|
--- |
|
license: osl-3.0 |
|
pipeline_tag: text-classification |
|
--- |
|
# Sentimental Analysis from text: A text analysis algorithm based on ETL approach with multiprocessing for parallel processing of multiple text files. |
|
This rule-based text analysis algorithm is able to utilize multiple cores of a processor, and is designed to run on a server as an engine and not a end to end web application. The algorithm takes URLs and id as input and calculates the positive, negative, ... scores from the article text and updates the values to output destination, data from here can be used for other applications. |
|
The data extraction and NLP problem is tasked with solving ways to design and implement a sentimental analysis algorithm, which analyzes textual data from web and assess its sentiments, positive, negative, or neutral, along with determining more variables such as subjectivity score, readability. Problem approach: The problem is solved using an extract, transform, load approach with multiprocessing for parallel processing of text files.<br> |
|
1) Data Extraction: |
|
• The function createdf() creates a dataframe containing URL and URL_ID from the Input.xlsx file. |
|
• The function extractdf() is a function that leverages requests liberary and a extract_urltext() function to fetch HTML content, then uses BeautifulSoup for efficient parsing of content. The title is extracted from "title" tag and article text is extracted using class td-pb-span8 td-main-content. |
|
• The extractdf() function constructs filenames with unique URL_ID and uses “with open” to write the extracted text to respective files in UTF-8 encoding.<br> |
|
2) Data Transformation: |
|
• The read_stopwords() function reads stop words from multiple files and returns a set of stop words. The clean_stopwords() uses NLTK’s word_tokenize function to tokenizes text and returns only those words not in stop words. The clean_stopwords_directory() applies clean_stopwords() function to all text files in the directory rewriting the files. |
|
• Create_posneg_dict() is a function that reads the provided files in MasterDirectiry and returns two sets, one with positive words and other with negative words.<br> |
|
3) Calculate variables: |
|
• The runengine() function runs a loop for each file the directory provided and tokenizes the texts into words (word_tokenize) and sentences (sentence_tokenize) and calculating each variable by passing these as parameters to separate functions made incorporated with the formulae provided with the problem. |
|
• The appendtodf() function appends a new columns with calculated variables for each row where URL ID matches to a dataframe. |
|
• Finally the result dataframe for each process running the runengine() function appends its respective output dataframes to a list which will be concatenated to one single dataframe.<br> |
|
4) Data Load: |
|
• The leadoutput() function stores the concatenated dataframe as Output.xlsx file to specified path.<br> |
|
|
|
The implementation of multiprocessing and error handling: |
|
The program uses multiprocessing library to create multiple processes equal to the number of logical processors(cores) found in the executing system. The number of files to process are split to each process almost equally. The processes execute the runengine() function parallelly processing separate files list divided for each process. |
|
1) Significant Performance Gains: Parallel processing effectively leverages multiple CPU cores, distributing workload and substantially reducing processing time, especially for large datasets. |
|
2) Shared Data Handling: The use of a managed list to store output dataframes by each process ensures efficient communication and data collection between parallel processes. |
|
3) Error Handling: Error handling is essential to prevent potential issues during multiprocessing, ensuring the integrity of the final output. |
|
<br> |
|
|
|
For a detailed instructions and documentation visit: https://github.com/suwesh/Multi-core_SentimentEngine |
