A Multilingual Social Listening Approach for Brands

In 2013, a global blue chip organisation commissioned DigitalMR to harvest and analyse social media data related to one of their Personal Care product categories. The objective was to explore the web landscape around the category in six different countries and languages around the world, in order to inform their future marketing campaigns.

What are people saying about the product category? How do they feel about using the product? Where are the conversations taking place? 

1. Noise Elimination
Posts harvested from the web often come with a lot of “noise”. To be able to clean the dataset and analyse only the truly relevant posts is the first challenge. In this case, we removed noise from tens of thousands of posts from the various countries, to obtain 166,105 actually relevant posts.

2. Sentiment Analysis
The next challenge is analysing the data for sentiment in an automated way, achieving high sentiment accuracy (i.e. ≥80%). We use a machine learning algorithm to determine whether each post is positive, negative, or neutral, the accuracy of which is checked by human curators.

3. Topic Annotation
In addition to the sentiment, it is important to understand the topic(s) of the posts, the conversation drivers. Manually reading the posts is not the way to go. DigitalMR uses complex hierarchical topic taxonomies tailored to the client’s needs for semantic analysis – assigning topics, subtopics, and attributes to each post. Taxonomies can be adapted to multiple languages, countries, cultures, and of course product categories. 

DigitalMR was able to provide the client with a Social Listening report on their product category that covered consumer posts from public websites in 6 different languages. Sentiment accuracy went as high as 91% at sentence level for German.

Our findings were utilised by the client and their advertising agency to inform their future marketing campaigns.

To see charts and a client testimonial, please refer to the attached infographic.