Which companies are interested in sentiment analysis?
Understand your own customers and industry better
By Oliver Schick on July 01, 2020
Which topics are currently being discussed in my industry? Which current trends are important to customers for my products? What is the attitude of my customers towards my company? If you have similar questions, an AI analysis of text data such as social media data could be useful to get a little closer to the answers. In recent years, social media applications in the B2C as well as B2B area have become increasingly important. Analyzing the data can provide unique insights into an industry. Due to the large amount of data, however, it is not possible to examine them manually. Automated analysis methods enable precisely these analyzes.
One type of analysis is the identification of the most discussed topics (topic modeling). So-called topic modeling is about assigning the content of many text files to higher-level topics. This enables topics that are discussed on social media platforms to be identified. On the one hand, this enables an overview of all relevant topics and, on the other hand, contributions can be assigned to the appropriate topic and, if necessary, examined manually.
In this showcase, we analyzed the posts on the Twitter platform. The analysis includes all German tweets that were published between May 30th and June 4th and contain either the word “KMU” or “Mittelstand”. Using a probabilistic model called Latent Dirichlet algorithm the topics were generated from the tweets. Figure 1 shows the results in a heat map. The topics are made up of the words on the axes. The darker a field, the more likely these words will appear together in a topic. If many of these dark fields lie next to one another, they form a cluster or a theme. The dendrograms to the left and above the matrix indicate how close the clusters are. The closer two topics are, the flatter the connection via a dendrogram. Another form of visualization is the interactive dashboard in Figure 2.
The circles on the right represent the overarching topics. The larger a circle, the more often this topic was discussed. By selecting the topics on the left, the keywords that make up a topic can be viewed on the right. The bars for the keywords give an insight into how often a keyword occurred overall in the documents (entire bar) or in the texts that deal with this topic (red part of the bar). The weighting (lambda parameter) of the keywords can be changed by using the slider. If lambda is one, the keywords that are absolutely most common in the selected topic are weighted higher. If the lambda value is reduced, it is also taken into account how high the relative frequency of a keyword is for the selected topic. If the lambda value is zero, only keywords that appear exclusively in texts on the selected topic are taken into account. In the analysis you can play with the parameter lambda. As a rule of thumb, values between 0.3 and 0.7 give the best insight.
Topic modeling is worthwhile
The use of topic modeling opens up these innovative solutions, among others:
- Automatic labeling of customer support tickets by topic.
- Recognize patterns and provide results in the form of common words and expressions.
- Automatic forwarding of messages to the most suitable team. For example, tickets that have billing problems, refunds, or contain expressions such as “credit card transaction”, “subscription error”, and so on, are sent to the accounting department.
- Automatic detection of urgency of a support ticket and corresponding prioritization. For example, if a ticket is flagged as “bug” or “urgent”, or a computer detects expressions like “immediately”, “immediate attention”, etc., it will be forwarded to the PR department to avoid a potential PR crisis.
- Improved product development by recognizing product features that are often mentioned, which can provide information on possible errors in the product or reveal that some features are not used at all.
Another useful form of machine text analysis is sentiment analysis. This is often applied to textual or acoustic customer ratings received via email, social media or other channels. The goal of sentiment analysis is to predict the emotions of a post's writer. This means that customer support can respond more specifically to customer questions or quickly to social media posts that are particularly emotional. In order to evaluate the emotions automatically, the grammar and semantics of contributions are systematically analyzed. In the case of spoken content, the voice color and the intonation can also be taken into account. Equipped with the knowledge of the mood of a user, the customer service, the marketing department or the sales department can react optimally to contributions.
Sentiment analysis can make a valuable contribution in the following areas:
- Addressing customers specifically: By tracking customers who have negative comments about a product or their business as a whole, customer service can target their issues. If a customer writes a comment in anger, the situation can be quickly clarified through personal contact or prioritization of the ticket.
- Track customer sentiment over time: The development can be followed by analyzing customer satisfaction over time. If you combine the mood trend with the topics identified as relevant, you can find explanations for the influencing factors.
- Identification of the customer segments that are more interested in your company: Combined with demographic and other quantitative data, it is possible to segment the customer base and look at their sentiment in isolation. Do customers who spend less write, for example, more negatively than customers who spend more? Or are there problems with the return policy of customers in Munich and not of customers in Stuttgart?
- Track the effects of product changes on customer satisfaction: When the business changes, so does customer sentiment. Publishing a marketing campaign or press release, changing the user interface or the pricing structure of your product can have an impact. This can be measured by automatically analyzing customer sentiment.
Topic modeling and sentiment analyzes are therefore very suitable tools for answering the questions raised in the introduction. In addition to the tools presented and the associated application examples, there are many other methods and areas of application that you can use to better understand your customers and industry. If you would like to take advantage of the opportunities for your company, we look forward to helping you!
- What's great to learn
- Habit building drugs are inherently bad
- What are the side effects of cellulose
- How strong can body protection be
- How can I feel pure
- Is price a good determinant of quality?
- What is 25 12 4 3
- How long is a cat memory
- What are subjective social evils
- Has LeBron already reached its heyday
- What is tax 6
- Is Kaka a legend
- What is SOX conformity
- What is a promise
- What is Medical Student Syndrome in Psychology
- Does RAM affect the GPU temperature
- What's wrong with Odisha
- What should everyone know about North Carolina
- Can illegal immigrants renew a passport
- Bacopa is a nootropic
- How should I perform in tomorrow's interview
- How to learn Japanese more easily
- Is Turkey ready for the space industry?
- What rhymes with rick