Analyzing product sentiments

We'd like to share our experience of using natural language processing to understand the trends in popularity of various brands. Our goal was to identify the most well-known and positively perceived brands and products, using data on web pages visited by millions of anonymous users. To achieve this, we had to extract the products mentioned on those web pages and then analyze the sentiments expressed about them.

We used Natural Language ToolKit (NLTK), a suite of powerful libraries for natural language processing in Python, to understand the attitudes expressed in the text from websites.

To represent the collective emotion associated with the entire piece of text, we introduced the concept of global sentiment of a given text with respect to a product. At the start, the global sentiment was set to zero, representing an unbiased sentiment score. We then analyzed each paragraph of the text separately.

In this paragraph Berthoud brand is described positively and are said to have a few advantages compared to Brooks (if you were wondering, these are leather bike saddles). The algorithm picks it up and updates the global sentiment.

The author has some reservations about the Aubisque brand even though they don’t criticize it. Still, the algorithm correctly identifies the sentiment with respect to that brand. This algorithm allows us to both understand general trends and determine the preferences of specific segments of customers.

If you have any questions about our sentiment analysis journey, feel free to reach out to us at info@open-d.io.