问题

I want to extract hashtags for my sentiment analysis project, however I'm getting a list of dictionary containing all the hashtags along with their indices in the tweet. I only want the text.

My code :

data = tweepy.Cursor(api.search, q, since=a[i], until=b[i]).items()
    tweet_data = []
    tweets = pd.DataFrame()
    tweets['Tweet_ID'] = map(lambda tweet: tweet['id'], tweet_data)
    tweets['Tweet'] = map(lambda tweet: tweet['text'].encode('utf-8'), tweet_data)
    tweets['Date'] = map(lambda tweet: time.strftime('%Y-%m-%d %H:%M:%S', time.strptime(tweet['created_at'],'%a %b %d %H:%M:%S +0000 %Y')), tweet_data)
    tweets['User'] = map(lambda tweet: tweet['user']['screen_name'], tweet_data)
    tweets['Follower_count'] = map(lambda tweet: tweet['user']['followers_count'], tweet_data)
    tweets['Hashtags']=map(lambda tweet: tweet['entities']['hashtags'], tweet_data)

Current Output :

df=pd.DataFrame({'Hashtags' : [{u'indices': [53, 65], u'text': u'Predictions'}, {u'indices': [67, 76], u'text': u'FreeTips'}, {u'indices': [78, 89], u'text': u'SoccerTips'}, {u'indices': [90, 103], u'text': u'FootballTips'}, {u'indices': [104, 110], u'text': u'Goals'}]})

Expected Output :

df=pd.DataFrame({'Hashtags' :["u'Predictions'", "u'SoccerTips'", "u'FootballTips'", "u'Goals'"]})

I've tried to use several methods to flatten/reduce/access a nested dictionary containing list of dictionaries. Please help.

Error :

as @MSeifert suggested, I've tried his method. The following error was generated:

dt=tweet.entities.hashtags
pd.io.json.json_normalize(dt, 'hashtags')
pd.io.json.json_normalize(dt, 'hashtags')['text'].tolist()

Traceback (most recent call last): <\br>

File "<ipython-input-166-be11241611d6>", line 1, in <module>
dt=tweet.entities.hashtags

AttributeError: 'dict' object has no attribute 'entities'

I've also tried doing this :-

dx = tweets['Hashtags']
for key, value in dx.items():
    print key, value

With the following error :-

Traceback (most recent call last):

File "<ipython-input-167-d66c278ec072>", line 2, in <module>
    for key, value in dx.items():

File "C:\ANACONDA\lib\site-packages\pandas\core\generic.py", line 2740, in __getattr__
    return object.__getattribute__(self, name)

AttributeError: 'Series' object has no attribute 'items'

UPDATE :

I'm able to access the text part of the nested hashtags dictionary

tweets['Hashtags'][1][1]['text']
Out[209]: u'INDvPAK'

I want to create a loop to append all the hashtags in the row.

回答1:

Instead of using the DataFrame constructor you could use the json_normalize function:

>>> import pandas as pd
>>> d = {'Hashtags' : 
...      [{u'indices': [53, 65], u'text': u'Predictions'}, 
...       {u'indices': [67, 76], u'text': u'FreeTips'}, 
...       {u'indices': [78, 89], u'text': u'SoccerTips'}, 
...       {u'indices': [90, 103], u'text': u'FootballTips'}, 
...       {u'indices': [104, 110], u'text': u'Goals'}]}
>>>  pd.io.json.json_normalize(d, 'Hashtags')
      indices          text
0    [53, 65]   Predictions
1    [67, 76]      FreeTips
2    [78, 89]    SoccerTips
3   [90, 103]  FootballTips
4  [104, 110]         Goals

Then you could just use the 'text' column:

>>> pd.io.json.json_normalize(d, 'Hashtags')['text'].tolist()
[u'Predictions', u'FreeTips', u'SoccerTips', u'FootballTips', u'Goals']

回答2:

Here's the solution :

After troubleshooting and trying various methods for a lot of time, I finally figured out how to split the nested dictionary. It is a fairly simple loop. I noticed that we can access the hashtag text by

tweets['Hashtags'][1][1]['text']
Out[209]: u'INDvPAK'

This was a valuable insight as i got to know I DON'T need to mention u'text as my index. text will be used.

Code :

ht=[]
for s in range(len(tweets['Hashtags'])):
    hasht=[]
    for t in range(len(tweets.Hashtags[s])):
        #zx = tweets['Hashtags'][s][t]['text']
        hasht.append(tweets['Hashtags'][s][t]['text'])
        t=t+1
    ht.append(hasht)
    s=s+1
tweets['HT']=zip(ht)

This is a simple nested for loop which iterates through first the inner key values in the { "Indices" : [], "u'text'" : []} and then iterates through the list of dictionaries under ["entities" : { "Hashtags" : [{1},{2},{3}]}]

Finally I used zip() to zip the lists of hashtags for a single row/user.