Task-oriented clustering using prompt learning
Systems, computer-implemented methods, and computer program products to facilitate capturing relative importance of relational entities for building database embedding models are provided. According to an embodiment, a system can comprise a processor that executes components stored in memory. The computer executable components can comprise a template component that utilized natural language as a prompt template to describe a perspective of clustering and assembles description information into the prompt template to generate a base model. The computer executable components can comprise a training component that can utilize data in the prompt template to automatically build training data of an adapter to generate a final model. The computer executable components can comprise a vector generator component that inputs the prompt template to the final model to generate one or more hidden layer vectors highlighting characteristics of the natural language.
1 . A system, comprising:
a processor that executes computer-executable components stored in a non-transitory computer-readable memory, wherein the computer-executable components comprise:
a template component that utilizes natural language as a prompt template to describe a perspective of clustering, and assembles description information into the prompt template to generate a base model that defines a clustering objective expressed in natural language;
a training component that utilizes data in the prompt template to automatically build training data derived from the prompt template itself, generating a final model by training an adapter of the base model using the automatically-built training data, without requiring externally labeled clustering, wherein the adapter comprises a neural network; and
a vector generator component that inputs the prompt template to the final model after the adapter training to generate one or more latent vectors from an intermediate representation of the final model highlighting characteristics of the natural language,
wherein the one or more latent vectors are used to perform a final clustering in latent space based on the clustering objective defined by the prompt template.
2 . The system of claim 1 , wherein the latent vectors comprise one or more hidden layer vectors generated by an intermediate layer of the neural network, and the one or more hidden layer vectors are used as a highlighted feature by the prompt template for the final clustering.
3 . The system of claim 2 , wherein the prompt template outputs a hidden state that is restored to an image enhanced with visual information.
4 . The system of claim 3 , wherein the hidden state reduces the visual information of the image in accordance with the natural language of the prompt template.
5 . The system of claim 4 , wherein the prompt template enlarges the visual information of the image in accordance with the natural language of the prompt template.
6 . The system of claim 1 , wherein the training component fine-tunes the base model with less than 1% of trainable parameters while freezing a backbone of the base model.
7 . The system of claim 2 , wherein the final model dynamically adjusts generation of the one or more hidden layer vectors in real time.
8 . A computer implemented method for utilizing prompt learning to perform topic-wise clustering of data, the computer implemented method comprising:
utilizing, by a device operatively coupled to a processor, natural language as a prompt template to describe a perspective of clustering by encoding semantic distinctions associated with a plurality of topics;
assembling, by the device, description information into the prompt template to generate a base model configured to respond to variations in the prompt template;
utilizing, by the device, data in the prompt template to automatically build training data derived from the prompt template itself;
generating a final model by training an adapter of the base model using the automatically-built training data, without relying on externally annotated topic labels to adapt the base model for topic-wise differentiation, wherein the adapter comprises a neural network;
inputting, by the device, the prompt template to the final model after the adapter training to generate one or more intermediate representations comprising vectors highlighting one or more characteristics of the natural language associated with respective topics; and
performing, by the device, a final clustering in latent space by grouping data based on similarities among the vectors highlighting the one or more characteristics of the natural language.
9 . The computer implemented method of claim 8 , further comprising: using, by the device, latent vectors that comprise one or more hidden layer vectors generated by an intermediate layer of the neural network, and using the one or more hidden layer vectors as a highlighted feature by the prompt template for the final clustering.
10 . The computer implemented method of claim 9 , further comprising: outputting, by the device, a hidden state from the prompt template that is restored to an image enhanced with visual information.
11 . The computer implemented method of claim 10 , further comprising: reducing, by the device, the visual information of the image in accordance with the natural language of the prompt template.
12 . The computer implemented method of claim 11 , further comprising: enlarging, by the device, the visual information of the image in accordance with the natural language of the prompt template.
13 . The computer implemented method of claim 9 , further comprising: tuning, by the device, the base model with less than 1% of trainable parameters while freezing a backbone of the base model, wherein the final model dynamically adjusts generation of the one or more hidden layer vectors in real time.
14 . A computer program product for utilizing prompt learning to perform topic-wise clustering of data, the computer program product comprising a non-transitory computer readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
utilize natural language as a prompt template to describe a perspective of clustering by encoding topic-related semantic distinctions within the prompt template;
assemble description information into the prompt template to generate a base model responsive to semantic variation in the prompt template;
utilize data in the prompt template to automatically build training data derived from the prompt template itself;
generating a final model by training an adapter of the base model using the automatically-built training data, without reliance on externally labeled topic data to adapt the base model for topic-wise clustering, wherein the adapter comprises a neural network;
input the prompt template to the final model after the adapter training to generate one or more hidden layer vectors highlighting one or more topic-dependent characteristics of the natural language; and
perform a final clustering in latent space by grouping data based on similarities among the one or more hidden layer vectors highlighting the one or more topic-dependent characteristics.
15 . The computer program product of claim 14 , wherein the program instructions are further executable to cause the processor to: utilize the one or more hidden layer vectors as a highlighted feature by the prompt template for the final clustering.
16 . The computer program product of claim 15 , wherein the program instructions are further executable to cause the processor to: output a hidden state that is restored to an image enhanced with visual information.
17 . The computer program product of claim 16 , wherein the program instructions are further executable to cause the processor to: reduce the visual information of the image in accordance with the natural language of the prompt template.
18 . The computer program product of claim 17 , wherein the program instructions are further executable to cause the processor to: enlarge visual information of the image in accordance with the natural language of the prompt template.
19 . The computer program product of claim 14 , wherein the program instructions are further executable to cause the processor to: tune the base model with less than 1% of trainable parameters while freezing a backbone of the base model.
20 . The computer program product of claim 14 , wherein the program instructions are further executable to cause the processor to: dynamically adjust generation of the one or more hidden layer vectors in real time via the final model.