Skip to content

Latest commit

 

History

History
91 lines (62 loc) · 5.27 KB

README.md

File metadata and controls

91 lines (62 loc) · 5.27 KB
Raw

Cryptocurrency Data Platform & AI Trading Algorithm

CI

This repository contains the source code, config files, and a short demo of the Cryptocurrency Data Platform & AI Trading Algorithm.

Key Objectives

The goal of the service is to provide a data platform that returns the most up-to-date information about the cryptocurrency market, and provide an AI Trading model that predicts the future sentiment of a currency to determine an optimal trading strategy.

The data regarding coin trends and prices was scraped using Telegram, Twitter and Coingecko APIs and the AI Trading model is using the coin price and sentiment information to determine whether to invest in a coin.

Architecture and Program Description

architecture_diagram drawio (3)

Telegram Bot

The Telegram Bot will by default scrape all messages from the Airdrop Alert Channel, which contains information about most recent coin releases and trends in the cryptocurrency market. The Telegram Bot will identify the key words (coins) from the channel messages and send them over to the Twitter Bot.

Please see the below for more details about how to set up API connection:

    * Add the following to your GitHub Codespace or Local file
    * API_ID, API_HASH, PHONE, USERNAME
    * api_id = os.getenv('API_ID')
    * api_hash = os.getenv('API_HASH')
    * phone = os.getenv('PHONE')
    * username = os.getenv('USERNAME')

Twitter Bot

The Twitter Bot accepts the key words generated from Telegram, and search for tweets related to the key words. The bot will return a dataframe containing all the tweets scraped for a specified timeframe. The bot uses the tweepy package in Python to connect to the Twitter API. See below for more details:

import tweepy

For more Twitter API registration details, read from website here: https://scottlai.com/2022/09/24/twitter-bot-101/

  • The API setting: tweepy.Client.
  • searching for recent tweets: client.search_recent_tweets.
  • Save the Tweets into CSV format.

Input:

searchTweets('key_words', max_tweets_amount(less than 100 for each time))

output: dataframe

Coin Price Query Bot

The Coin price query bot will return a payload that contains the most up-to-date cryptocurrency prices for a specified timeframe. The bot uses the Coingecko API to scrape the data. See below for more details:

@Aditya to add more information.

Sentiment Analyzer

The sentiment analysis component will predict the sentiment of a given tweet and generate a sentiment score between -1 and 1 for a tweet.

The sentiment analyzer is a roBERTa-base model that was trained on ~124M Tweets. More information about the model used can be found here.

Please see below for an example use case / pipeline for the sentiment analysis:

from sentiment_analyzer import sentiment_generator

df = sentiment_generator(df, calculate_scores=False, task="sentiment-latest", remove_stopwords=False)

The sentiment_generator function will accept a pandas dataframe df and output another pandas dataframe df with one or more columns reporting information about the sentiment of the tweet (see below for more details). All tweets are stored under the field text. The function will also take additional three parameters: 1) calculate_scores will generate a sentiment score in addition to a sentiment label ('Positive', 'Negative', 'Neutral') if set to True. 2) 'remove_stopwords' will remove stopwords from the tweets if set to 'True'. 3) task: the Roberta-Base sentiment model can perform other tasks as well including emotion, degree of offensiveness of text. These are the possible parameters a user can input for task: emoji, emotion, hate, irony, offensive, sentiment-latest . NOTE: besides for 'sentiment-latest', all other tasks have not been tested with the model.

Data Cleaning steps:

  • lowercase all tweets
  • remove any URLs.
  • an option to remove all stopwords. If user sets remove_stopwords=True in the sentiment_generator function then stopwords will be removed. By default, stopwords are preserved.
  • Handles English tweets only.
  • replace mentions with a flag @User
  • removes hashtag signs

Algorithm and Automation

The AI Algorithm will take in the scraped twitter data and apply sentiment analysis for each tweet. Along with the coin price information, it will then generate a trading strategy for the user based on the signals that the model generates. We are using a time-series model trained on the twitter and sentiment analysis. Please see below for more details about the architecture of the model: framework

API

Please perform the following steps to access the Swagger Documentation of the data platform:

  1. run python ./api/main.py and open the application in a browser
  2. type in /docs after the root url to open up the Swagger Documentation

The user should be able to access Telegram, Twitter and Coin price data after running the API.

Demo Video

@All - add video