Sentiment Analysis Module

Social media is a window to people mind. Sentiment Analysis module uses social networks to learn what their users think about a brand, product, or event to provide valuable information for businesses.

 

SOLUTION

Understanding customers’ reactions, opinions, and perception is key to perform marketing analysis, plan and adjust advertising campaigns, and evaluate social media response to specific events. Our Machine Learning team decided to create a handy tool for that and developed a module that accesses social media via their APIs and processes millions of comments, reviews, posts, or tweets related to a specific topic.

To ensure efficient opinion mining, the team has compared different machine learning approaches to get most evaluated algorithms and implement pre-processing trained on different datasets. Best linear algorithms and neural networks they have combined in a model that became a basis for Sentiment Analysis module.

Sentiment Analysis module works as follows:

  • Preprocesses texts of comments, reviews, posts, or tweets received from social networks; removes stop words; extracts features based on vector representation of words using SentiWordNet;
  • Returns probability of Positive/Negative sentiment associated with the event or notion under analysis;
  • If required, provides a full package of quality metrics, such us accuracy, f1 score, ROC-AUC, etc.

The module was trained on millions reviews from different sites, and reached high quality of analysis.

 

RESULT

Due to meticulous preprocessing and prediction refinement, accuracy of sentiment analysis reached 70-90% depending on the area and dataset balance.

Client: 

In-house solution developed by Machine Learning experts group at Sigma Software

Description: 
An opinion-mining module to evaluate and calculate metrics of positive and negative sentiment
Role: 
Full-cycle development from a concept to the final solution
Team and Duration: 
Project duration is 3 months with an average of 4 FTE
Country: 
Technologies: 
Python, Scikit-learn, NLTK, Pandas, Keras; Google word2vec, SantiWordNet
Solutions: 
Contact us
Contact us