Electronic device and method for data classification

Abstract

An electronic device and a method of data classification are provided. The method includes: obtaining class data, wherein the class data includes a data resource; inputting the class data to a reinforcement learning model to update a first weight of a first data resource from a first weight value to a second weight value; performing regression analysis according to the second weight value of the first weight to update the first weight from the second weight value to a third weight value; and outputting the first weight.

Description

[Technical Field]

[0001] This invention relates to a data processing technique, and more particularly to an electronic device and method for data classification. [Previous Technology]

[0002] The future trend for telecom operators is to gradually adopt common application programming interfaces (APIs), such as the CAMA API, to align with the standards of the Global System for Mobile Communications Association (GSMA). However, due to the inconsistent formats and significant quality differences of multi-source data, such as network exposure functions (NEF), operations support systems (OSS), or business support systems (BSS), inconsistencies and inaccuracies arise during API generation. With the increasing demand for real-time data, the system needs a solution that can dynamically adapt to different data sources to improve the accuracy and timeliness of API output and align with the global standardization process in the telecom industry. [Summary of the Invention]

[0003] The present invention provides an electronic device and method for data classification, which can perform data classification and dynamic optimization, ensuring that the system can provide efficient and accurate data processing and automated early warning capabilities while supporting the CAMARA API specification.

[0004] An electronic device for data classification according to the present invention includes a transceiver and a processor. The processor is coupled to the transceiver and configured to perform: acquiring classification data, wherein the classification data includes data sources; inputting the classification data into a reinforcement learning model to update a first weight of a first data source from a first weight value to a second weight value; performing regression analysis based on the second weight value of the first weight to update the first weight from the second weight value to a third weight value; and outputting the first weight through the transceiver.

[0005] In one embodiment of the present invention, the action space of the reinforcement learning algorithm corresponding to the reinforcement learning model includes: weight reduction corresponding to the data source.

[0006] In one embodiment of the present invention, the reward function of the above-mentioned reinforcement learning algorithm is associated with the difference between the value in the classification data and the reference value, wherein the larger the difference, the greater the weight reduction.

[0007] In one embodiment of the present invention, the processor is further configured to perform: obtaining a plurality of historical weight values ​​corresponding to a plurality of historical classification data, wherein the plurality of historical weight values ​​correspond to a first data source; and using a data set containing the plurality of historical weight values ​​and a second weight value as input to a regression analysis to generate a regression model, wherein the regression model generates estimated weight values ​​corresponding to the classification data and the first data source, wherein the processor updates the first weight from the second weight value to a third weight value based on the estimated weight values.

[0008] In one embodiment of the present invention, the processor is further configured to perform: detecting whether the classification data is abnormal; and in response to the abnormality of the classification data, restoring the first weight from the third weight value to the first weight value.

[0009] In one embodiment of the present invention, the processor is further configured to perform: in response to normal classification data, output classification data via transceiver.

[0010] In one embodiment of the present invention, the processor is further configured to perform: receiving offline data via a transceiver; performing a k-average algorithm on the offline data to classify the offline data into one of a plurality of data sources; and, in response to the offline data being classified into a first data source, inputting the offline data as classification data into a reinforcement learning model to update the first weight of the first data source.

[0011] In one embodiment of the present invention, the processor is further configured to perform: in response to offline data being classified to a first data source, inputting offline data into a transformer-based bidirectional encoder representation model to classify the offline data into one of a plurality of subcategories; and in response to offline data being classified to a first subcategory matching user needs, inputting offline data as classification data into a reinforcement learning model.

[0012] In one embodiment of the present invention, the processor is further configured to perform: receiving real-time data via a transceiver; detecting whether the real-time data is abnormal; and in response to the real-time data being abnormal, inputting classification data into a reinforcement learning model to update the first weight of the first data source.

[0013] In one embodiment of the present invention, the processor described above is further configured to perform: receiving multiple data from a first data source via a transceiver, wherein the larger the first weight, the greater the number of multiple data; and generating an application interface based on the multiple data.

[0014] A method for data classification according to the present invention includes: obtaining classification data, wherein the classification data includes data sources; inputting the classification data into a reinforcement learning model to update a first weight of a first data source from a first weight value to a second weight value; performing regression analysis based on the second weight value of the first weight to update the first weight from the second weight value to a third weight value; and outputting the first weight.

[0015] Based on the above, the electronic device of the present invention can extract and preprocess data from multiple data sources such as NEF, OSS, or BSS, perform intelligent classification and summarization of data through a large language model (LLM), and support dynamic weight adjustment based on real-time data. The electronic device can not only continuously optimize the accuracy of API generation based on changes in real-time data, but also issue immediate alarms when data anomalies or potential problems are detected, ensuring stable operation of the electronic device. Through AI self-adjustment learning, the electronic device provides efficient and accurate intelligent decision support and problem early warning in complex and ever-changing data environments.

Implementation Method

[0016] This invention proposes a multi-source data intelligent processing and dynamic optimization method based on artificial intelligence and LLM. First, data is collected from different sources, and telecommunications data is preprocessed and classified. Next, LLM is used for model training and inference to ensure accurate classification and summarization of various types of data. Then, an AI adaptive learning mechanism dynamically adjusts the weights of data sources, optimizing the accuracy and reliability of API generation based on actual data changes. When data anomalies or potential problems are detected, the electronic device of this invention can automatically issue an alarm and notify relevant personnel for timely intervention and handling.

[0017] FIG1 illustrates a schematic diagram of an electronic device 10 for data classification according to an embodiment of the present invention. In one embodiment, the electronic device 10 (and its modules) may be deployed on the same hardware platform. Communication between modules of the electronic device 10 may be performed via API. In one embodiment, the electronic device 10 (and its modules) may be deployed on different hardware platforms. The electronic device 10 may perform inter-module communication via IP-based communication protocols. The electronic device 10 may include a processor 110, a storage medium 120, and a transceiver 130.

[0018] The processor 110 is, for example, a central processing unit (CPU), or other programmable general-purpose or special-purpose microcontroller (MCU), microprocessor, digital signal processor (DSP), programmable controller, application-specific integrated circuit (ASIC), graphics processing unit (GPU), image signal processor (ISP), image processing unit (IPU), arithmetic logic unit (ALU), complex programmable logic device (CPLD), field programmable gate array (FPGA), or other similar elements or combinations thereof. The processor 110 may be coupled to storage medium 120 and transceiver 130, and access and execute multiple modules and various applications stored in storage medium 120.

[0019] The storage medium 120 is, for example, any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory, hard disk drive (HDD), solid state drive (SSD), or similar components or combinations thereof, used to store multiple modules or various applications that can be executed by the processor 110. In this embodiment, the storage medium 120 can store multiple modules including a data classification and LLM establishment module 11, an intelligent weight adjustment and optimization module 12, and an intelligent feedback and weight adjustment module 13, the functions of which will be described later.

[0020] Transceiver 130 transmits or receives signals wirelessly or via a wired connection. Transceiver 130 can also perform operations such as low-noise amplification, impedance matching, mixing, up- or down-frequency conversion, filtering, amplification, and similar operations. Processor 110 can communicate with external systems via transceiver 130 to receive or output data.

[0021] The data classification and LLM building module 11 can collect offline or real-time data from multiple data sources (such as NEF, OSS, or BSS), and perform data cleaning, classification, and feature extraction. The data classification and LLM building module 11 can input this data into a reinforcement learning model (e.g., a trained LLM model) for model training, ensuring that the model can accurately classify and summarize data from different sources. Through model pre-training and fine-tuning, the electronic device 10 can cope with diverse data sources, maintaining the model's efficiency and accuracy.

[0022] The intelligent weight adjustment and optimization module 12 can dynamically adjust the weights of the model according to preset rules and AI calculations. The electronic device 10 can analyze the data sources and adjust their weights according to the accuracy and relevance of different data to ensure that the API generated from the collected data can output more accurate and relevant data. At the same time, the intelligent weight adjustment and optimization module 12 can continuously update the rules based on the analysis results of real-time data to ensure that the weight allocation among the data sources is reasonable and optimized.

[0023] The intelligent feedback and weight adjustment module 13 can collect feedback information and incorporate real-time data and feedback information into the weight adjustment reference. The electronic device 10 can self-adjust based on the feedback data to further optimize model performance and data processing accuracy. The intelligent feedback and weight adjustment module 13 can also have an instant alarm function. When the electronic device 10 detects data anomalies, it will automatically issue an alarm and notify the administrator for immediate handling. This feedback loop enables the electronic device 10 to maintain high efficiency and reliability in a dynamically changing environment.

[0024] FIG2 illustrates a flowchart of data classification according to an embodiment of the present invention, wherein the data classification process can be implemented by the electronic device 10 shown in FIG1. ​​First, the data classification and LLM creation module 11 can receive classification data from multiple data sources 200, wherein the classification data can be offline data or real-time data. The classification data can be used to output to an external system so that the external system can perform API creation based on the classification data. The classification data at least records the data source (e.g., NEF, OSS, or BSS). In one embodiment, the classification data may further record information such as phone number, location information, SIM card information, network logs, traffic, or data type.

[0025] In step S201, the data classification and LLM establishment module 11 can perform k-means clustering on the offline data to classify the offline data into one of multiple data sources (e.g., NEF, OSS or BSS).

[0026] In step S202, the data classification and LLM establishment module 11 can input the offline data classified in step S201 into a bidirectional encoder representations from transformers (BERT) model to further classify the offline data into one of multiple subcategories. For example, the BERT model can classify offline data into a specific subcategory based on the traffic (or network logs, data type) of the offline data. All offline data classified into that subcategory has the same or similar traffic.

[0027] In one embodiment, the data classification and LLM establishment module 11 can train a BERT model based on multiple historical classification data.

[0028] If the offline data is classified into a subcategory that matches the user's needs (e.g., the user needs data with a bandwidth between 50MB and 100MB), in step S203, the intelligent weight adjustment and optimization module 12 can input the offline data into the reinforcement learning model to update the weight values ​​corresponding to the data source of the offline data. For example, if the offline data is classified into NEF, the reinforcement learning model can update the weight values ​​corresponding to the NEF weights based on the offline data.

[0029] In one embodiment, the action space of the reinforcement learning algorithm of the reinforcement learning model may include weight reductions corresponding to the data sources. The reward function of the reinforcement learning model may be associated with the difference between the values ​​in the categorical data and a reference value. The larger the difference, the greater the weight reduction. The smaller the difference, the smaller the weight reduction. The reference value may be associated with user requirements or international organization standards.

[0030] For example, suppose a user needs to create an API with categorized data that has a bandwidth of 100MB. The user can configure the reference value to 100MB. If the difference between the bandwidth value in the categorized data sourced from NEF and the reference value of 100MB is extremely small (e.g., the bandwidth field of the categorized data records a value of 99MB), it means that the categorized data is very close to the user's needs. Accordingly, the reinforcement learning model can either not reduce the weight of NEF or reduce the magnitude of the NEF weight reduction based on the function value of the reward function.

[0031] In step S204, after the weights of the data sources corresponding to the offline data are updated, the intelligent weight adjustment and optimization module 12 can perform regression analysis based on the updated weight values ​​to update the weight values ​​of the weights again.

[0032] Specifically, the electronic device 10 can obtain multiple historical weight values ​​corresponding to multiple historical classification data in a manner similar to steps S201 to S203, wherein the multiple historical weight values ​​may correspond to a specific data source (e.g., the data source NEF for offline data). The intelligent weight adjustment and optimization module 12 can use the multiple historical weight values ​​as input to regression analysis to generate a regression model that can be used to estimate unknown weights. After obtaining the updated weight values ​​of the offline data, the intelligent weight adjustment and optimization module 12 can use a data set containing the aforementioned multiple historical weight values ​​and the updated weight values ​​of the offline data as input to the regression model to generate a new regression model. The regression model can be used to generate estimated weight values ​​corresponding to the offline data and its data source. The intelligent weight adjustment and optimization module 12 can update the weights of the data source of the offline data (e.g., the weights of NEF) to these estimated weight values.

[0033] In step S205, the processor 110 can determine whether the offline data is abnormal. For example, the processor 110 can determine whether the offline data conforms to the format, user requirements, or international organization specifications according to preset rules. If the processor 110 determines that the offline data is abnormal, then step S206 is executed. If the processor 110 determines that the offline data is normal, then step S207 is executed.

[0034] In step S206, the processor 110 can restore the weights of the abnormal offline data. Specifically, the processor 110 can restore the weights of the offline data from the estimated weight values ​​obtained in step S204 to the weight values ​​before step S203 that have not yet been updated by the reinforcement learning model. In other words, the processor 110 can cancel the impact of abnormal offline data on the weight adjustment of the data source.

[0035] In step S207, processor 110 may output categorized data (e.g., transmitting offline or real-time data to a database for storage). The output categorized data can be used as material for creating an API. The weight of a data source may be associated with the amount of data. The larger the weight corresponding to a data source, the more data the processor 110 can collect from that data source. Processor 110 may output the data as material for creating an API, or may generate an API based on the data. Conversely, the smaller the weight corresponding to a data source, the less suitable the data from that source is for creating an API. Accordingly, processor 110 may collect fewer data from that data source. Processor 110 may output the data as material for creating an API, or may generate an API based on the data. In one embodiment, processor 110 may output the weight corresponding to the data source through transceiver 130 as a reference when an external system collects data.

[0036] In step S208, the intelligent feedback and weight adjustment module 13 can detect real-time data. In step S209, the processor 110 can determine whether the real-time data is abnormal based on the detection result. If the real-time data is abnormal, step S210 is executed. If the real-time data is normal, step S207 is executed. For example, the processor 110 can detect whether the real-time data is abnormal based on an anomaly detection model (e.g., XGBoost).

[0037] In step S210, the processor 110 can output an alarm message through the transceiver 130 to prompt the management system or relevant personnel to intervene and correct the abnormal data situation.

[0038] In one embodiment, after completing step S210, the processor 110 can further process real-time data in the same way as offline data, that is, begin executing step S201 to process real-time data, thereby changing the weight of the data source of real-time data as abnormal data, so that the function of the electronic device 10 can adapt to the dynamically changing environment. Abnormal offline data can eventually be output to the CAMARA API for application. For example, the SIM swap API can detect SIM card status changes in real time. If there is abnormal swapping behavior, the electronic device 10 can automatically send an alarm to notify users and operators to prevent potential SIM fraud. The device status API can update the device connection status based on real-time data from NEF and OSS, providing accurate device diagnostic results to operators to ensure service stability.

[0039] FIG3 illustrates a flowchart of a method for data classification according to an embodiment of the present invention, wherein the method can be implemented by the electronic device 10 shown in FIG1. ​​In step S301, classification data is obtained, wherein the classification data includes data sources. In step S302, the classification data is input into a reinforcement learning model to update the first weight of the first data source from a first weight value to a second weight value. In step S303, regression analysis is performed based on the second weight value of the first weight to update the first weight from the second weight value to a third weight value. In step S304, the first weight is output.

[0040] In summary, the present invention has the following characteristics and effects: The present invention can collect and process various data from multiple sources (such as NEF, OSS, or BSS) in real time, and use artificial intelligence (AI) and large-scale language model (LLM) for data classification and summarization; The present invention can dynamically adjust the weight of each data source and continuously optimize the accuracy of API generation according to actual data changes, realizing adaptive data processing and decision support; The present invention can issue real-time alarms based on the analysis results of real-time data. When abnormal data is detected, the present invention can immediately notify the management system or relevant personnel for intervention and correction; The present invention continuously uses feedback data and AI technology for model retraining, and can be widely applied to scenarios such as KYC verification, device status management, SIM card exchange, or billing processing, providing efficient and intelligent data processing and problem prediction functions. [Simplified Explanation of the Diagram]

[0041] FIG1 is a schematic diagram of an electronic device for data classification according to an embodiment of the present invention. FIG2 is a flowchart of data classification according to an embodiment of the present invention. FIG3 is a flowchart of a method for data classification according to an embodiment of the present invention.

Claims

1. An electronic device for classifying data, comprising: transceiver; and a processor, coupled to the transceiver, and configured to perform: acquiring classification data, wherein the classification data includes a data source; The classification data is input into a reinforcement learning model to update the first weight of the first data source from a first weight value to a second weight value; regression analysis is performed based on the second weight value of the first weight to update the first weight from the second weight value to a third weight value; the first weight is output through the transceiver; anomalies in the classification data are detected; and in response to anomalies in the classification data, the first weight is restored from the third weight value to the first weight value.

2. The electronic device as claimed in claim 1, wherein the action space corresponding to the reinforcement learning algorithm of the reinforcement learning model includes: The weights corresponding to the data sources are reduced.

3. The electronic device as claimed in claim 2, wherein the reward function of the reinforcement learning algorithm is associated with the difference between a value in the classification data and a reference value, wherein the larger the difference, the greater the weight reduction.

4. The electronic device of claim 1, wherein the processor is further configured to perform: obtaining a plurality of historical weight values ​​corresponding to a plurality of historical categorical data, wherein the plurality of historical weight values ​​correspond to the first data source; and taking a data set including the plurality of historical weight values ​​and the second weight value as input to the regression analysis to generate a regression model, wherein the regression model generates estimated weight values ​​corresponding to the categorical data and the first data source, wherein the processor updates the first weight from the second weight value to the third weight value based on the estimated weight values.

5. The electronic device as claimed in claim 1, wherein the processor is further configured to: output the classification data via the transceiver in response to the classification data being normal.

6. The electronic device of claim 1, wherein the processor is further configured to perform: receiving offline data via the transceiver; performing a k-average algorithm on the offline data to classify the offline data into one of a plurality of data sources; and, in response to the offline data being classified into the first data source, inputting the offline data as the classification data into the reinforcement learning model to update the first weight of the first data source.

7. The electronic device of claim 6, wherein the processor is further configured to perform: in response to the offline data being classified to the first data source, inputting the offline data to a transformer-based bidirectional encoder representation model to classify the offline data into one of a plurality of subcategories; and in response to the offline data being classified to a first subcategory matching user needs, inputting the offline data as the classification data to the reinforcement learning model.

8. The electronic device of claim 1, wherein the processor is further configured to perform: receiving real-time data via the transceiver; detecting whether the real-time data is abnormal; and in response to the real-time data being abnormal, inputting the classification data into the reinforcement learning model to update the first weight of the first data source.

9. The electronic device of claim 1, wherein the processor is further configured to perform: receiving multiple data entries from the first data source via the transceiver, wherein a larger first weight corresponds to a larger number of data entries; and generating an application interface based on the multiple data entries.

10. A method for classifying data, comprising: Obtain categorized data, wherein the categorized data includes data sources; The classification data is input into a reinforcement learning model to update the first weight of the first data source from a first weight value to a second weight value; regression analysis is performed based on the second weight value of the first weight to update the first weight from the second weight value to a third weight value; the first weight is output. Detect whether the classification data is abnormal; and in response to the abnormality of the classification data, restore the first weight from the third weight value to the first weight value.

Images