Sentiment Analysis with Topic Modelling

What is it?

Sentiments, feelings and emotions are an integral part of human nature. Over 2.5 quintillion bytes of data are created every single day. Twitter itself generates 500 million tweets every day. Sentiment Analysis deals with examining the study of textual data like posts, blogs, reviews, etc. expressed by users regarding their views and opinions about a specific product, service, event, idea, news or person.

The most common use of Sentiment Analysis is classifying the text into class(es). It can either be binary like positive/negative, yes/no, etc or multi-class like positive-negative-neutral, yes-no-maybe, etc. Sentiment Analysis aids organisations to track the following:

  • Brand popularity and acceptance
  • New product anticipation and perception
  • Company reputation
  • Understand customer experience

What we did?

During our analysis of social media comments for one of the biggest cosmetic giants of US and France, we followed the protocol of Natural Language Processing (NLP). We first analyse raw data which contains not only text but also tons of emoticons. Besides analysing long reviews and posts that mention multiple products, our challenges were complicated sentences, sarcasm, slangs. For example, in beauty parlance, “It just sucks everything from my pores” would mean a positive sentiment OR “I am dying to use ABC product” would mean a positive sentiment.

Our solution design not only concentrated around classification of data as positive/negative/neutral, but also topic modelling. Topic modelling is a process to automatically detect topics present in the text and derive hidden patterns in the corpus and thus assist in better decision making. Topics can also be defined as repeated pattern of most occurring terms in a corpus of text.

As shown above, in a snapshot, the terms which were relevant for the highlighted topic in red gives an indication of what is important to the user for that term. For e.g. When the topic “glow” is used, terms like “abhcosmetics”, “Maybelline”, “purpleheart”, etc. are most frequently used. This provides a detailed insight to the business owners for marketing the right products in the right category, or to provide the right offers on brands when the customer is looking for “glow”.

Where is the future?

Topic Models are very advantageous in domains like HR Recruitments where the right candidates can be mapped to the right job profile. It can also aid the News industry where the correct articles are recommended to the right set of users. Or to gauge public opinion to policy announcements and campaign messages ahead of presidential elections. Thus, sentiment analysis along with topic modeling can not only help you know what people are thinking about you but also about your competitors. For a customized solution to your business problem, please feel free to get in touch with us.

Thanks! You've already liked this
No comments