Given a Dictionary with user_id and a list of alert_words (words and phrases too look for in an sentence) and a string content. We have to look if the alert_words appears in the content and return the list of user_ids who's alert_words appears in the content


input = { 1 : ['how are you'], 2 : ['you are'] }

content = 'hello, how are you'

output = [1]

user_id = 1 has 'how are you' while user_id = 2 has the words but not in the correct order so only user 1 is returned.


I'm using Google's pygtrie implementation of Trie data structure to achieve this. [pygtrie documentation]


  • For each word in the given sentence
  • Check if the word is a key, if yes add it to the list of user_ids
  • check if the word has a subtrie i.e. that current word is the starting of a alert_word. So add it to another set alert_phrases
  • for each word in alert_phrases we check if we can extent with the current word and do the same set of operations if it is a key/subtrie


import pygtrie
from typing import Dict, List, Set

def build_trie(realm_alert_words : Dict[int, List[int]]) -> pygtrie.StringTrie:
    trie = pygtrie.StringTrie()
    for user_id, words in realm_alert_words.items():
        for word in words:
            alert_word = trie._separator.join(word.split())        
            if trie.has_key(alert_word):
                user_ids_for_word = trie.get(alert_word)
                trie[alert_word] = set([user_id])
    return trie

def get_user_ids_with_alert_words(trie : pygtrie.StringTrie, content : str) -> Set[int]:
    """Returns the list of user_id's who have alert_words present in content"""
    content_words = content.split()
    alert_phrases = set()
    user_ids_in_messages = set()
    for possible_alert_word in content_words:
        #has_node returns 1(HAS_VALUE) if the exact key is found, 2(HAS_SUBTRIE) if the key is a sub trie, 
        # 3 if it's both 0 if it's none 

        alert_word_in_trie = trie.has_node(possible_alert_word)
        if alert_word_in_trie & pygtrie.Trie.HAS_VALUE:
            user_ids = trie.get(possible_alert_word)

        deep_copy_alert_phrases = set(alert_phrases)

        # Check if extending the phrases with the current word in content is a subtrie or key. And
        # Remove the word if it is not a subtrie as we are interested only in continuos words in the content
        for alert_phrase in deep_copy_alert_phrases:
            extended_alert_phrase = alert_phrase + trie._separator + possible_alert_word
            alert_phrase_in_trie = trie.has_node(extended_alert_phrase)

            if alert_phrase_in_trie & pygtrie.Trie.HAS_VALUE:
                user_ids = trie.get(extended_alert_phrase)

            if alert_phrase_in_trie & pygtrie.Trie.HAS_SUBTRIE:

        if alert_word_in_trie & pygtrie.Trie.HAS_SUBTRIE:

    return user_ids_in_messages


input = {1 : ['hello'], 7 : ['this possible'], 2 : ['hello'], 3 : ['hello'], 5 : ['how are you'], 6 : ['hey']}
alert_word_trie = build_trie(input)
content = 'hello how is this possible how are you doing today'
result  = get_user_ids_with_alert_words(alert_word_trie, content)
assert(result == set([1, 2, 3, 5, 7]))

input = {1 : ['provisioning', 'Prod deployment'], 2 : ['test', 'Prod'], 3 : ['prod'], 4 : ['deployment'] }
alert_word_trie = build_trie(input)
content = 'Hello, everyone. Prod deployment has been completed'
result  = get_user_ids_with_alert_words(alert_word_trie, content)
assert(result == set([1, 2, 4]))

input = {1 : ['provisioning/log.txt'] }
alert_word_trie = build_trie(input)
content = 'Hello, everyone. Errors logged at provisioning/log.txt '
result  = get_user_ids_with_alert_words(alert_word_trie, content)
assert(result == set([1]))

The two methods are part of a larger classes which have some not so related code. You get a list of user_ids and their alert_words from the database and you process every message content based on the trie already build up.

This is for a chat application so frequency of running the get_user_id_with_alert_words is high when the build_trie is relatively less since it will be cached.

  • \$\begingroup\$ If content is Hello, how [some words] are [more words] you, what the result should be? \$\endgroup\$ – vnp Jan 12 '19 at 22:16
  • \$\begingroup\$ @Carcigenicate you solution iterates the entire alert_words which grows in size as users increase. Also i don't think the in a list is of linear time complexity. \$\endgroup\$ – thebenman Jan 13 '19 at 5:23
  • 1
    \$\begingroup\$ @thebenman: in is O(n) for lists (and strings): wiki.python.org/moin/TimeComplexity \$\endgroup\$ – Graipher Jan 13 '19 at 9:13
  • \$\begingroup\$ Since the time complexity is important to you, what do you think is the time complexity of using Tries? \$\endgroup\$ – Graipher Jan 13 '19 at 9:14
  • 1
    \$\begingroup\$ @Graipher Finding a word in a trie is O(m) where m is the length of the word and finding it in a list would be O(n) where n is the total number of elements in the list which can grow as opposed to the number of words in the message. \$\endgroup\$ – thebenman Jan 13 '19 at 11:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.