Innosuisse - the Swiss Innovation Agency promotes science-based innovation projects by companies and universities every year. Only a few projects are awarded funding. Eventfrog impressed the jury with its business model, its vision and the passion with which it prepared its application. Eventfrog's multi-award-winning business model is disrupting the event industry with free ticketing services.
And that's what it's all about: A two-year collaboration with Lucerne University of Applied Sciences and Arts (HSLU) in the field of machine learning is primarily intended to improve data quality, which poses a major challenge due to imports from various cultural magazines, among other things. The aim is also to improve data quality with regard to event venues. The platform should also automatically recognise if an entry does not comply with the GTCs. For example, events with a racist or violent background are not tolerated and should be automatically sorted out, as should incorrectly published test events.
This includes data-based event recommendations for ticket buyers and organisers, specific recommendations and tips in the area of ticketing and event management in general.
Severin Wyss and Michael Hochstrasser, both Machine Learning Engineers at Eventfrog, answer a few questions about the exciting project and provide insights into the topic of machine learning. Marc Bravin, data scientist and doctoral student at Lucerne University of Applied Sciences and Arts, also shares his views on the multi-year collaboration between research and the private sector.
In the picture (from left to right): Severin Wyss (Eventfrog), Marc Pouly, Solange Emmenegger (both HSLU), Michael Hochstrasser (Eventfrog), Reza Kakooee, Marc Bravin (both HSLU)
Michael and Severin, congratulations on being awarded the Innosuisse project! First of all, what exactly does "machine learning" mean and what does a machine learning engineer work on in their day-to-day work?
The term machine learning covers various models and algorithms that allow a computer to extract information from data and make independent decisions based on it. While earlier programs usually worked with hard rules created by programmers, the program now "learns" these rules independently from as much data as possible.
For example, such a rule might previously have been: "If a frog is green, it is a tree frog". With machine learning, the frog species can be determined within seconds using images, sounds and lots of other unstructured data, without anyone having to formulate such a rule.
As machine learning engineers, we select the best models for a specific question and train them using many examples, which then teaches the model to perform precisely this one task.
Why is machine learning the future?
Over the last thirty years, computers have brought us a huge increase in efficiency by automating clearly structured tasks. However, until now, there have always been tasks that are monotonous and simple for humans, but pose unsolvable problems for computers. For example, it is very easy for humans to determine whether two event descriptions are the same event. However, a rule-based algorithm already reaches its limits here, as no simple rules can be derived - free text or images are too unstructured for this and even small differences in the data make it impossible for classic programs to answer this question. With machine learning, we give programs this ability and enable them to answer such questions much faster and on a larger scale than humans can. Machine learning thus increases efficiency and scalability.
Can you explain to readers in simple terms what the process of a multi-year science-based project looks like in the world of machine learning?
The process is very similar to general research for industrial purposes. We have set ourselves specific questions that we want to answer with the help of machine learning. The most important basis for this is a large amount of good quality data - because these are the learning examples for our models. Training and evaluating the models is a very time-consuming task, and the results and progress cannot be estimated at the beginning - this is where the scientific work takes place. As soon as a model is good enough "in the lab", we will integrate it into the existing application to determine whether it holds up in reality. In many areas, this task coincides with classic software development.
How was Eventfrog able to convince Innosuisse to accept our application? In your opinion, what convinced the jury about this project?
It is important to Innosuisse that academic universities and companies work together to implement science-based projects that create added value for the economy and society. We were able to convince Innosuisse because we have concrete ideas for machine learning applications that are both scientifically innovative and economically interesting. For example, the requirement for recommender systems is an innovation: These systems aim, for example, to give a customer a personal recommendation of the products & services that are most relevant to her. In our case, the products at the events are very short-lived and almost no data on purchasing behavior is known at the time of the recommendation. An additional step is needed to compare the new event with past events about which there is already information on user preferences.
The Lucerne University of Applied Sciences and Arts is a very strong partner for Eventfrog. What do you expect from the joint collaboration?
Eventfrog has two experts in the field of machine learning who will be working with three experts from Lucerne University of Applied Sciences and Arts. We maintain a close and active exchange of knowledge and can benefit from the accumulated experience of the entire research group from a large number of other projects at the university. Accordingly, our ideas will be fully scrutinized and expanded with valuable input.
What do you personally hope to gain from this two-year collaboration?
We have the wonderful opportunity to work with experts from research and industry at the same time. The most important thing for us is to gain practical experience that we can use to Eventfrog's advantage. Thanks to the collaboration with the Lucerne University of Applied Sciences and Arts, this should combine a high level of scientific complexity with real business use cases.
What positive effects can event organisers and ticket buyers look forward to?
First and foremost, the quality of the event agenda will be improved, as the entries will be automatically checked and examined for duplicates. We already offer the largest selection of events in Switzerland - in future, we want to provide customers with even better support in finding the events that are most relevant to them quickly and conveniently. We will also provide our event organisers with more personalised tips and recommendations before, during and after the event.
Interview with Marc Bravin, PhD student at the Lucerne University of Applied Sciences and Arts (HSLU)
Marc, HSLU and Eventfrog recently started a two-year project: What are you personally looking forward to the most?
I am particularly looking forward to the exciting discussions in the project team and the results of using the developed models in productive operation. In this project, we have the opportunity to work closely with experienced data scientists from the field. This enables rapid industrialization of the developed models.
Research and industry come together: What are the advantages of this?
The latest machine learning algorithms from research can be tested directly in practice and used productively. This results in direct added value for research, as well as for the real economy and, in particular, for event organisers and ticket buyers of Eventfrog. In addition, close collaboration between research and industry enables an effective transfer of knowledge and experience.
Why is the collaboration with Eventfrog and this project in particular exciting for you?
User-generated content, especially unstructured data such as text or images, will become increasingly important in the future. Efficiently monitoring such content and then deriving insights from it is enormously complex. As an up-and-coming, data-orientated company, Eventfrog has a valuable pool of data with which the latest findings from research can be tested in practice.
What goals should be achieved by the end of the project?
By using machine learning, the quality of the content entered by event organisers on Eventfrog is to be continuously analysed and improved so that the platform can be scaled as required and ticket buyers have the most attractive event agenda possible at their disposal. In addition, new recommendation algorithms are to be developed, which, for example, suggest suitable events to ticket buyers or support event organisers in increasing their reach and reaching the right target group.
Thank you for your answers and insights into the world of machine learning. We look forward to the first results!
