Architecture and method for providing insights in networks domain

Patent metadata

Field	Value
Publication number	US-11922336-B2
Application number	US-202117384484-A
Country	US
Kind code	B2
Filing date	Jul 23, 2021
Priority date	Oct 4, 2016
Publication date	Mar 5, 2024
Grant date	Mar 5, 2024

Abstract

The invention is ontology-based architecture for performing semantic data mining for insights. The major components of the architecture are illustrated using Network Ontology (NO), an ontology generated for the network domain for delivering improved network management. In addition, algorithms for summarizing performance profiles in the form of rank tables and for extracting insight rules (concrete action plan) from the rank tables are presented. By using this approach, an actionable plan for assisting decision maker can be obtained, as domain knowledge is incorporated in the system. Experimental results on a network dataset show that this model provides an optimal action plan for a network to improve its performance by encoding data-driven rules into the ontology and suggesting changes to its current network configuration.

First claim

We claim:

1. A method for providing insights in a network domain to a decision maker, comprising:
(a) building a wireless network ontology for the network domain by manually extracting semantics from a relational database;
(b) applying a rank table generation algorithm;
(c) inputting the rank tables into an insight rule generation algorithm;
(d) applying a insight rule generation algorithm to generates insight rules;
(e) encoding the insight rules into the ontology;
(f) using a graphic user interface to operate an experiments lookup services;
(g) selecting an experiment to perform;
(h) submitting the chosen experiment;
(i) invoking the reasoner;
(j) displaying at least one inferred recommended configurable setting to the decision maker; and
wherein each network configuration responds to a different experiment.

2. The method of claim 1 , wherein applying a rank table generation algorithm comprises:
(a) the rank table generation algorithm receiving input from the ontology and the decision maker, said input comprising:
i. controllable parameters;
ii. uncontrollable parameters;
iii. performance metrics; and
iv. discretization of the parameters and performance metrics;

(b) the rank table generation algorithm generates one or more rank tables; and
(c) using information in the rank tables to update the ontology.

3. The method of claim 1 , wherein the rank table generation algorithm comprises:
(a) providing an input;
(b) providing an output, comprising a list of rank tables;
(c) providing a list of Combination of Rank Table Names, wherein a Combination Rank Table Names List holds all possible combinations of uncontrollable parameter values;
(d) providing a list of Combination of Each Row in Rank, wherein the Combination of Each Row in Rank list holds all possible combinations of the controllable parameters, with respect to the number of categories for each controllable parameter;
(e) categorizing the performance metrics and controllable parameters of the experiment;
(f) generating a rank value for each experiment;
(g) computing an average of the rank values of all experiments under decision-maker selected controllable parameters of each rank table;
(h) generating values for a Rank column of the rank tables;
(i) generating a rank table; and
(j) adding the rank table to the rank table list.

4. The method of claim 3 , wherein the input comprises:
(a) a network ontology;
(b) a list of uncontrollable parameters;
(c) a set of possible values for each uncontrollable parameter;
(d) a list of controllable parameters;
(e) a number of categories for each controllable parameters; and
(f) a number of categories for each performance metric.

5. The method of claim 3 , wherein categorizing the performance metrics and controllable parameters of the experiment comprises the following steps:
(a) for each Combination of Rank Table Names, retrieve all triples pertaining to the pertinent Combination of Rank Table Names;
(b) for the retrieved triples, for each performance metric of the experiment in the Combination of Rank Table Names, categorize each into bins, wherein the number of bins corresponds to the number of performance metrics;
(c) replacing the original performance metric values with a corresponding category label;
(d) repeating the above categorizing process for each controllable parameter using the number of categories for each controllable parameter;
(e) completing the experiments in each Combination of Rank Table Names; and
(f) updating the ontology.

6. The method of claim 3 , wherein computing the sum of the performance metrics of each experiment to obtain a rank value for each experiment comprises the following steps:
(a) for each Combination of Rank Table Names, retrieve all the triples pertaining to the pertinent Combination of Rank Table Names; and
(b) for the received triples for each experiment compute the sum of all performance metrics to obtain the Rank of each experiment under decision-maker selected combination of controllable parameters that belongs to the rank table.

7. The method of claim 1 , wherein the rank table generation algorithm comprises:
(a) providing as inputs the following:
i. providing a list of Combination of Rank Table Names;
ii. providing a list of Combination of Each Row in Rank;
iii. categorizing the performance metrics and controllable parameters of the experiment;

(b) computing the sum of the performance metrics of each experiment to generate a rank value for each experiment;
(c) computing an average of the rank values of all experiments under decision-maker selected controllable parameters of each rank table;
(d) providing outputs in the following:
i. generating values for a Rank column of the rank tables;
ii. generating a rank table for the combination rank table names, comprising the combination of each row in rank table and its rank; and

(e) adding the rank table to the rank table list.

8. The method of claim 7 , wherein categorizing the performance metrics and controllable parameters of the experiment comprises the following steps:
(a) for each Combination of Rank Table Names, retrieve all triples pertaining to the pertinent Combination of Rank Table Names by performing a SPARQLQUERY test;
(b) for the retrieved triples, for each performance metric of the experiment in the Combination of Rank Table Names, categorize each into bins, wherein the number of bins corresponds to the number of performance metrics;
(c) replacing the original performance metric values with a corresponding category label after discretization;
(d) repeating the above categorizing process for each controllable parameter using the number of categories for each controllable parameter;
(e) completing the experiments in each Combination of Rank Table Names; and
(f) updating the ontology.

9. The method of claim 7 , wherein computing the sum of the performance metrics of each experiment to obtain a rank value for each experiment comprises the following steps:
(a) for each Combination of Rank Table Names, retrieve all the triples pertaining to the pertinent Combination of Rank Table Names; and
(b) for the received triples for each experiment compute the sum of all performance metrics to obtain the Rank of each experiment under decision-maker selected combination of controllable parameters that belongs to the rank table.

10. The method of claim 1 , wherein insight rule generation algorithm comprises:
(a) an input, comprising a list of rank tables;
(b) an output, comprising a list of insight rules;
(c) performing insight extraction for each row in a rank table; and
(d) generating at least one insight rule.

11. The method of claim 10 , wherein performing insight extraction comprises the following steps:
(a) applying the get_insight_rule(rk, ranktablei) function, which returns an optimal row rk′ based on the distances between rk and different combinations of controllable parameters;
(b) construct an insight rule for row rk in rank table by comparing the values in rk and rk′; and
(c) create list of insight rules.

Assignees

• Univ Louisiana At Lafayette

Inventors

• Duggimpudi Maria Bala
• Moursy Abdelhamid
• Ali Elshaimaa
• Raghavan Vijay V

Classifications

• G06N7/01 Primary

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06F3/0484

  for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range · CPC title

• G06F16/24578

  using ranking · CPC title

• G06F16/367

  Ontology · CPC title

• G06N5/013

  Automatic theorem proving · CPC title