The study deep dives into how to work with text mining techniques to extract insights from unstructured consumer-generated data. The purpose of the thesis was to develop a text mining method to derive brand association networks from unstructured online customer reviews and to explore how this method can assist in developing more effective brand management strategies. The study focused on the hotel industry and examined how brand association could be derived using a text mining approach, how these associations were related to hotel ratings, and what managerial insights could be gained.
“Students, in general, tend to focus on a qualitative method for their thesis, but we were a bit tired of that; we wanted to use a quantitative approach and do something new and exciting. Given how much information there is in the form of customer reviews, we thought there must be a way of structuring such data, and that’s how we got into that topic. Luckily, one of us had some prior knowledge within coding, and hence, we were able to develop a method where we built most of the automation from scratch. With our method, we hope to contribute with an understanding of how to retrieve this type of text data and how to structure it to be able to gain valuable insights.”
– Thesis authors
The winning thesis was chosen by an internal jury comprised of co-workers at Precis, and as the motivation state below, they were impressed by the students choice of method;
“With a high degree of innovativeness, the winning students have used a relevant methodology and contributed new insights to the field of digital marketing. The jury agreed that the thesis holds an overall high quality; it is well-written, investigates an interesting topic, as well as introduced a methodology that could be relevant for creating another dimension in digital marketing. We were also very impressed that the students used automation, a good amount of data, and the latest technologies to come to their conclusions.”
– The Precis Jury
Even if the study focused on the hotel industry, the text mining framework can be applied to other industries that have unstructured consumer-generated data such as social media posts or open survey answers. The study shows that vast amounts of unstructured user-generated content on the web can derive valuable strategic insights within industries where competition is dependent on maintaining a favourable brand image. The techniques and models proposed in the study can, hence, provide insights valuable for the planning and implementation of their digital marketing strategies.
“There are, of course, studies around customer reviews and where such data has been examined, but we have been given the impression that most companies retrieve the information, but they do not link it with marketing. Text mining approach to online customer reviews are kind of unexplored, but you can gain a lot of useful insights from them. You can have thousands of answers to what your customers really think about your brand, but you often have a hard time getting a complete picture. That goes for unstructured data in general, it is often difficult to get an overview and a complete picture”.
– Thesis authors
The study explores and integrates new data science and machine learning techniques in the field of marketing, and can be viewed as a complement to other data sources and methods to gain insights within brand management strategies.
“We think that this method should not be a substitute for the other methods or ways of collecting data, but more as a complement. Of course, one should not ignore other ways of collecting data, such as questionnaires and focus groups, because they have such great value and that you can deep dive into details and get narrow insights.
Our data gives a more general picture. The advantage is since we have so much data, you can actually say something general about the result. We have done T-tests, and other statistical tests, where we have seen that the results of the show a high probability. This is something you may not be able to do in a focus group, for example, where there is such a small selection.
On the other hand, because our insights are on a fairly general level, it may be difficult to get a more detailed picture. With our method, you remove the extreme opinions on both ends, and we sort of only get the broad mass and not the “niche insights”. But of course, there are advantages and disadvantages to all methods.”
– Thesis authors
The thesis can be found here.
Great work and again, congratulations Viktor, Rebin and Jakob!