Machine learning success

The initiative is a European innovation program for startups and SMEs. A promising solution approach to optimize customer data for marketing, for example, is deep learning, a subfield of machine learning (ML). Together with the start-ups Frosha.io and Recogn.ai, Uniserv wants to advance this technology.

Customer databases are prone to errors

As part of the Data Pitch initiative, Uniserv challenged start-ups in the summer to develop an adaptive machine solution. This should be trained to combine data from various internal and external sources, such as websites, open data, social media profiles or e-mails.

These sources are extremely heterogeneous. In addition, the solution had to automatically correct errors in existing data records - and thus be able to optimize customer databases. This is because errors often creep in, especially in master data such as name and address.

Simone Braun, Business Development, Uniserv, explains: "For example, information about a person is often stored in data records about the company, or company names are mentioned in the address field.

It also often happens that information about several people is only stored in one data record for a single person. Although a human being recognizes these errors immediately, they are difficult for a machine to correct. However, the self-learning technology used should surpass human capabilities."

Deep Learning detects faulty structures

The deep learning solution approach of the two European startups Frosha.io and Recogn.ai was particularly convincing. Frosha.io used so-called recurrent neural networks within the scope of the task. Recogn.ai convinced with its approach of combining knowledge graphs and deep learning for natural language processing.

Using deep learning neural networks, a machine enables itself to recognize structures, evaluate them, and improve itself in multiple runs.

In this way, the technology succeeds in making even unstructured customer data records usable for marketing measures, among other things. Both solutions also had to take into account the EU's General Data Protection Regulation (GDPR).

As part of the competition, Uniserv provided the participating start-ups with data sets containing synthetic name and address information in unstructured and semi-structured form. Frosha.io and Recogn.ai prevailed among 142 applicants in the fall - and subsequently successfully faced the critical questions of the Uniserv jury, the EU Commission and representatives of the Data Pitch Initiative in London at the end of October.

They were then accepted into the EU-funded incubator program. This gives them six months from February 1, 2018 to put their solution idea into practice with the help of financial support from the EU. During this time, Uniserv will support both start-ups as a cooperation partner by providing expertise and data sets, which the machine learning system can use to learn.

In the future, Uniserv intends to further develop the respective projects together with both start-ups. The goal is a continuous exchange of knowledge on the use of machine learning and artificial intelligence for the processing of customer data and its use in marketing initiatives.

Opportunity for startups and companies

Data Pitch is a European Union-funded incubator program that connects companies and public institutions with startups from across Europe.

As part of the initiative, a competition was held in the summer and fall of 2017. In this, companies involved in the analysis of data were able to assign tasks to start-ups and SMEs. The initiative promotes start-ups by providing financial support, and participating companies in turn benefit from new ideas.