Emotion Inspired by an Idea: Sentiment Analysis

• 

  Sentiment Analysis: Key to Good Insights

Inspiration

Emotions are essential, not only in personal life but in business as well. How we feel about certain products or brand provides companies the context necessary to evaluate and improve the product, business, marketing, and communications strategy. Irrespective of the industry or vertical, brands have become imperative to understand consumers’ feelings about the brand and products. Many times, bad reviews online damaged the reputation of companies. We can’t always make bad review and go away, but the key is to find it before it gets out of hand and take action on time.

What it does

Keywords and phrases commonly used in customer service conversations can reveal product and brand insight. Negative and positive words used in product reviews can be identified using customer feedback analytics tools, which would indicate issues earlier. Analysing product sentiment requires a solid understanding of how customers think. My solution incorporates Azure Cognitive Service Sentiment Analysis feature evaluates text and returns sentiment scores and labels for each sentence. The sentiment analysis feature provides sentiment labels (such as "negative", "neutral" and "positive") based on the highest confidence score found by the service at a sentence and document-level. This feature also returns confidence scores between 0 and 1 for each document & sentences within it for positive, neutral and negative sentiment. A Power BI report are built to analyse the sentiment scores obtained over the product reviews.

How we built it

I am using Azure Cognitive Services to integrate AI and machine learning in my application. Unstructured reviews are imported into as Spark table dataset that contains a text column for sentiment analysis using Spark pool in your Azure Synapse Analytics workspace. I am using PySpark (Python) language in Synapse Studio notebook (notebook_amazon, notebook_imdb, notebook_yelp). In this application, the pre-built model for text analytics (sentiment analysis) is used. Azure Cognitive Services is configured to as linked service Azure Synapse Analytics using Azure Key Vault linked service for permission. Azure Synapse Analytics us used for data warehousing and analytics. Azure Synapse Analytics workspace with an Azure Data Lake Storage Gen2 storage account configured as the default storage. Results of Sentiment Analysis consist of a sentiment label and a confidence score. The four possible labels are positive, mixed, negative, and neutral.

Challenges we ran into

It was kind of difficult to get started because there are many ways to choose from and as a first timer in a hackathon, I tried Visual Code, Jupyter Notebook, Machine Learning Studio and many more before I decided to use the easiest. Ha ha.

Accomplishments that we're proud of

Submitting to the Hackathon is an accomplishment already due to my busy schedule, new to the knowledge and too many things to pick up in a short time.

What we learned

Practice makes perfect. Building a project for Hackathon definitely helped me to pass my AI-900 exam.

What's next for Emotion Inspired by an Idea: Sentiment Analysis

Currently, the solution uses a pre-built model. I would like to enrich the model or try different models for the sentiment analysis. Also, it would be interesting to understand how sentiment analysis treat sarcastic text. Sarcastic people express their negative sentiments using positive words.

Built With

• azure-cognitive-services
• azure-key-vault
• azure-synapse-analytics
• pyspark