个人博客地址在 https://mengjiexu.com/post/door-to-twitter-analysis-ii/

Introduction

Please see the background information about Twitter API v2 in my last blogpost Twitter API: Door To Social Media Analysis I. In this blogpost, I will replicate the Tweets extraction porcess of Bartov, Faurel, and Mohanram (2018, TAR), which found that opinions of individuals tweeted just prior to a firm’s earnings announcement can predict its earnings and announcement returns.

From a perspective of implementation, we need to get all the tweets which mentioned a given firm in a specified time range. Consistent to the paper’s definition, the targeted tweets should:

• Include stock symbols preceded by the dollar sign (e.g., $AAPL for Apple Inc.; $PEP for PepsiCo Inc.)
  • In which the stock symbols belong to Russell 3000 firms
• Be released between March 21, 2006 and December 31, 2012

RoadMap

1. Import access token from the environment variables and put the token into the headers. Regarding how to get access token and make it an environment variable, please check Get Access to Twitter API.

   bearer_token = os.environ.get("BEARER_TOKEN")
   def create_headers(bearer_token):headers = {"Authorization": "Bearer {}".format(bearer_token)}return headers
   

2. Set parameters for queries. Regarding the available fields and their usages for Tweet object, please check Fields of Tweet Object.

   By specifying start_time, end_time, and query, I require the returned tweets:

   • Is released no earlier than start_time and no later than end_time
   • Contain the ticker of a firm firm preceded by the dollar sign

   By specifying tweet.fields, I expect the following information for each tweet is returned:

   • created_at, public_metrics, author_id, lang

   By specifying max_results, one can get 10 tweets for each request. This figure could be changed.

   # Input filter parameters: firm tickers, start time stamp, end time stamp
   # Optional params: start_time,end_time,since_id,until_id,max_results,next_token,
   # expansions,tweet.fields,media.fields,poll.fields,place.fields,user.fields
   def setpara(firm,start_time,end_time):query_params = {'start_time': "%s"%start_time, 'end_time':"%s"%end_time, 'query': "$%s"%firm, 'tweet.fields':"created_at,public_metrics,author_id,lang",'max_results': 10}return(query_params)
   

3. Submit a query and check the returned data

   For now, we have obtained the necessary elements for submitting a query:

   • Access token bearer_token
   • Query parameters start_time, end_time, and query
   • Returned fields created_at, public_metrics, author_id, lang
   • Parameter for page splitting max_results

   With those elements, we can submit our very first query and analyse the returned data to decide on how to clean it.

   import requests, os, json# To set your environment variables in your terminal run the following line:
   # export 'BEARER_TOKEN'='<your_bearer_token>'
   bearer_token = os.environ.get("BEARER_TOKEN")search_url = "https://api.twitter.com/2/tweets/search/all"# Optional params: start_time,end_time,since_id,until_id,max_results,next_token,
   # expansions,tweet.fields,media.fields,poll.fields,place.fields,user.fields
   def setpara(firm,start_time,end_time):query_params = {'start_time': "%s"%start_time, 'end_time':"%s"%end_time, 'query': "$%s"%firm, 'tweet.fields':"created_at,public_metrics,author_id,lang",'max_results': 10}return(query_params)def create_headers(bearer_token):headers = {"Authorization": "Bearer {}".format(bearer_token)}return headersdef connect_to_endpoint(url, firm, start_time, end_time):response = requests.request("GET", search_url, headers = create_headers(bearer_token), params=setpara(firm,start_time,end_time))print(response.status_code)if response.status_code != 200:raise Exception(response.status_code, response.text)return response.json()def search(firm, start_time, end_time):json_response = connect_to_endpoint(search_url, firm, start_time, end_time)print(json.dumps(json_response, indent=4, sort_keys=True))if __name__ == "__main__":search("AAPL", '2012-03-21T00:00:00Z', '2012-03-21T00:20:00Z')
   

   If the code is executed successfully, we will get response displayed as follows (To save the space, I removed the 2nd to 9th tweets returned).

   200
   {"data": [{"author_id": "215869072","created_at": "2012-03-21T00:18:35.000Z","id": "182259875408121856","lang": "en","public_metrics": {"like_count": 0,"quote_count": 0,"reply_count": 0,"retweet_count": 1},"text": "@FortuneMagazine Can #Apple's stock reach $1,368? http://t.co/wJlWogk1 $AAPL  @albertosrt"},...{"author_id": "341817759","created_at": "2012-03-21T00:06:37.000Z","id": "182256865110003712","lang": "en","public_metrics": {"like_count": 0,"quote_count": 0,"reply_count": 0,"retweet_count": 0},"text": "was forced to buy a small position in $aapl today to offset shorts in other names, in at 594 with $3 trailing"}],"meta": {"newest_id": "182259875408121856","next_token": "1jzu9lk96gu5npvzel7a7nb3ogobtgzn9360fuunqj5p","oldest_id": "182256865110003712","result_count": 10}
   }
   

4. Update query paramters with next_token to get the returned data of the next page

   Clearly, the tweets are returned in a json format. Different from the fields we have requested (created_at, public_metrics, author_id, lang), the meta part is the key for jumping to the next page. If the meta has the element next_token, that means there are more pages contained in your submitted query and we need to add the value of next_token into the query parameters to get the returned data in the next page. Apparently, we need to repeat this procedure until the value of next_token is empty.

   # Add the 'next_token' returned in the search result into the parameters for the subsequent query for the next page
   def updatepara(oldpara, next_token):newpara = {**oldpara,**{'next_token':next_token}}return(newpara)# Request the content of next page with updated parameters
   def dealjson(firm, start_time, end_time, json_response):headers = create_headers(bearer_token)# Retrieve the 'next_token' from the returned jsonnext_token = csvjson(firm, start_time, end_time, json_response)while next_token:# If the 'next_token' is not empty, add it to the search parameters and 		submit a new querynewpara = updatepara(setpara(firm,start_time,end_time), next_token)time.sleep(3)json_response = connect_to_endpoint(search_url, headers, newpara)next_token = csvjson(firm, start_time, end_time, json_response)
   

5. Record the request history and transfer the received data to csv format.

   # Clean data received and export in csv format
   def csvjson(firm, start_time, end_time, json_response):# Record meta data of request historymeta = json_response['meta']callnum = meta['result_count']try:next_token = meta['next_token']except:next_token = Falseempty = (callnum==0)with open("/xx/meta.csv",'a') as g:now = time.strftime('%Y-%m-%d %H:%M:%S',time.localtime(time.time()))g.write(",".join([firm, start_time, end_time,now, repr(callnum), str(next_token)])+'\n')# Clean received json-formatted tweets and export in csv formatif not empty:with open("/xx/%s.csv"%firm,'a') as f:w = csv.writer(f)tweets = json_response['data']for tw in tweets:list = [tw['created_at'], tw['id'], tw['text'],tw['public_metrics']['retweet_count'],tw['public_metrics']['reply_count'],tw['public_metrics']['like_count'],tw['public_metrics']['retweet_count'],tw['author_id'], tw['lang']]w.writerow(list)return(next_token)
   

6. Iterate the above steps through the pre-specified ticker list.

   # Input the ticker list
   tickerlist = open("/xx/code/tickerlist.txt",'r')def mainparse(firm,start_time,end_time):headers = create_headers(bearer_token)json_response = connect_to_endpoint(search_url, headers, setpara(firm,start_time,end_time))dealjson(firm, start_time, end_time, json_response) if __name__ == "__main__":for ticker in tqdm(tickerlist):mainparse('$%s'%ticker.strip(),'2006-03-21T00:00:00Z','2012-12-31T23:59:59Z')time.sleep(1)
   

Sample Results

All th extracted tweets for a given firm are saved in a same csv file named by the firm’s ticker.

Figure 1: Sample File List

Recall a typical returned data for each tweet is as follows. suppose the root json node for a given tweet is tw.

{	"author_id": "215869072","created_at": "2012-03-21T00:18:35.000Z","id": "182259875408121856","lang": "en","public_metrics": {"like_count": 0,"quote_count": 0,"reply_count": 0,"retweet_count": 1
}

For each firm, there are following columns:

• Release time stamp tw['created_at']
• Tweet ID tw['id']
• Tweet text tw['text']
• retweet_count tw['public_metrics']['retweet_count']
• reply_count tw['public_metrics']['reply_count']
• like_count tw['public_metrics']['like_count']
• Author ID tw['author_id']
• Language of Tweet tw['lang']

Figure 2: Sample Tweets List

Codes

import requests, time
import os, json, csv
from tqdm import tqdm# To set your environment variables in your terminal run the following line:
# export 'BEARER_TOKEN'='<your_bearer_token>'
# If you don't successfully set the token as an enviornment variable, please use the following line instead:
# bearer_token = '<Your Bearer Token>'
bearer_token = os.environ.get("BEARER_TOKEN")# Request url
search_url = "https://api.twitter.com/2/tweets/search/all"# Input the ticker list
tickerlist = open("/xx/code/tickerlist.txt",'r')# Verify your developer access to Twitter 
def create_headers(bearer_token):headers = {"Authorization": "Bearer {}".format(bearer_token)}return headers# Input filter parameters: firm tickers, start time stamp, end time stamp
# Optional params: start_time,end_time,since_id,until_id,max_results,next_token,
# expansions,tweet.fields,media.fields,poll.fields,place.fields,user.fields
def setpara(firm,start_time,end_time):query_params = {'start_time': "%s"%start_time, 'end_time':"%s"%end_time,'query': "%s"%firm, 'tweet.fields': "created_at,public_metrics,author_id,geo,lang,non_public_metrics,promoted_metrics",'max_results': 10}return(query_params)# Get response for the server and return in json format
def connect_to_endpoint(url, headers, params):response = requests.request("GET", search_url, headers=headers,params=params)if response.status_code != 200:raise Exception(response.status_code, response.text)return response.json()# Add the 'next_token' returned in the search result into the parameters for the subsequent query for the next page
def updatepara(oldpara, next_token):newpara = {**oldpara,**{'next_token':next_token}}return(newpara)# Clean data received and export in csv format
def csvjson(firm, start_time, end_time, json_response):# Record meta data of request historymeta = json_response['meta']callnum = meta['result_count']try:next_token = meta['next_token']except:next_token = Falseempty = (callnum==0)with open("/xx/meta.csv",'a') as g:now = time.strftime('%Y-%m-%d %H:%M:%S',time.localtime(time.time()))g.write(",".join([firm, start_time, end_time,now, repr(callnum), str(next_token)])+'\n')# Clean received json-formatted tweets and export in csv formatif not empty:with open("/xx/%s.csv"%firm,'a') as f:w = csv.writer(f)tweets = json_response['data']for tw in tweets:list = [tw['created_at'], tw['id'], tw['text'],tw['public_metrics']['retweet_count'],tw['public_metrics']['reply_count'],tw['public_metrics']['like_count'],tw['public_metrics']['retweet_count'],tw['author_id'], tw['lang']]w.writerow(list)return(next_token)# Request the content of next page with updated parameters
def dealjson(firm, start_time, end_time, json_response):headers = create_headers(bearer_token)# Retrieve the 'next_token' from the returned jsonnext_token = csvjson(firm, start_time, end_time, json_response)while next_token:# If the 'next_token' is not empty, add it to the search parameters and submit a new querynewpara = updatepara(setpara(firm,start_time,end_time), next_token)time.sleep(3)json_response = connect_to_endpoint(search_url, headers, newpara)next_token = csvjson(firm, start_time, end_time, json_response)print(next_token)def mainparse(firm,start_time,end_time):headers = create_headers(bearer_token)json_response = connect_to_endpoint(search_url, headers, setpara(firm,start_time,end_time))dealjson(firm, start_time, end_time, json_response) if __name__ == "__main__":for ticker in tqdm(tickerlist):mainparse('$%s'%ticker.strip(),'2006-03-21T00:00:00Z','2012-12-31T23:59:59Z')time.sleep(1)

Summary and Upcoming Posts

In this blogpost, I replicated the Tweets extraction process of Bartov, Faurel, and Mohanram (2018, TAR). But retreiving the tweets archive is just a very small part of the Twitter API’s functions. In the next blogpost, I will introduce how to request the detailed user account information as well as how to check the bi-direction links among multiple Twitter accounts.

References

https://github.com/twitterdev/Twitter-API-v2-sample-code