Skip to main content
Lucidworks AI custom embedding model training lets you train and deploy custom models using the Custom model training user interface or the Lucidworks AI Models API.

​
Training data

Custom embedding model training uses the following types of data:
  • Signals acquired from user interactions on your website
  • Customer questions and answers your organization collects from logs and forums your organization collects
  • Information from documents obtained from queries on your website
  • Datasets and other information labeled on your website
For more detailed information, see use case training data.

​
Custom model training process diagram

This diagram displays the process flow for custom model training. Custom model training process diagram

​
RNN models

Recurrent neural network (RNN) models are excellent for understanding sequential data such as text, logs, conversations, and user behavior patterns. This provides comprehensive and accurate predictions. Lucidworks AI supports the following types of recurrent neural network (RNN) models:
  • General. The general RNN model is typically used by knowledge management sites that contain information, such as news or documentation. Time-efficient and accurate information based on semantic understanding enhances the user search experience.
  • ecommerce. An ecommerce RNN model is typically used by sites that provide products for purchase. For B2B organizations, these models effectively analyze long-term customer account activity, which provides information to enhance customer product relevancy. For B2C organizations, one use of ecommerce RNN models is to provide the ability to search real-time product recommendations, which improves conversions.
For more detailed information, see RNN models.

​
Transformer RNN models

Lucidworks AI supports the transformer with the RNN head model, which is a hybrid model that combines the extensive semantic understanding of transformer encoders with client data tuned and more compact output of the RNN head model. Best practices recommend that only B2B and knowledge management organizations use transformer RNN models. B2B organizations receive accurate tuned relevance in addition to improved associations of domain specific terms and product correlations. Knowledge Management organizations can implement recommenders based on user_sessions-item or item-to-item where the item could be content like a document. For more detailed information, see Transformer RNN models.

​
Custom configuration

Custom configuration parameters can be used to train a custom model using either the Custom model training user interface or the Lucidworks AI Models API. For more detailed information, see Custom configuration.