6. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Narratives
Narratives are what you believe; stories are how you communicate that.
● Stories don’t have to be true or believed. They can be used as signals
● We tell ourselves stories about
○ who we are and want to be (identity narratives);
○ who we belong to and don’t belong to (in-groups and out-groups);
○ what our world is, and
○ what is happening around us.
Narratives define us as individuals and groups: families, communities, nations.
6
8. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Stories carry meaning, not truth
“the currency of story is not truth, but meaning.
That is, what makes a story powerful is not necessarily facts, but how the story creates
meaning in the hearts and minds of the listeners.
Therefore, the obstacle to convincing people is often not what they don’t yet know but
actually what they already do know.
In other words, people’s existing assumptions and beliefs can act as narrative filters to
prevent them from hearing social change messages.”
- Beautiful Trouble
8
20. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Marketing: Brand Metrics
● Share of Voice - how much this brand is mentioned relative to competing brands.
● Consumer engagement
● Net sentiment (against competitor brands), and co-occurrence of brand with specific positive concepts (“brand health”),
across consumers and influencers (and the deltas between these).
● Volume (against competitor brands)
● Message uptake
● Issue tracking - how much this brand is mentioned in conjunction with other narratives, and the themes inside those
narratives.
● Relative visibility of brand executives (CEO etc), both relative to each other, and to executives from competing brands.
● Posts, reach, and engagement of mentions by influencers (journalists etc).
○ paid media is bought by the brand (paid search optimisation, advertising online, in print, TV, direct mail, affiliate
marketing),
○ owned media is controlled by the brand (own websites, blogs, mobile apps, social media accounts, brochures,
stores),
○ earned media is publicity generated through PR (targeting influencers, creating word of mouth discussion
through engaging in social media, community conversations, blogs, and other user-generated content, brand
advocates, and viral marketing in these spaces).
● The simplest sets of brand measures are counts of brand mentions, news mentions, social mentions, and
engagement, and the changes in these over time.
20
21. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Basics: election example
● What: rumours from people’s uncertainties.
○ based on truth and community stories, unknown untruths, community
signals, false information, false contexts, existing narratives.
● Who: people looking for information, political operators, countries
● When: before, during, after events
● How: seeded across user-generated content sites, nation state-controlled media,
influencers
● Why: Goals include boosting one candidate/party, reducing support for other
candidates/parties, creating panic, preventing people from voting, reducing trust
in the election system, etc.
21
23. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
4D Model of disinformation campaigns (Nimmo)
Dismiss: if you don’t like what your critics say, insult them.
Distort: if you don’t like the facts, twist them.
Distract: if you’re accused of something, accuse someone else of the same thing.
Dismay: if you don’t like what someone else is planning, try to scare them off.
31. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Prebunking
“Inoculates” people against disinfo
FirstDraft lists 3 types:
● fact-based: correcting a specific false
claim or narrative
● logic-based: explaining tactics used
to manipulate
● source-based: pointing out bad
sources of information
Image: https://firstdraftnews.org/articles/a-guide-to-prebunking-a-promising-way-to-inoculate-against-misinformation/ 31
34. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Engage: Dennett’s 4 points
1. Attempt to re-express your target’s position so clearly, vividly and
fairly that your target says: “Thanks, I wish I’d thought of putting it that
way.”
2. List any points of agreement (especially if they are not matters of
general or widespread agreement).
3. Mention anything you have learned from your target.
4. Only then are you permitted to say so much as a word of rebuttal or
criticism.
37. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
List of popular narratives and counters
37
For elections, these include:
● Election date and location changes
● Scandals around politicians
● An election being won by a different candidate to the one declared winner.
Work with your communities to create this list
● You will likely have top-level narratives with subnarratives, and narratives
linked to different stages of an election
● List against prebunk/debunk
43. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
NCFG catechism
Situation:
● 1. situation or opportunity the narrative campaign is being conducted in response to or in preparation for?
● Mission:
● 2. desired end-state of the campaign?
Approach:
● 3. relevant target audience?
● 4. attitudes, behaviors, conclusions, and decisions within target audience to be influenced or generated?
● 5. sources and information channels considered most credible by the target audience? Why?
● 6. medium(s) and format(s) target audience most familiar with?
● 7. adversarial campaigns?
● 8. messages target audience already being exposed to?
● 9. methods used to influence target audience?
● 10. resources necessary to carry out this approach?
● 11. desired end-state of campaign? How is end-state measurable?
Milestones:
● 1. metrics used to indicate success of campaign?
● 2. milestones that indicate progress toward desired end-state?
43
49. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Text as “bags of words”
● Sentences:
○ “Oh really?”,
○ ”Just like with radiation poisoning then.”
● Words: “just”, “like”, “with”, etc
● Trigrams: “jus”, “ust”, “st “, “t l”, “ li”, “lik” “ike”
● Bigrams: “just like”, “like with”, “with radiation”, “radiation
poisoning”
● Stopwords: “with”, “on”, “then”, “what”, “have”, “they”,
“been”, “out”, “in”, “the”, etc
"RT @KateShemirani: Oh really? Just like
with radiation poisoning then. Put enough
symptoms down on the diagnosis sheet
and you can just abo… RT
@Walletwalking1: @Sterling2143
@AAureilus Anyone noticed COVID
symptoms are same as 5G exposure.
What have they been rolling out in the…
RT @ADDiane: Let's tell the people who
won't wear masks that it's not for covid,
it's for tricking the facial recognition
software that dee… Discourse "
50. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Word Importance: Named Entity Recognition
Finds names of people, organisations, locations etc in text
Can use to create social graphs
import spacy
nlp = spacy.load('en_core_web_sm')
sentence = "Bill Gates is selling 5G Covid19 data to Microsoft"
doc = nlp(sentence)
for ent in doc.ents:
print(ent.text, ent.label_)
Bill Gates PERSON
5 CARDINAL
Microsoft ORG
51. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Sentiment: (some of) the feels
● Word-based: give (some) words positive/negative scores
○ Use an existing ‘sentiment dictionary’
○ Score some words, use machine learning on the rest
● Document-based: score documents and use machine learning
○ ‘positive’/’negative’ for each sentence
● Semantic/pragmatic: use natural language processing
○ Satire is hard to detect
○ “Nice work bro!”
○ Emoticons are a language too
Sentiment dictionaries:
● Wordstat:
https://provalisresearch.com/products/content-analysis-software/wordstat-dictionary/sentiment-dictionaries/
● Sentiwordnet: https://github.com/aesuli/SentiWordNet
● Emoticon sentiment lexicon: http://people.few.eur.nl/hogenboom/files/EmoticonSentimentLexicon.zip
Very positive
Positive
Neutral
Negative
Very negative
58. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Json array to Pandas dataframe
Reading Twitter json can be a pain.
● I wrote library clean_twitter.py to convert these to a CSV file
● Or you can use “df = pd.DataFrame([tweetdata]).transpose().reset_index()”
(If you use Spacy, you might get a run error: fix (“python =m spacy download en” from the terminal
window) is at
https://stackoverflow.com/questions/54334304/spacy-cant-find-model-en-core-web-sm-on-windows-10-
and-python-3-5-3-anacon )
58
59. Disinformation/Malign
Influence
Training,
Disarm
Foundation
|
2022
Topic detection using Latent Dirichlet Analysis
no_features = 1000
no_topics = 7
tfidf_vectorizer = TfidfVectorizer(max_df=0.95,
min_df=2, max_features=no_features,
stop_words=stop_words)
tfidf =
tfidf_vectorizer.fit_transform(dftweets['text'])
tfidf_feature_names =
tfidf_vectorizer.get_feature_names()
59
lda =
LatentDirichletAllocation(n_components=no_topics
, max_iter=5, learning_method='online',
learning_offset=50.,random_state=0).fit(tfidf)
no_top_words = 10
for topic_idx, topic in enumerate(lda.components_):
print("Topic %d:{}".format(topic_idx))
print(" ".join([feature_names[i]
for i in topic.argsort()[:-no_top_words
- 1:-1]]))
from https://blog.mlreview.com/topic-modeling-with-scikit-learn-e80d33668730