TakeBlipInsightExtractor Package

Data & Analytics Research

Overview

Here is presented these content:

• Intro
• Run
• [Example of initialization e usage](#Example of initialization e usage)

Intro

The Insight Extractor offers a way to analyze huge volumes of textual data in order to identify, cluster and detail subjects. This project achieves this results by way of applying a proprietary Named Entity Recognition (NER) algorithm followed by a clustering algorithm. The IE Cloud also allows any person to use this tool without having too many computational resources available to themselves.

The package outputs four types of files:

• Wordcloud: It's an image file containing a wordcloud describing the most frequent subjects on the text. The colours represent the groups of similar subjects.

• Wordtree: It's an html file which contains the graphic relationship between the subjects and the examples of uses in sentences. It's an interactive graphic where the user can navigate along the tree.

• Hierarchy: It's a json file which contains the hierarchical relationship between subjects.

• Table: It's a csv file containing the following columns:

    Message                   |  Entities                                                                                    | Groups     | Structured Message
    sobre cobranca inexistente|[{'value': 'cobrança', 'lowercase_value': 'cobrança', 'postags': 'SUBS', 'type': 'financial'}]|['cobrança']|sobre cobrança inexistente
  

Parameters

The following parameters need to be set by the user on the command line:

• embedding_path: path to the embedding model, the file should end with .kv;
• postagging_model_path: path to the postagging model, the file should end with .pkl;
• postagging_label_path: path to the postagging label file, the file should end with .pkl;
• ner_model_path: path to the ner model, the file should end with .pkl;
• ner_label_path: path to the ner label file, the file should end with .pkl;
• file: path to the csv file the user wants to analyze;
• user_email: user's Take Blip email where they want to receive the analysis;
• bot_name: bot ID.

The following parameters have default settings, but can be customized by the user;

• node_messages_examples: it is an int representing the number of examples outputed for each subject on the Wordtree file. The default value is 100;
• similarity_threshold: it is a float representing the similarity threshold between the subject groups. The default value is 0.65, we recommend that this parameter not be modified;
• percentage_threshold: it is a float representing the frequency percentile of subject from which they are not removed from the analysis. The default value is 0.9;
• batch_size: it is an int representing the batch size. The default value is 50;
• chunk_size: it is an int representing chunk file size for upload in storaged. The default value is 1024;
• separator: it is a str for the csv file delimiter character. The default value is '|'.

Example of initialization e usage:

1. Import main packages;
2. Initialize main variables;
3. Initialize eventhub logger;
4. Initialize Insight Extractor;
5. Insight Extractor usage.

An example of the above steps could be found in the python code below:

1. Import main packages

import uuid
from TakeBlipInsightExtractor.insight_extractor import InsightExtractor
from TakeBlipInsightExtractor.outputs.eventhub_log_sender import EventHubLogSender

2. Initialize main variables

embedding_path = '*.kv'
postag_model_path = '*.pkl'
postag_label_path = '*.pkl'
ner_model_path = '*.pkl'
ner_label_path = '*.pkl'

user_email = 'your_email@host.com'
bot_name = 'my_bot_for_insight_extractor'
application_name = 'your application'

eventhub_name = '*'
eventhub_connection_string = '*'

file_name = '*'
input_data = '*.csv'
separator = '|'

similarity_threshold = 0.65
node_messages_examples = 100
batch_size = 1024
percentage_threshold = 0.7

3. Initialize eventhub logger

correlation_id = str(uuid.uuid3(uuid.NAMESPACE_DNS, user_email + bot_name))
logger = EventHubLogSender(application_name=application_name,
                           user_email=user_email,
                           bot_name=bot_name,
                           file_name=file_name,
                           correlation_id=correlation_id,
                           connection_string=eventhub_connection_string,
                           eventhub_name=eventhub_name)

4. Initialize Insight Extractor

insight_extractor = InsightExtractor(input_data,
                                     separator=separator,
                                     similarity_threshold=similarity_threshold,
                                     embedding_path=embedding_path,
                                     postagging_model_path=postag_model_path,
                                     postagging_label_path=postag_label_path,
                                     ner_model_path=ner_model_path,
                                     ner_label_path=ner_label_path,
                                     user_email=user_email,
                                     bot_name=bot_name,
                                     logger=logger)

5. Insight Extractor usage

insight_extractor.predict(percentage_threshold=percentage_threshold,
                          node_messages_examples=node_messages_examples,
                          batch_size=batch_size)