For a brand to be successful it has to address challenges including but not limited to differentiation, brand identity, cost, complexity, consistency, and adaptation to changing market trends. The health of a brand is determined by its awareness, reputation, positioning, delivery, and engagement. To assess the health, data is collected using multiple surveys and analyzed for insights. Traditionally, the process is slow, resource intensive and hence cannot be repeated very often. Results of changes made by the brand can only be observed after another iteration. Thus, traditional methods of brand health monitoring are slow, resource-intensive, and geographically limited.
Social media provides a valuable source of information through its broad reach and candid user feedback. Online sources like Google reviews and Facebook offer valuable insights into consumer opinions. Data aggregation and processing from social sites can provide valuable insights into customer sentiment and business aspects. Google reviews also enables data analysis for individual locations, allowing for comparisons between different locations for the same brand and determining its health.
How can it be done?
This can be achieved by using advanced Natural Language Processing (NLP) techniques based on Language Models (LLMs). Natural language understanding of LLMs can effectively identify customer sentiment and categorize comments for businesses. Sentiment analysis that represents customer’s sentiment regarding different categories of a business is called Aspect Based Sentiment Analysis (ABSA), where each aspect represents a business category.
What does our system do?
Our solution automates data gathering from social media to analyze public opinion about a brand. Customer comments are categorized as positive or negative. BrandSense classifies food sector based on factors like: Food Quality, Service Quality, Physical Environment, Staff, Menu, Price Fairness and telecommunication sector reviews based on factors like: Area Coverage, Fiber Cut, No Service, Browsing Speed, Customer Service, and Billing Issues.
To automate the process of brand health monitoring, the BrandSense algorithm utilizes fine-tuned natural language processing algorithms to perform three major tasks:
The process starts with the language model identifying the language in which the text data is written. The language identification model, which uses a transformer model, was trained specifically on roman Urdu language as this helps the BrandSense to identify the language according to the data.
Aspect identification is the process of categorizing sentences into specific categories of interest, defined by businesses. The classifier must be trained for each brand or category i-e for food sector such as food quality, service quality , physical environment, staff, menu, and price fairness. The algorithm can be customized based on client requirements, and the differences in categories vary by sector. For example, in telecommunications, different categories may exist, such as area coverage, fiber cut, no service, browsing speed, customer service, and billing
Sentiment analysis is done by selecting the classifier based upon the language of the review. We have trained and fine-tuned separate classifiers for Urdu and English languages. The selection of the classifier depends upon the result of the language classifier in the first stage. The sentiment analyzers based on transformer technology are also fine-tuned using customer specific data.
- Location based sentiment analysis using data from Google Reviews
- Fine tuning of aspect & sentiment classifiers on the data of the client in Roman Urdu
- Prediction of customer churn from reviews
- Ability to identify requests made by customers regarding certain aspects of service
- Identify multiple aspects and sentiments in a single review
- Capability: Ability to perform sentiment analysis for different categories (aspects) of the business.
- Accuracy: Our aspect based sentiment analysis system has an accuracy of more than 85%.
- Customization: Our ABSA system can be quickly fine-tuned for specific test cases by using customer data thus making it customizable.
- Flexibility of deployment: The system can easily be deployed on local and cloud infrastructure.
- Robustness: The system is capable of identifying multiple aspects and sentiments in a single statement.
- Languages: Currently our ABSA system supports English and Roman Urdu