From June 2016 to February 2018, we conducted the Customer Prediction Platform (CPP) research project. This strategically important project was co-financed by the European Regional Development Fund (ERDF) and the Investitionsbank Berlin Brandenburg (IBB). Now that the reporting and evaluation phase has been completed, we would like to present some highlights of this project below.
CPP is a middleware for assessing the customer potential of corporate customers. The paradigm shift of CPP is that public contextual knowledge about companies is incorporated into a fully automated recommendation for end users in B2B marketing. In the case of inbound leads, for example, individual product recommendations can be calculated.
Concrete subgoals over the entire period consisted of:
- Increased transparency in customer recommendations
- Automatic categorization of new companies into user-defined categories
- Setting up a comprehensive event monitoring system
- Analyses of company websites and recorded news
Three results of the research project that we consider to be particularly important:
First, we launched the open source project DALPHI Active Learning Platform for Human Interaction to implement the continuous iterative process of improving machine learning models. This iterative improvement is achieved with the help of the Active Learning technique. From unannotated data, the model independently identifies those for which knowledge of the correct output would offer the greatest added value. These are typically those examples that are “close to the decision limit”. These examples are then presented to a human annotator. Then these now annotated examples are added to the training data set and a new model is learned with this now larger training data set. The current version of the software includes the administration of different annotation interfaces, which can be adapted to different questions.
Next, we would like to highlight the CATEGORIZE function developed in the project. This feature will particularly enrich our service Implisense. It allows you to score any company for affinity to custom profiles. This means that a lead can no longer be qualified just according to traditional criteria such as region, industry or size, but according to one’s own profiles, such as the profile of the customers of a specific product. From a technological point of view, we have succeeded in “reversing” our well-known recommendation function and finding the right profile for the individual company.
Finally, we would like to draw attention to the comprehensive event monitoring, which was also developed within the scope of this project. In addition to announcements in the commercial register, we now provide you with news, job advertisements as well as blog and social media texts. We do not only record articles from a high five-digit number of sources, but we also process each recorded article and assign it to the correct companies. Technologically, questions of Named Entity Recognition (NER), Named Entity Disambiguation and relation extraction were addressed here. In the future we will also present further approaches for automatic tagging and prioritizing. Many of our customers are already benefiting from this development in products such as our Software-as-a-Service solution Implisense Pro or our Implisense API.