28 Apr 2021 How AI can Bring Social Listening to the Next Level
Nowadays, being present on social media is almost compulsory for an enterprise. As Jeff Bezos said: “Your brand is what other people say when you are not in the room”.
However, what is not so prevalent is a proper analysis of social media data. It is often difficult for organisations to create an automated solution which can enable them to fetch this data from platforms, process it, or enable their business users to make correct decisions based on it. There is a vast amount of information on social media sites, and this can be overwhelming for many organisations. Moreover, organisations need to understand how to make use of Advanced Analytics techniques to go beyond prescriptive and descriptive analysis.
1. Social Listening Advanced Analytics “Pill”
At ClearPeaks, we were aware of this gap; which is why we decided to develop a solution that could help marketing and strategy professionals address it. We focused on the following features while building the solution:
- Focused on competitive analysis
- Real-time capabilities
- Use of Advanced Analytics techniques
- Customisable
- Easily deployable
For the Proof of Concept (PoC), we went for Twitter as a social media platform to analyse, due to the following reasons:
- It is one of the top 3 social media platforms with a massive daily active customer base.
- Use of hashtags (#) and mentions (@), which are easy to identify.
- Twitter has like and retweet features, directly related to impact.
- It has the best API facilities, which are key to creating an automated solution.
2. Advanced Analytics techniques
Understanding how Data Science can help analyse your Twitter data is particularly important. Most of the well-known applications of Data Science are related to supervised predictions, for example: how many units will I sell next month, or what is the probability for each one of my customers to churn? However, these scenarios are not extremely useful in social media analysis.
What is in fact useful in this context are algorithms that work with words and context. This type of language-related algorithm lies within an AI subdomain called Natural Language Processing (NLP), which is the field that is leveraged for this solution.
Specifically, the data enrichment processes used for this solution are:
- Sentiment Analysis: Every tweet will have a sentiment value that ranges from -1 (bad sentiment) to 1 (good sentiment).
- Topic modelling: Tweets will be clustered into topics, depending on the similarity between them considering their words. However, the process of interpreting and naming the topics needs to be manual, as the algorithm only provides cluster assignments. A Business Analyst, with the help of a Data Analyst, should be able to interpret these topics and label them. This is the only manual intervention in the entire solution.
3. Solution Dashboard
As the final product of the solution, there is a complete dashboard for all relevant metrics and insights. After some research, we found out that one of the most valuable use cases is the ability to compare concepts based on keywords, for example:
- Compare Company X to Company Y
- Compare Product A to Product B
In our instance, we decided to use two of the most known sport retail brands out there: Nike and Adidas. We will go through the dashboard while explaining some of the potential use cases that could arise from it.
3.1. Summary
The summary page of the dashboard contains the following main metrics and dimensions:
- Sentiment Analysis
- Topic Modelling
- Top Influential Users
This page enables business users to quickly identify significant problems or opportunities. For example, users can look for relationships between sentiment and topic for a specific day and keyword.
3.2. Comparison
The comparison page is a fantastic way to compare the evolution of multiple metrics against time by keyword at once. What is interesting is seeing how they can evolve in different ways, and trying to get indicators of behavioural structures. It is important to also note that these metrics (number of views, virality, etc.) are not intrinsically good; it could be that bad tweets (having low sentiment analysis) go viral and have many views. So, careful consideration is required while incorporating this information into your decision-making.
3.3. Trending
The trending page is directly related to topics. Here, it is possible to go deeper and delve into the relationship between sentiment and topics over time, by keyword. You can also see the most frequently used words in the tweets. Additionally, this dashboard enables business users to check that topic naming makes sense. When selecting a topic, the most frequent words related to it will be displayed. It is important to remember that topic naming is a process that requires business understanding, as topics and associated words can (and will) evolve with time.
3.4. Influencers
This page puts together information about influence, sentiment, and meaning. Top users can be seen from different perspectives: topics, views, etc. The scatterplot is an enormously powerful tool, as it maps influential users considering reach (y axis) and sentiment (x axis). It is clearer with an example: let us suppose that there is an influential user with high reach that is tweeting good things about your company. It may be interesting to partner with this user to make use of their influence and start promoting your brand with them.
3.5. Detail
As in most dashboards, it is especially important to be able to have the lowest granularity in terms of information. This last page is extremely useful if there are specific tweets that the user needs to check, for example viral tweets or tweets from influential users.
Conclusion
At ClearPeaks, we believe in enabling business users with the right set of insights. To this end, we have created a comprehensive solution to provide competitive analysis to organisations based on social media data by applying Machine Learning techniques to enrich the analysed information. This can be a great tool for any company to understand and leverage social media data properly. Please visit the solution page for more details and to schedule a quick demo. If you have any questions or comments, simply drop us a line and we’ll get in touch!
