Goal: Given a real time stream of social media posts, proper news posts, and journalistic videos, we want to.
- Recognize entities and track sentiment towards them over time by demographic. E.g. how do russian men feel about Xi Jinping these days?
- Define what an "event" or "narrative" is, identify which pieces of media seem to propagate this narrative, and have information about the narrative such as time sequence, most likely place of origin, etc. E.g., fake example, in July 2019 there was a strong push against universal basic income that started dominating twitter for americans, and interestingly, in June 2019 there were a lot of Chinese social media accounts and news about the very same issue, and then some maybe chinese english tweeters who started the discussion up over again.
- Broadly categorize by province/nation, potentially by deeper demographics if possible, general sentiment towards these events and entities and highlight the difference between nations over time. E.g., Maybe in june, russians loved modi, but in august, they hate modi, but Chinese hated modi in june and in august.
- Collect and perform some type of metadata analysis to try and detect "influence operations." It is not a question of whether or not states are interfering in online discourse, the incentives are too strong for them not to, the question is how to detect them. My hypothesis is that by knowing what narratives are popular in a country, and by having an analysis of that countries national news, and their entity opinions, we may be able to use those heuristics to potentially flag "suspicious" accounts, even if they're not
- This is the big ticket thing that they really want. Lets say that you want to change the narrative, could you make a list of potential actions that can be taken? E.g., if the russian government is pushing some narrative about israel-palestine particularly hard, what are strong arguments that can be taken against it? (recommended by the primer guy)
Generally narratives don't exist in a vacuum. They exist and there's also their opposition. Perhaps if we can flag what we think are russian bots pushing something like, israel is really bad, then we could identify what the most positively engaged with pro-israel arguments in response to them are, and suggest that to them? The guy said we didn't need to give them it verbatim, we could just give it in the form of a "cheatsheet" or something
Even past that, lets say you have to give a press release, your job is to figure out what to say in advance, it would be beneficial to know what would be the best received things in that regard.
Something the guy said would be useful was in the formatting and display of the thing, like an autogenerated love/hate list, like according to state sponsored chinese media, whos on the love list and hate list in terms of entitties.
Comments -
Generally, a common trend that i'm realizing is that while the data may be there somewhere, it certainly isnt universally available or easily accessible to the individual intelligence officer, and the channels from which they have to do it often times mean they're doing a lot of their own research on these types of things. The guy literally said yeah they just use google. So simply collecting it and putting it on a dashboard is nice, but we need to show some type of technological advance, and I', thinking that the technological advance of detecting influence operations is best.
Presentation looks like, shill advances in natural language processing, say some strangely mathy things,
H4D problem description - When assessing success and strategic messaging within information operations, the evaluation of how widely information is disseminated stands as a crucial factor. Social media platforms serve as key tools in gauging the authenticity and effectiveness of communicated messages. However, due to the vast expanse of social media across various languages, understanding sentiment becomes exceedingly complex. Analyzing diverse social media platforms in different languages further complicates measurement, making analysis subjective. ● For instance, in humanitarian missions supporting oppressed groups, broadcasting messages becomes crucial in countering suppressive narratives. Assessing sentiment changes towards America, measuring from, say, a 30% to 50% positivity rate or demonstrating positive shifts after disseminating specific messages, holds significant importance. ● An all-in-one capability to scrape social media sites is desired, given the current anecdotal and sporadic observation-based methods, especially on platforms like Twitter. The existing process of sentiment analysis remains highly manual and lacks comprehensive coverage. This approach is essential not only for strategic campaigns but also for understanding how the United States is perceived by other nations.
Key takeaways:
- It's actually manual?
- They do care about tracking the information and efficacy of their productivity. Sydney said before; what about sales? it's harder when it's not a company trying to literally track a product.