Multi-layer micro model analytics framework in information processing system

Patent metadata

Field	Value
Publication number	US-12450535-B2
Application number	US-202217969827-A
Country	US
Kind code	B2
Filing date	Oct 20, 2022
Priority date	Oct 20, 2022
Publication date	Oct 21, 2025
Grant date	Oct 21, 2025

Abstract

Techniques are disclosed for a multi-layer micro model analytics framework for analyzing or otherwise processing data. For example, a method comprises building two or more micro models respectively for two or more stages of a given process, wherein each micro model of the two or more micro models comprises a user interaction layer and a predictive learning layer that coordinate to train the micro model based on input to the user interaction layer and data accessible by the predictive learning layer for the corresponding stage of the two or more stages of the given process. The method then assembles the two or more micro models to perform analysis for the given process. In one non-limiting example, the given process is a new product introduction process and each micro model is built and trained to perform analytics for a specific lifecycle stage of the process.

First claim

What is claimed is:

1. An apparatus, comprising:
a processing platform comprising at least one processor and at least one memory storing computer program instructions wherein, when the at least one processor executes the computer program instructions, the apparatus is configured to:
train a plurality of micro models respectively for each of a plurality of tasks, wherein the plurality of tasks collectively correspond to a stage of two or more stages of a given process, wherein each micro model of the plurality of micro models comprises a user interaction layer and a predictive learning layer that coordinate to train the micro model and a dynamic memory network engine comprising at least an episodic memory model, wherein the user interaction layer is trained based on input to the user interaction layer and wherein the predictive learning layer is trained based on data stored in one or more data sources accessible by the predictive learning layer for the respective task of the plurality of tasks of the corresponding stage of the two or more stages of the given process;
perform analysis on additional input to the user interaction layer using the trained user interaction layer and the trained predictive learning layer of each of the trained plurality of micro models for the respective task of the plurality of tasks for the corresponding stage of the two or more stages of the given process;
generate one or more recommendations and one or more details about the respective task for each of the trained plurality of micro models by inputting an output of the trained user interaction layer and an output of the trained predictive learning layer into the dynamic memory network engine;
perform analysis for the corresponding stage of the two or more stages of the given process by assembling the trained plurality of micro models for each of the plurality of tasks for the corresponding stage of the two or more stages of the given process, wherein at least micro model for at least one task of the plurality of tasks provides input for at least one other micro model for at least one other task of the plurality of tasks for each stage;
generate one or more recommendations and one or more details about the corresponding stage of the two or more stages of the given process based on the analysis of the assembled trained plurality of micro models for each of the plurality of tasks for the corresponding stage of the two or more stages of the given process; and
continuously trigger further training for the predictive learning layer of a given micro model of the plurality of micro models based on a rejection of the one or more recommendations about the respective task of the plurality of tasks of the corresponding stage of the two or more stages of the given process generated by the given micro model.

2. The apparatus of claim 1 , wherein each of the user interaction layer and the predictive learning layer of each micro model comprises one or more machine learning algorithms configured to cooperatively train the micro model.

3. The apparatus of claim 1 , wherein the user interaction layer comprises an intent analyzer configured to determine an intent from the input to the user interaction layer.

4. The apparatus of claim 3 , wherein the intent analyzer is further configured to utilize natural language understanding and a machine learning model to analyze the input and classify the intent.

5. The apparatus of claim 4 , wherein the machine learning model comprises a bi-directional recurrent neural network with long short-term memory.

6. The apparatus of claim 1 , wherein the predictive learning layer comprises a sentiment analyzer configured to perform context-based opinion mining with respect to at least a relevant portion of the data accessible by the predictive learning layer.

7. The apparatus of claim 6 , wherein the sentiment analyzer comprises a recurrent neural network with target dependent long short-term memory.

8. The apparatus of claim 6 , wherein the sentiment analyzer derives a sentiment polarity for the relevant portion of the data.

9. The apparatus of claim 1 , wherein the predictive learning layer comprises a similarity analyzer configured to match one or more features in the data accessible by the predictive learning layer.

10. The apparatus of claim 9 , wherein the similarity analyzer is configured to use a distance metric-based algorithm to perform the matching.

11. The apparatus of claim 10 , wherein the distance metric-based algorithm utilizes a Euclidean distance metric.

12. The apparatus of claim 1 , wherein the predictive learning layer comprises an opportunity analyzer configured to determine one or more potential risks in the given process based on at least a portion of the data accessible by the predictive learning layer.

13. The apparatus of claim 12 , wherein the opportunity analyzer executes one or more statistical programming algorithms to identify the one or more potential risks.

14. The apparatus of claim 1 , wherein the user interaction layer and the predictive learning layer are operatively coupled by the dynamic memory network engine configured to generate a response to the input to the user interaction layer.

15. The apparatus of claim 14 , wherein the dynamic memory network engine computes a representation for the input, performs an iterative attention process on the representation to retrieve relevant data, and reasons over at least a portion of the retrieved relevant data to generate the response.

16. The apparatus of claim 1 , wherein the given process comprises a new product introduction process.

17. A method, comprising:
training a plurality of micro models respectively for each of a plurality of tasks, wherein the plurality of tasks collectively correspond to a stage of two or more stages of a given process, wherein each micro model of the plurality of micro models comprises a user interaction layer and a predictive learning layer that coordinate to train the micro model and a dynamic memory network engine comprising at least an episodic memory model, wherein the user interaction layer is trained based on input to the user interaction layer and wherein the predictive learning layer is trained based on data stored in one or more data sources accessible by the predictive learning layer for the respective task of the plurality of tasks of the corresponding stage of the two or more stages of the given process;
performing analysis on additional input to the user interaction layer using the trained user interaction layer and the trained predictive learning layer of each of the trained plurality of micro models for the respective task of the plurality of tasks for the corresponding stage of the two or more stages of the given process;
generating one or more recommendations and one or more details about the respective task for each of the trained plurality of micro models by inputting an output of the trained user interaction layer and an output of the trained predictive learning layer into the dynamic memory network engine;
performing analysis for the corresponding stage of the two or more stages of the given process by assembling the trained plurality of micro models for each of the plurality of tasks for the corresponding stage of the two or more stages of the given process, wherein at least micro model for at least one task of the plurality of tasks provides input for at least one other micro model for at least one other task of the plurality of tasks for each stage;
generating one or more recommendations and one or more details about the corresponding stage of the two or

Assignees

• Dell Products Lp

Inventors

• Sahoo Sujit Kumar
• Kumar Dhilip S
• Maikhuri Ajay
• Marappa Devaraj

Classifications

• G06F40/40

  Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30) · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06F40/216

  using statistical methods · CPC title

• G06F40/30

  Semantic analysis · CPC title

• G06N3/045

  Combinations of networks · CPC title