Domain Name Generation Using Latent Space Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current domain name generation techniques are limited in their ability to produce targeted and customized domain names that are relevant to individual users, often requiring extensive search and registration processes, leading to increased bandwidth usage, processor usage, and wasted user time.
Innovation Solution
A computer-implemented method using a variational autoencoder with a trained encoder and decoder neural network to map seed domain names into a low-dimensional information-rich latent space, allowing for the generation of new domain names that are syntactically and semantically similar by modifying and decoding these representations, enabling fine-grained control over features like length, topic, and syntax.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional domain name generation techniques are used, then domain names can be generated, but the generated domain names lack targeting and customization, requiring extensive search and registration processes
Solution Approach 1:
The patent transforms domain name generation by changing parameters from traditional keyword-based methods to a latent space representation approach. By encoding domain names into a continuous latent space and manipulating vectors in this space, the system can generate customized domain names by adjusting latent parameters, enabling fine-grained control over domain name characteristics and significantly reducing search time
Solution Approach 2:
The patent introduces a new dimensional approach by mapping discrete domain names into a continuous low-dimensional latent space. This dimensionality transformation allows for smooth interpolation and manipulation between domain names, enabling targeted generation of customized domain names without requiring extensive sequential searching
2Productivity
If traditional domain name generation techniques are used, then domain names can be generated, but bandwidth usage and processor usage increase due to extensive search processes
Solution Approach 1:
The patent applies preliminary action by pre-training the encoder-decoder model on a large corpus of domain names before actual generation. This offline training phase captures domain name patterns and structures in the latent space, so that during actual use, domain names can be generated quickly through simple vector operations without requiring extensive online searching and processing
Solution Approach 2:
The patent uses copying by creating representations of domain names in latent space rather than working with the actual domain names themselves. By manipulating these compressed vector representations, the system can generate new domain names with minimal computational overhead, avoiding the need to process and compare numerous actual domain name strings
3Adaptability or versatility
If a low-dimensional latent space is used to represent domain names, then targeted and customized domain names can be generated, but the system complexity increases due to the variational autoencoder model
Solution Approach 1:
The patent applies segmentation by dividing the domain name generation task into distinct components: an encoder that maps domain names to latent space, a latent space for storage and manipulation, and a decoder that reconstructs domain names from latent vectors. This segmentation allows each component to be optimized independently and simplifies the overall system architecture despite the sophistication of the variational autoencoder model
Data Source
AI summary
A computer device, a computer-readable medium and a computer-implemented method is provided. The computer device and the computer-readable medium perform the computer-implemented method of generating unregistered internet domain names using a trained encoder function and a trained decoder function. The method includes obtaining one or more seed domain names; mapping, using a hardware processor implementing the trained encoder function, the one or more seed domain names in a low-dimensional information-rich latent space; modifying a representation of one domain name of the one or more seed domain names that is encoded in the low-dimensional information-rich latent space; and generating, using the hardware processor implementing the trained decoder function, one or more unregistered domain names for domain name registration from the representation that is modified.