and its spread (see Part II: Identifying Disinformation).
This Playbook is designed for EMBs. As a template, it will obviously need to be adapted, depending on jurisdiction. It describes social media monitoring as a function within an EMB, and assumes that the EMB has access to a AI-based social media monitoring tool, such as KI Social.
To be effective, this process should be put in place well before the first milestone of the election period.
|1. Setting up
Before the project begins:
· Ensure there is clarity regarding the goals of the social media monitoring. A key issue is scope: does the EMB mandate include monitoring of operational issues, or disinformation and misinformation, or all of the above? (see Part III: Managing Operational Issues).
· Does the mandate include tracking voter issues expressed within national borders, or will it also include voters travelling or living abroad? This is important for the many nations with large expatriate communities. If results should include social media posts from voting citizens residing outside their country, it won’t be possible to limit data by geographic location.
· Will the monitoring function be active (occurring continuously), passive (occurring once a week, for example), or retroactive (taking place after each election milestone is completed)?
· In this phase, EMBs should inventory its internal staff capacity; for example, does its staff include social media content producers, policy personnel, social media analysts, social media monitors, or data scientists? If not, arrange with an experienced vendor such as KI Design to provide these services.
Technical set-up: The technical team within the EMB should document:
a. All EMB web and social media assets
b. All relevant national and international news sites
c. Lessons learned from the previous election
d. A list of all confirmed candidates when it is finalized
e. All political party data and web assets
f. Details of political spending on Twitter, Reddit, Facebook pages, and other platforms
|2. Acquire a social media service provider with the following capabilities:
a. Full firehose access to data, going back at least to one previous election. It’s vital to be able to analyse data from the previous election, to understand what potential issues may occur in the current one. For example, there may be specific complaints related to a particular location, or to the capacity of polling station staff. That said, historic analysis will not provide a complete picture; new issues and ambiguities will arise.
b. Ability to track keywords that are not necessarily related to elections; for example, power outages, roadblocks, protests, etc.
c. Geolocation capacities:
i. For posts with geolocation tags, the tool should display post locations on a map. For example, posts may complain about ballot non-delivery in a certain region.
ii. For posts without geolocation tags, the tool should have the ability to group them and map them visually, to show concentration; for example, a post without a geolocation tag may state “unable to find [named] polling station on EMB website”; the visualisation component is important so that logistical issues can be dealt with collectively rather than as individual instances.
d. The tool should permit custom views for various EMB staff skillsets. For example: content producers would want to measure the volume of incoming and outgoing messages on the EMB’s social media channel; data scientists may want to write sophisticated queries; issue managers may need views that show whether or not posts have been responded to.
e. Your vendor should be able to provide data science analytics and application dashboard customization expertise.
f. Your vendor needs to have experience in provisioning data science services to EMBs.
|3. Noise elimination
A query contains an expression that’s composed of keywords, emojis, and urls to be tracked. These will include both keywords you are looking for, and many keywords that you don’t want to be in the search. For example: In a national election, if the query contains the phrase “election monitor,” the result could include any and all election monitoring occurring anywhere in the world, as well as any election monitoring within municipalities, cities, unions, associations, or the UN, or from other regional-level elections.
In an election context, without noise elimination, some 90% of the results of a query are irrelevant. Hence, a large portion of the query should be dedicated to eliminating these irrelevant results.
Issues to be aware of include:
a. Elections in other countries may be taking place concurrently; for example, an Indian provincial election and a national UK election. Especially if both countries share a common language, online discourses may include overlapping content; for example, place names or street names.
b. Name similarities of candidates with other citizens.
c. Election talk on social media will be dominated by countries with higher per capita access to the Internet, and in particular those whose citizens most frequent Twitter; such as the US, UK, and France.
d. In multi-lingual countries, where many unofficial languages are spoken, queries should aim to capture election discourse in languages other than the official one/s. With that comes the need for noise elimination related to the nation/s where those other languages are dominant.
|4. Your social media monitoring tool should include three distinct functions:
a. Dashboards and reports: Real time, periodic (e.g., every four hours), daily, weekly, or monthly
b. Data feeds: Each feed is dedicated to:
i. Operational issue/s
ii. Capturing of EMB’s footprint; this means that this feed would be dictated to finding any occurrences of posts that mention the EMB, its leadership, or the relevant legislation
iii. All parties and all candidates (including any events or investments or announcements by the political parties)
iv. Data feeds specific to disinformation and misinformation
c. Alerts of any media mentions that are of particular interest to the EMB
|5. Create specific filters to target election milestones
a. Content regarding election steps prior to election day:
i. Voter registration
ii. Ballot mailing
iii. Citizens moving residence
iv. Allowed pieces of identification
v. Election date
vi. References to bias by EMB officials
vii. Impersonation of EMB or political candidates
b. On voting days (advance polling and election day):
ii. Availability of paper ballots
iii. Registration issues
iv. Staff issues
v. Directions to polling station
vi. Power outages
vii. Poll relocation
c. Ballot counting hours: Analysing concerns and content appearing after polling stations are closed and before the results were issued.
d. Post-election reporting: providing aggregate data on the monitoring activity and the number of situations averted, mitigated, or responded to.
Part of a 5-part series on
Monitoring Online Discourse During An Election:
PART FIVE: Monitoring Political Financing Issues