Stock_Chatbot

This is a Chatbot project by Kenneth Sun to help clients with stock information.

Brief video (2mins)

For Mainland China users, please see https://www.bilibili.com/video/av43594521/

Project report

For further information about what this project does, please see Chatbot_Project_Report.pdf

Note: If the online viewer is not available, you can DOWNLOAD the report in the link above.

Introduction

This is a chatbot to help clients with stock information. With this chatbot, clients can query various stock indicators conveniently. And he can also give brief investment suggestions.
The chatbot is associated with Wechat app via wxpy API. Model training is based on Rasa-nlu. The following techniques or methods are implemented:

Multiple selective answers to the same question and provide a default answer.
Intent recognition based on sklearn and spacy.
Named entity recognition using conditional random fields.
Construction of a local chatbot system based on Rasa-NLU.
Single-round incremental query for multiple times based on the incremental filter.
Multiple rounds of multi-query technology on state machines, and can provide explanations and answers based on contextual issues.
Handling pending state transitions and pending actions.
Complex pandas Dataframe processing and data cleaning, and producing a corresponding matplotlib figure.

Example

Identification sample

Input: "I want to know the highest price of TSLA in the past few days"

# Output:
{'intent': {'name': 'vague_historical_data', 'confidence': 0.5648109407370152},
 'entities': [{'start': 19,
   'end': 26,
   'value': 'high',
   'entity': 'hst_data_type',
   'confidence': 0.7486755036905514,
   'extractor': 'ner_crf',
   'processors': ['ner_synonyms']},
  {'start': 36,
   'end': 40,
   'value': 'tsla',
   'entity': 'company',
   'confidence': 0.8320076082894126,
   'extractor': 'ner_crf'}],
 'intent_ranking': [{'name': 'vague_historical_data',
   'confidence': 0.5648109407370152},
  {'name': 'current_price', 'confidence': 0.09569968257227997},
  {'name': 'finish', 'confidence': 0.08864097295750097},
  {'name': 'advice', 'confidence': 0.06723593774902344},
 'text': 'i want to know the highest price of TSLA in the past few days'}

Environment

Python 3.4-3.6.
Installed iexfinance, wxpy, pandas, matplotlib, spacy.
Installed Rasa-nlu. (https://www.rasa.com/docs/nlu/installation/)

Usage

Train the model

You can either use the given model

trainer = Trainer(config.load("config_spacy.yml"))
training_data = load_data('stock_training.json')
interpreter = trainer.train(training_data)

Or train a customized model by yourself

# Build a training file
customized_training = {
  rasa_nlu_data = {
    # Your training example here
  }
}
# Write the data into json file
with open("stock_training.json","w") as f:
    json.dump(stock_training,f)
    print('Done')
# Train the model
trainer = Trainer(config.load("config_spacy.yml"))
training_data = load_data('stock_training.json')
interpreter = trainer.train(training_data)

Extract the intent and entities

interpreter.parse('I want to get the historical close price of tesla from Oct 12 2017 to Jan 22 2018')

Generate line chart of stocks

message = 'Tell me the historical close price of TSLA from 2017-5-8 to 2017-6-8.'

generate_figure(message)

Build your Wechat bot

from wxpy import *

# Create a new Bot object
bot = Bot()

# Set target client account
my_friend = bot.friends().search('YOUR PARTNER HERE')[0]

# Register
@bot.register(my_friend, TEXT)
def auto_reply(msg):
    # Your chatbot action here

Contact

Email: yzjshz1998@outlook.com
Personal website: www.kennethsun.com (currently under maintenance)

HaozhanSun/Stock_Chatbot