Skip to content

Latest commit

 

History

History
174 lines (146 loc) · 9.33 KB

README.md

File metadata and controls

174 lines (146 loc) · 9.33 KB
Raw

Research on all kinds of NLP for market forecasting, experts estimation, etc.

%%{init: {"theme": "default","fontFamily": "cursive", "sequence": { "securityLevel":"loose" }}}%%
graph TD
    title[<u>Table of Contents:</u>]
    title---F
    style title fill:#FFF,stroke:#FFF
    linkStyle 0 stroke:#FFF,stroke-width:0;

    %%FirstStep-->...
    %%classDef containers fill:#E6E1F7
    
    %% styling the nodes and arrors https://dompl.medium.com/produce-great-looking-flowcharts-in-seconds-7f3bea64f2e2
    %% mermaid editor https://mermaid.live
    Mo[[<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#models'>Models</a>]]
    A(<b>NLP</b>) -->|for| F{<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#1-market-forecasting'>Market Forecasting</a>} 
    
    style A stroke:#333,stroke-width:4px
    subgraph MF [<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#sentiment-based' >Sentiment based</a>]
    Se{{<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#sentiment-analysis'>Sentiment<br>Analysis</a>}}
    Fc{{<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#forecasting'>Forecasting</a>}}
    %% Al{{?}}
    end

    
    Ts[(<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#text-sources'>Text Sources</a>)]
    
    %% Se<-.-Mo
    %% Fc<-.-Al[[<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#algorithms'>Algorithms</a>]]
    F--> Di{{Direct}}

    Sc(((<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#scrapping'>Scrapping</a>)))
    A -->|to identify| M((<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#2-market-movers-identification'>Market Movers</a>))
    A -->|to estimate| E[<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#3-experts-competence-estimation'>Expert's Competence</a>]
    F -->|"  "| MF{{?}}
    style Se stroke:#f66,stroke-width:2px,stroke-dasharray: 5 5,text-decoration: none
   
    
    %%I><a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#interpretability'>Interpretability?</a>]
    %%V<-.-I
    %%Q<-.-I
    %%Di<-.-I
    %%MF<-.-I
    %% class A,B,F,E,E,M,F,Q containers;
Loading

Models

  • FinancialBERT - A Pretrained Language Model for Financial Text Mining 2022 Ahmed Rachid Hazourli [pdf] [demo]
  • FinBert - A Large Language Model for Extracting Information from Financial Text 2019-2022 Allen H. Huanga, Hui Wang, Yi Yang [pdf] [repo] [website] [demo]
  • NFinBERT: A Number-Aware Language Model for Financial Disclosures 2021 Hao-Lun Lin and Jr-Shian Wu, Yu-Shiang Huang, Ming-Feng Tsai, Chuan-Ju Wang [pdf]

1. Market Forecasting

Projection of the market states (high/low volatility, up/down index movements).

Sentiment Based

Text source: newspapers
Sentiment model: BERT
Predictions model: LSTM?

Claimed predictions for Tesla based solely on newspaper text

Sentiment Analysis

Models
Hugging Face Hub #financial-sentiment-analysis

Forecasting

Algorithms

To be added

Interpretability

Essentially we want an answer to a question "what..", but can we as well know "why"?
Is this problem related to long-form question-answering (LFQA)? I don't know. If so, here's something to read:

  • WebGPT: Browser-assisted question-answering with human feedback 2021 OpenAI [blog] [pdf] [demo] + the explanation on habr (by @stalkermustang I presume)
  • GopherCite: Teaching language models to support answers with verified quotes 2022 DeepMind [blog] [pdf]
  • (And obviously) Toolformer: Language Models Can Teach Themselves to Use Tools 2023 Meta [pdf]

Can the task of prediction can be solved as collateral to the Market Movers identification?

2. Market Movers Identification

Market moving information is a term used in stock market investing, defined as information that would cause any reasonable investor to make a buy or sell decision. (Wikipedia)

Example of a stock mover identification taken from Benzinga:

How can such pipeline look:

%%{init: {"theme": "default","fontFamily": "cursive", "sequence": { "securityLevel":"loose" }}}%%
stateDiagram

    classDef Important stroke:#f66,stroke-width:2px,stroke-dasharray: 5 5,text-decoration: none
    direction LR
    A: Get top N gainers/losers
    b: Get latest tweets/news
    c: Summarize as an explanation
    B: ~
    C: Show
    [*] --> A
    A --> B
    B --> C
    state B {
      direction LR
      b --> c
    }
    C --> [*]
    class c Important
Loading

3. Expert's Competence Estimation

Ranking experts, qualification of experts, based on their past predictions/performance.

Tools

Pythonic ways to download market data from Yahoo!Ⓡ finance:
yfinance
yahooquery

Text Sources

%%{init: {"theme": "default","fontFamily": "cursive", "sequence": { "securityLevel":"loose" }}}%%
graph TD
    classDef containers fill:#E6E1F7
    Sc(((<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#scrapping'>Scrapping</a>)))
    Ts[(<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#text-sources'>Text Sources</a>)]
    subgraph Ws [<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#websites'>Websites</a>]
        subgraph Tw [<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#websites'>Twitter</a>]
            subgraph Fb [<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#websites'>Facebook</a>]
                subgraph Tg [<a href='https://github.com/completelyboofyblitzed/NLP-in-STOCK-MARKET/blob/main/README.md#websites'>Telegram</a>]
                Cm[(Plain mentions/tags)]
                Ex[("''Experts''")]
                In[(Highly influential<br> individuals)]
                Of[(Official Media)]
                end
            end
        end
    end
    Ts-->Cm
    Ts-->Ex
    Ts-->In
    Ts-->Of
    Sc-.-Fb
    Sc-.-Tw
    Sc-.-Ws
    Sc-.-Tg
Loading

Scrapping

To be added

Websites

To be added

Twitter

To be added

Facebook

To be added

Telegram

To be added

🤝 Contributing

Contributions, issues and feature requests are welcome.