💭Sentiment Analysis

Individual Telegram messages, Twitter posts, and replies/comments are used to calculate sentiment against a list of established narratives (DeFi, Meme, AI, etc) from market aggregators such as Coingecko, CMC, et al. as well as curated sub-narratives, specific divisions such as "AI-big data", AI-"model production", or "AI-companion models."

Our large language model (LLM) is tasked with filtering and assigning each contextual message (within a larger message set, or conversation) with a sentiment score, ranging from 1 to 10, representing low and high sentiment, respectively.

Tickers associated with this message+context can be implied or directly inferred, such as when mentioned within a specific message, or associated with a continued conversation or thread.

StrawberryAI then builds historical data for each ticker, picking up on and adding to each ticker's and tertiary narrative, giving it the ability to more confidently sort new projects into nuanced sub-narratives, allowing the model to build upon itself recursively.

Putting it Together with DYR, and Other Products

Once $TICKER and narrative are confidently paired, the results are then candidates for broadcasting overall sentiment when reaching outside of the normal range (ie, "Sentiment for $TICKER has increased significantly over the past four hours). An example template below, is how information is aggregated and arranged for dissemination:

• Fastest growing narratives

• Fastest growing tickers

[Summarization of changes within Twitter] [Average sentiment shift, tickers]

[Summarization of changes within Telegram] [Average sentiment shift, tickers]

[Summarization of changes within trusted News feeds] [Average sentiment shift, tickers]

[Aggregated median sentiment]

[Description of tickers detected, including price movements and holders +/- ]

[Rank detected tickers, by putting bullish cases first in the list ]

• Explain rationale about potential upside/downside