Multiplier case: Natural Language Processing at Volkswagen Infotainment<\/h2><\/div>The use of natural language processing in Volkswagen Infotainment\u2019s patent process<\/h4><\/div><\/div><\/div><\/div><\/div>Summary<\/h3><\/div><\/div><\/div>Patents are a key factor in competitiveness, particularly for technology companies. At the same time, however, the patent application process involves high costs due to the significant amount of manual work involved and the specialist knowledge required. This is particularly true of the novelty search stage. The use of AI technology offers great potential here for shortening processing times, reducing costs and boosting innovation.<\/p>\n<\/div><\/div><\/div>
![](\"https:\/\/humaine.info\/wp-content\/uploads\/2025\/04\/Bildschirmfoto-2025-04-22-um-11.54.25.png\" "\\"Bildschirmfoto")
<\/span><\/div><\/div><\/div>Description of the Approach<\/h3><\/div>The project focuses on investigating the extent to which related features from domain-specific descriptions (technical development and patenting) can be embedded in similar vectors. To this end, the following steps will be undertaken: collection of data linking development and patent terminology; transformation of the data into training examples for a neural network; design and implementation of a training architecture; and validation of the output.<\/p>\n<\/div><\/div><\/div>
Goals & Results<\/h3><\/div>It became apparent even during the training phase that ideas described in domain-specific terminology can be superimposed on one another in the vector space and made searchable. Lists of unknown patents can be sorted according to their relevance to a particular idea. It was also demonstrated that individual, sufficiently long chunks of patents contain the semantics required for linking. The approach developed thus shows potential for further generalisation.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
Partner<\/h1><\/div><\/div><\/div>Practice Partners<\/h3><\/div>
The use of natural language processing in Volkswagen Infotainment\u2019s patent process<\/h4><\/div><\/div><\/div><\/div><\/div>Summary<\/h3><\/div><\/div><\/div>Patents are a key factor in competitiveness, particularly for technology companies. At the same time, however, the patent application process involves high costs due to the significant amount of manual work involved and the specialist knowledge required. This is particularly true of the novelty search stage. The use of AI technology offers great potential here for shortening processing times, reducing costs and boosting innovation.<\/p>\n<\/div><\/div><\/div>
<\/span><\/div><\/div><\/div>Description of the Approach<\/h3><\/div>The project focuses on investigating the extent to which related features from domain-specific descriptions (technical development and patenting) can be embedded in similar vectors. To this end, the following steps will be undertaken: collection of data linking development and patent terminology; transformation of the data into training examples for a neural network; design and implementation of a training architecture; and validation of the output.<\/p>\n<\/div><\/div><\/div>
Goals & Results<\/h3><\/div>It became apparent even during the training phase that ideas described in domain-specific terminology can be superimposed on one another in the vector space and made searchable. Lists of unknown patents can be sorted according to their relevance to a particular idea. It was also demonstrated that individual, sufficiently long chunks of patents contain the semantics required for linking. The approach developed thus shows potential for further generalisation.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
Partner<\/h1><\/div><\/div><\/div>Practice Partners<\/h3><\/div>
Summary<\/h3><\/div><\/div><\/div>Patents are a key factor in competitiveness, particularly for technology companies. At the same time, however, the patent application process involves high costs due to the significant amount of manual work involved and the specialist knowledge required. This is particularly true of the novelty search stage. The use of AI technology offers great potential here for shortening processing times, reducing costs and boosting innovation.<\/p>\n<\/div><\/div><\/div>
<\/span><\/div><\/div><\/div>Description of the Approach<\/h3><\/div>The project focuses on investigating the extent to which related features from domain-specific descriptions (technical development and patenting) can be embedded in similar vectors. To this end, the following steps will be undertaken: collection of data linking development and patent terminology; transformation of the data into training examples for a neural network; design and implementation of a training architecture; and validation of the output.<\/p>\n<\/div><\/div><\/div>
Goals & Results<\/h3><\/div>It became apparent even during the training phase that ideas described in domain-specific terminology can be superimposed on one another in the vector space and made searchable. Lists of unknown patents can be sorted according to their relevance to a particular idea. It was also demonstrated that individual, sufficiently long chunks of patents contain the semantics required for linking. The approach developed thus shows potential for further generalisation.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
Partner<\/h1><\/div><\/div><\/div>Practice Partners<\/h3><\/div>
Patents are a key factor in competitiveness, particularly for technology companies. At the same time, however, the patent application process involves high costs due to the significant amount of manual work involved and the specialist knowledge required. This is particularly true of the novelty search stage. The use of AI technology offers great potential here for shortening processing times, reducing costs and boosting innovation.<\/p>\n<\/div><\/div><\/div>
<\/span><\/div><\/div><\/div>Description of the Approach<\/h3><\/div>The project focuses on investigating the extent to which related features from domain-specific descriptions (technical development and patenting) can be embedded in similar vectors. To this end, the following steps will be undertaken: collection of data linking development and patent terminology; transformation of the data into training examples for a neural network; design and implementation of a training architecture; and validation of the output.<\/p>\n<\/div><\/div><\/div>
Goals & Results<\/h3><\/div>It became apparent even during the training phase that ideas described in domain-specific terminology can be superimposed on one another in the vector space and made searchable. Lists of unknown patents can be sorted according to their relevance to a particular idea. It was also demonstrated that individual, sufficiently long chunks of patents contain the semantics required for linking. The approach developed thus shows potential for further generalisation.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
Partner<\/h1><\/div><\/div><\/div>Practice Partners<\/h3><\/div>
The project focuses on investigating the extent to which related features from domain-specific descriptions (technical development and patenting) can be embedded in similar vectors. To this end, the following steps will be undertaken: collection of data linking development and patent terminology; transformation of the data into training examples for a neural network; design and implementation of a training architecture; and validation of the output.<\/p>\n<\/div><\/div><\/div>
Goals & Results<\/h3><\/div>It became apparent even during the training phase that ideas described in domain-specific terminology can be superimposed on one another in the vector space and made searchable. Lists of unknown patents can be sorted according to their relevance to a particular idea. It was also demonstrated that individual, sufficiently long chunks of patents contain the semantics required for linking. The approach developed thus shows potential for further generalisation.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
Partner<\/h1><\/div><\/div><\/div>Practice Partners<\/h3><\/div>
It became apparent even during the training phase that ideas described in domain-specific terminology can be superimposed on one another in the vector space and made searchable. Lists of unknown patents can be sorted according to their relevance to a particular idea. It was also demonstrated that individual, sufficiently long chunks of patents contain the semantics required for linking. The approach developed thus shows potential for further generalisation.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
![](\"https:\/\/humaine.info\/wp-content\/uploads\/2025\/02\/Logo_VW-Infotainment.jpg\"){kind=logo}
<\/a><\/span><\/div>![](\"https:\/\/humaine.info\/wp-content\/uploads\/2025\/02\/Logo_Lehrstuhl-fuer-Produktionssysteme-LPS.png\"){kind=logo}
<\/a><\/span><\/div>