Method and device for automatically classifying digital content into categories suitable for specific domain by using large language

Abstract

A content management method is disclosed. For example, the content management method may comprise the steps in which: a control unit (processor) fine-tunes a language model by using content data collected from a user; the control unit tokenizes the content data by using a tokenizer included in the language model, and generates a PyTorch dataset including training data and test data by using the tokenized data and label-encoded data; the control unit configures weights for respective multiple classes included in the training data by using frequencies of the multiple classes; the control unit trains the language model by using an optimizer and a loss function to which the weights have been applied; and the control unit outputs a category classification result for the test data by using the trained language model.

Description

Method and device for automatically categorizing digital content into specific domains using macrolanguage

[0001] The present invention relates to a content management method, and more particularly, to a technology for outputting a category classification result using content data collected from a user.

[0002] In the information age, the amount of data transmitted between users and platforms is rapidly increasing. In particular, algorithms that provide services through user interaction can categorize user data into specified categories and then provide it back to the user.

[0003] Classification techniques for user data can be applied, for example, to natural language processing (NLP). Based on artificial intelligence technologies like NLP, large amounts of data can be processed at relatively low cost.

[0004] However, as the volume of data being processed continues to grow and demands for more detailed and sophisticated processing results increase, existing natural language processing technologies alone face the challenge of increasing the time and effort required to classify and process data. Furthermore, there are limitations in assessing the reliability of the data being processed, and class imbalance issues can degrade classification performance.

[0005] Furthermore, as the amount of data increases, the amount of unnecessary or unreliable data also increases, which can cause harm to users who demand accurate information.

[0006] Embodiments of the present invention can provide a content management method that trains a language model using content data collected from a user and labeled data (or label encoding data) from an expert or a service providing server, and outputs an accurate classification result (or labeling result) for the content data through the trained language model.

[0007] Embodiments of the present invention can provide a content management method that evaluates the reliability of content data provided by a user, manages the data based on the evaluation result and label type, and dynamically updates the reliability evaluation result based on update information such as content sharing.

[0008] The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0009] According to an embodiment of the present invention, a content management method may include a step in which a control unit (processor) fine-tunes a language model using content data collected from a user, a step in which the control unit tokenizes the content data using a tokenizer included in the language model and generates a PyTorch dataset including training data and test data using the tokenized data and label-encoded data, a step in which the control unit sets weights for each of a plurality of classes included in the training data using the frequencies of the classes, a step in which the control unit trains the language model using a loss function and an optimizer to which the weights are applied, and a step in which the control unit outputs a category classification result for the test data using the trained language model.

[0010] According to one embodiment, the language model may correspond to BERT (Bidirectional Encoder Representations from Transformers) including a plurality of transformer layers, the tokenizer may correspond to WordPiece Tokenizer, the loss function may correspond to cross entropy, and the optimizer may correspond to AdamW.

[0011] According to one embodiment, the plurality of transformer layers may include a multi-head self-attention layer, a feed-forward neural network (FFNN), a residual connection layer, and a layer normalization layer.

[0012] According to one embodiment, the content management method may further include a step in which the control unit labels the content data based on location, business hours, parking availability, and phone number to generate text data, a step in which the control unit fine-tunes the language model based on the text data, a step in which the control unit trains the language model by setting a learning rate to 2e-5, and a step in which the control unit outputs the category classification result based on JSON (JavaScript Object Notation).

[0013] According to one embodiment, the category classification result includes classification information on restaurants, travel, tickets, daily necessities, fashion, miscellaneous goods, food, beauty, and home appliances, and the classification information may include location, price, business hours, rating, purchasing class, and phone number.

[0014] According to one embodiment, the content management method may further include a step of the control unit receiving the integerized label encoding data from an external server that provides a service corresponding to the content data using an interface.

[0015] According to one embodiment, the content management method may further include a step in which the control unit sets a weight for each of the plurality of classes so as to be inversely proportional to the frequency of the class. For example, the content management method may further include a step in which the control unit sets the weight by further utilizing the user's age, height, gender, weight, place of residence, the time at which the content data was collected, the type of service provided by the external server that collected the content data, the location corresponding to the content data, the ratio of the training data to the test data, the amount of the training data, and the number of the plurality of classes.

[0016] According to one embodiment, the content management method may further include a step in which the control unit vectorizes a plurality of paper data including paper data regarding the content data using the language model to generate a plurality of paper vector data, a step in which the control unit stores the plurality of paper vector data in a vector database, and a step in which the control unit outputs a reliability score of the content data based on a comparison result between the content data and the plurality of paper vector data.

[0017] According to one embodiment, the content management method may further include a step in which the control unit crawls an image file corresponding to the content data and extracts text from the crawled image file, a step in which the control unit combines the content data and the text to generate analysis data, a step in which the control unit vectorizes the analysis data using a Retrieval-Augmented Generation (RAG) model to generate content vector data and then stores the vector data in the vector database, a step in which the control unit identifies target paper vector data corresponding to the content vector data among the plurality of paper vector data, a step in which the control unit identifies reliability evaluation criteria including scientific evidence, clinical test results, ingredient information, advertising techniques, and authentication information based on the contents of the target paper vector data, and a step in which the control unit compares the content vector data and the target paper vector data to calculate a score for each of the reliability evaluation criteria, a step in which the control unit outputs a final reliability score of the content data based on the score, and a step in which the control unit assigns a first label to the content data based on the final reliability score being less than or equal to a specified score.

[0018] According to one embodiment, the content management method may further include a step in which the control unit calculates a new reliability score for the new content data when new content data regarding the content vector data is collected from the user, a step in which the control unit assigns a second label distinct from the first label to the new content data when the new reliability score exceeds a specified score, and a step in which the control unit trains the language model based on the final reliability score, the new reliability score, and a change history of the content data and the label assigned to the new content data.

[0019] According to one embodiment, the content management method may further include a step in which the control unit outputs the final reliability score, the new reliability score, the first label, or the second label to the user based on JSON (JavaScript Object Notation).

[0020] The effects of the content management method according to embodiments of the present invention are described as follows.

[0021] Embodiments of the present invention can provide an efficient and accurate content management method by quickly and accurately classifying various pieces of information contained in content-related data in the information age overflowing with unnecessary data, and providing content based on the classified results to other users.

[0022] Embodiments of the present invention can provide a content management method that evaluates the reliability of data regarding content in real time using a language model, updates the reliability when the data is updated, and allows users to select optimal data for content use.

[0023] In addition, various effects may be provided, either directly or indirectly, through this document.

[0024] FIG. 1 is a block diagram illustrating components of a content management device according to an embodiment of the present invention.

[0025] Figure 2 is a flowchart of a content management method according to one embodiment of the present invention.

[0026] Figure 3 is a flowchart of a content management method according to one embodiment of the present invention.

[0027] In connection with the description of the drawings, the same or similar reference numerals may be used for identical or similar components.

[0028] Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. When designating components in each drawing, it should be noted that, where possible, identical components will be given the same reference numerals, even if they appear in different drawings. Furthermore, when describing embodiments of the present invention, detailed descriptions of related known structures or functions will be omitted if they are deemed to hinder understanding of the embodiments of the present invention.

[0029] In describing components of embodiments of the present invention, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only intended to distinguish the components from other components, and the nature, order, or sequence of the components are not limited by these terms. Furthermore, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning they have in the context of the relevant technology, and shall not be interpreted in an idealized or overly formal sense unless explicitly defined in this application.

[0030] Hereinafter, embodiments of the present invention will be described in detail with reference to FIGS. 1 to 3.

[0031]

[0032] FIG. 1 is a block diagram illustrating components of a content management device according to an embodiment of the present invention.

[0033] According to one embodiment, the content management device (100) may include an interface (110), a memory (120), and a control unit (130). The configuration of the content management device (100) illustrated in FIG. 1 is exemplary, and embodiments of the present invention are not limited thereto. For example, the content management device (100) may further include components not illustrated in FIG. 1 (e.g., at least one of a communication unit, a display unit, a notification unit, a sensor unit, or any combination thereof).

[0034] According to one embodiment, the interface (110) may support one or more designated protocols that may be used to directly or wirelessly connect the content management device (100) to an external electronic device (e.g., a content provision server). According to one embodiment, the interface (110) may include, for example, a high definition multimedia interface (HDMI), a universal serial bus (USB) interface, an SD card interface, or an audio interface.

[0035] For example, the interface (110) may include a communication unit. The communication unit may, for example, support the establishment of a direct (e.g., wired) communication channel or a wireless communication channel between the content management device (100) and an external electronic device, and the performance of communication through the established communication channel. The communication unit may include one or more communication processors that operate independently from the control unit (130) (e.g., an application processor) and support direct (e.g., wired) communication or wireless communication. According to one embodiment, the communication unit may include a wireless communication module (e.g., a cellular communication module, a short-range wireless communication module, or a global navigation satellite system (GNSS) communication module) or a wired communication module (e.g., a local area network (LAN) communication module, or a power line communication module). Any of these communication modules may communicate with an external electronic device via a first network (e.g., a short-range communication network such as Bluetooth, wireless fidelity (WiFi) direct, or infrared data association (IrDA)) or a second network (e.g., a long-range communication network such as a legacy cellular network, a 5G network, a next-generation communication network, the Internet, or a computer network (e.g., a local area network or a wide area network)). These various types of communication modules may be integrated into a single component (e.g., a single chip) or implemented as multiple separate components (e.g., multiple chips). The wireless communication module may use subscriber information stored in the subscriber identification module (e.g., an international mobile subscriber identity (IMSI)) to identify or authenticate the content management device (100) within a communication network such as the first network or the second network.

[0036] According to one embodiment, the memory (120) may store commands or data. For example, the memory (120) may store one or more instructions that, when executed by the control unit (130), cause the content management device (100) to perform various operations.

[0037] For example, the memory (120) may be implemented as a single chipset with the control unit (130). The control unit (130) may include at least one of a communication processor or a modem.

[0038] For example, the memory (120) can store various information related to the content management device (100). For example, the memory (120) can store information regarding the operation history of the control unit (130). For example, the memory (120) can store the history of category classification results output by the content management device (100), labeling results, content data, update history of the language model, etc.

[0039] For example, the memory (120) may include a plurality of storage devices of different types. For example, the memory (120) may include volatile and / or nonvolatile storage media. For example, the memory (120) may include at least one of random-access memory (RAM), read only memory (ROM), embedded multi-media card (eMMC), or any combination thereof.

[0040] The steps of the method or algorithm described in connection with the embodiments disclosed herein may be implemented directly in hardware, software modules, or a combination of the two executed by the control unit (130). The software modules may reside in a storage medium (i.e., memory (120)) such as RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM.

[0041] For example, the memory (120) is coupled to a control unit (130), and the control unit (130) can read information from a storage medium and write information to the storage medium. Alternatively, the memory (120) may be integrated with the control unit (130). The memory (120) and the control unit (130) may reside within an application-specific integrated circuit (ASIC). The ASIC may reside within a user terminal. Alternatively, the memory (120) and the control unit (130) may reside as separate components within the user terminal.

[0042] According to one embodiment, the control unit (130) may be operatively connected to the memory (120). For example, the control unit (130) may control the operation of the memory (120).

[0043] For example, the control unit (130) may be a central processing unit (CPU) or a designated type of semiconductor device that executes processing for instructions stored in the memory (120).

[0044] For example, the control unit (130) can tune a language model using content data collected from the user.

[0045] For example, the control unit (130) can receive content data collected from a user by an external device (e.g., an external server that provides content corresponding to the content data) using the interface (110).

[0046] For example, the control unit (130) can fine tune a language model using content data.

[0047] For example, the control unit (130) can generate text data by labeling content data based on location, business hours, parking availability, and phone number, and fine-tune a language model based on the generated text data. The text data may include text corresponding to training data among the content data.

[0048] For example, the language model may correspond to BERT (Bidirectional Encoder Representations from Transformers), which includes multiple transformer layers. The multiple transformer layers may include a multi-head self-attention layer, a feed-forward neural network (FFNN), a residual connection layer, and a layer normalization layer.

[0049] For example, the control unit (130) can tokenize content data using a tokenizer included in a language model, and generate a PyTorch dataset including training data and test data using the tokenized data and label-encoded data.

[0050] For example, the control unit (130) can receive label encoding data including label information encoded as an integer by an external device (e.g., an external server providing content corresponding to content data) using the interface (110).

[0051] For example, the control unit (130) can generate a PyTorch dataset using tokenized data and label-encoded data.

[0052] For example, the control unit (130) can divide a PyTorch dataset into training data and test data based on specified criteria. For example, the control unit (130) can divide a PyTorch dataset into training data and test data based on the amount, type, quality, and acquisition time of the data.

[0053] For example, the tokenizer could be WordPiece Tokenizer.

[0054] For example, the control unit (130) can set weights for each of the plurality of classes using the frequencies of the plurality of classes included in the training data.

[0055] For example, the control unit (130) can distinguish and identify each of the multiple classes included in the training data. For example, the control unit (130) can identify the frequency of each of the multiple classes and set a weight inversely proportional to the frequency. That is, the control unit (130) can set a weight for each of the multiple classes inversely proportional to the frequency of the class. The frequency can be determined based on, for example, the amount of data for each of the multiple classes, the acquisition cycle, the proportion of the data relative to the total data, etc.

[0056] For example, the control unit (130) can set weights for each of the multiple classes based on the mathematical expression 1 below.

[0057] [Mathematical Formula 1]

[0058]

[0059] For example, the above W i represents the weight corresponding to the i-th class, n represents the sum of the frequencies of multiple classes, and n i represents the frequency of the ith class, and I i represents the predefined importance of the i-th class, and D irepresents the data capacity of the i-th class, and P i can represent the data acquisition cycle of the i-th class

[0060] For example, the control unit (130) can set weights by further utilizing the user's age, height, gender, weight, place of residence, time at which content data was collected, type of service provided by an external server that collected the content data, location corresponding to the content data, ratio of training data to test data, amount of training data, and number of multiple classes.

[0061] The control unit (130) may, for example, set weights inversely proportional to the user's age. For example, the control unit (130) may set the weight for a class corresponding to a female user to be lower than the weight for a class corresponding to a male user. For example, among the multiple classes included in the content data, the control unit (130) may set the weight for a class obtained from a female user in his 30s to be lower than the weight for content data obtained from a male user in his 30s.

[0062] For example, the control unit (130) can train a language model using a weighted loss function and an optimizer.

[0063] For example, the control unit (130) can apply weights set for each of a plurality of classes to a loss function for training a language model.

[0064] For example, the control unit (130) can train a language model based on a specified learning rate. The specified learning rate may be 2e-5 (or 0.00002).

[0065] For example, the loss function could correspond to cross entropy.

[0066] For example, the optimizer could correspond to AdamW.

[0067] For example, the control unit (130) can output a category classification result for test data using a learned language model.

[0068] For example, the control unit (130) may output a category classification result based on JSON (JavaScript Object Notation). That is, the control unit (130) may output a category classification result in JSON format and provide it to the user. The control unit (130) may display the category classification result, for example, through a display device formed in one area of ​​the content management device (100).

[0069] For example, category classification results may include classification information on restaurants, travel, tickets, household goods, fashion, miscellaneous goods, food, beauty, and home appliances. This classification information may include, for example, location, price, business hours, ratings, purchasing demographics, and phone number.

[0070] According to one embodiment, the content management device (100) can evaluate the reliability of content data collected from users and assign labels, and update the reliability and change the labels as content data continues to be received.

[0071] For example, the control unit (130) can use a language model to vectorize a plurality of paper data including paper data regarding content data to generate a plurality of paper vector data, and store the generated data in a vector database.

[0072] For example, multiple paper data may include paper data related to various contents processed in the content management device (100).

[0073] For example, the control unit (130) can receive or crawl multiple paper data from a site, repository, server, etc. where paper data is stored.

[0074] For example, the control unit (130) can calculate and output a reliability score of the content data based on the comparison result between the content data and a plurality of paper vector data.

[0075] For example, the control unit (130) can crawl image files corresponding to content data in a site, storage, server, memory (120), etc., and extract text from the crawled image files.

[0076] For example, the control unit (130) can generate analysis data by combining content data and text.

[0077] For example, the control unit (130) can generate content vector data by vectorizing analysis data using the RAG (Retrieval-Augmented Generation) model, and then store the generated content vector data in a vector database.

[0078] For example, the control unit (130) can identify target paper vector data corresponding to content vector data among multiple paper vector data.

[0079] For example, the control unit (130) can identify reliability evaluation criteria including scientific evidence, clinical test results, ingredient information, advertising techniques, and authentication information based on the contents of the target paper vector data.

[0080] For example, the control unit (130) can compare content vector data and target paper vector data to calculate a score for each reliability evaluation criterion. For example, the control unit (130) can calculate and output a final reliability score for the content data based on the score. The final reliability score may be, for example, the sum of the scores for each reliability evaluation criterion.

[0081] For example, the control unit (130) may assign a first label to the content data based on a final reliability score being below a specified score. The first label may be, for example, a label indicating that the content data has low reliability or that the content is false.

[0082] For example, when content data regarding content vector data is collected from a user, the control unit (130) can calculate a new reliability score for the new content data. The new reliability score can be calculated based on an algorithm substantially identical to the reliability score calculation algorithm described above.

[0083] For example, if the new confidence score exceeds a specified score, the control unit (130) may assign a second label distinct from the first label to the new content data. For example, when the first new content data is acquired, the control unit (130) may assign the same first label to the content data in response to identifying that the new content data is data regarding the same content as the content corresponding to the content vector data. Thereafter, a new confidence score for the new content data is calculated, and if the new confidence score exceeds the specified score, the control unit (130) may change the first label assigned to the new content data to a second label. In this case, the second label may be, for example, a label indicating that the new content data is highly reliable or that the content is true.

[0084] For example, the control unit (130) can display a reliability score through a display device formed in one area of ​​the content management device (100).

[0085] For example, the control unit (130) can train a language model based on the final confidence score, the new confidence score, and the second label.

[0086] For example, the control unit (130) can train a language model for calculating a reliability score based on a final reliability score corresponding to content data and a new reliability score corresponding to new content data.

[0087] For example, the control unit (130) can identify content corresponding to content data and new content data, and continuously crawl paper data related to the identified content (e.g., restaurants, travel, tickets, daily necessities, fashion, miscellaneous goods, food, beauty, and home appliances, etc.) based on a specified cycle.

[0088] For example, the control unit (130) may train a language model based on the change history of labels assigned to content data and new content data.

[0089] For example, the control unit (130) can output a final confidence score, a new confidence score, a first label, or a second label to the user based on JSON (JavaScript Object Notation).

[0090] For example, the control unit (130) can display the final reliability score, the new reliability score, the first label, or the second label through a display device formed in one area of ​​the content management device (100).

[0091] For example, the control unit (130) may display the first label and the second label by setting the color, size, font, shape, etc. differently. For example, the first label may be displayed in a color with relatively high saturation compared to the second label, displayed in a relatively large size, displayed in a bold font, or displayed in a form with relatively high visibility.

[0092]

[0093] Figure 2 is a flowchart of a content management method according to one embodiment of the present invention.

[0094] According to one embodiment, a content management device (e.g., content management device (100) of FIG. 1) can perform the operations disclosed in FIG. 2. For example, at least some of the components included in the content management device (e.g., interface (110), memory (120), and control unit (130) of FIG. 1) can be configured to perform the operations of FIG. 2.

[0095] In the following embodiments, operations S210 to S250 may be performed sequentially, but are not necessarily performed sequentially. For example, the order of the operations may be changed, and at least two operations may be performed in parallel. Furthermore, any content corresponding to or overlapping with the content described above with respect to FIG. 2 may be briefly described or omitted.

[0096] According to one embodiment, the content management device can fine-tune a language model using content data collected from a user (S210).

[0097] For example, a content management device may receive content data from an external device (e.g., a content delivery server).

[0098] For example, the language model could be BERT.

[0099] According to one embodiment, the content management device can tokenize content data using a tokenizer included in a language model, and generate a PyTorch dataset including training data and test data using the tokenized data and label-encoded data (S220).

[0100] According to one embodiment, the content management device can set weights for each of the plurality of classes using the frequencies of the plurality of classes included in the training data (S230).

[0101] For example, a content management device may set weights for each of a plurality of classes to be inversely proportional to the frequency of the class.

[0102] According to one embodiment, the content management device can learn a language model using a weighted loss function and an optimizer (S240).

[0103] According to one embodiment, the content management device can output a category classification result for test data using a learned language model (S250).

[0104]

[0105] Figure 3 is a flowchart of a content management method according to one embodiment of the present invention.

[0106] According to one embodiment, a content management device (e.g., content management device (100) of FIG. 1) can perform the operations disclosed in FIG. 3. For example, at least some of the components included in the content management device (e.g., interface (110), memory (120), and control unit (130) of FIG. 1) can be configured to perform the operations of FIG. 3.

[0107] In the following embodiments, operations S310 to S320 may be performed sequentially, but are not necessarily performed sequentially. For example, the order of the operations may be changed, and at least two operations may be performed in parallel. Furthermore, any content corresponding to or overlapping with the content described above with respect to FIG. 3 may be briefly described or omitted.

[0108] Additionally, the operations illustrated in FIG. 3 may be performed together with the operations illustrated in FIG. 2. For example, at least some of the operations illustrated in FIG. 3 may be performed between, before, or after the operations illustrated in FIG. 2.

[0109] According to one embodiment, the content management device can use a language model to vectorize a plurality of paper data including paper data regarding content data, thereby generating a plurality of paper vector data and storing the vector data in a database (S310).

[0110] According to one embodiment, the content management device can output a reliability score of the content data based on a comparison result between the content data and a plurality of paper vector data (S320).

[0111] The above description is merely an example of the technical idea of ​​the present invention, and those skilled in the art will appreciate that various modifications and variations can be made without departing from the essential characteristics of the present invention.

[0112] Accordingly, the embodiments disclosed in the present invention are intended to illustrate, rather than limit, the technical concept of the present invention, and the scope of the technical concept of the present invention is not limited by these embodiments. The scope of protection of the present invention should be interpreted by the following claims, and all technical concepts within the scope equivalent thereto should be construed as being included within the scope of the present invention.

Claims

1. In terms of content management methods, A step in which a control unit (processor) fine-tunes a language model using content data collected from a user; A step in which the control unit tokenizes the content data using a tokenizer included in the language model, and generates a PyTorch dataset including training data and test data using the tokenized data and label-encoded data; A step in which the control unit sets a weight for each of the plurality of classes using the frequency of the plurality of classes included in the training data; A step of the control unit training the language model using the weighted loss function and optimizer; and The control unit includes a step of outputting a category classification result for the test data using the learned language model; How to manage content.

2. In claim 1, The above language model corresponds to BERT (Bidirectional Encoder Representations from Transformers) which includes multiple transformer layers, The above tokenizer corresponds to WordPiece Tokenizer, The above loss function corresponds to cross entropy, The above optimizer corresponds to AdamW, The above multiple transformer layers are: Containing a multi-head self-attention layer, a feed-forward neural network (FFNN), a residual connection layer, and a layer normalization layer. How to manage content.

3. In claim 2, The above content management method is, A step in which the control unit generates text data by labeling the content data based on location, business hours, parking availability, and phone number; A step of the control unit fine-tuning the language model based on the text data; The step of the control unit training the language model by setting the learning rate to 2e-5; and The above control unit further includes a step of outputting the category classification result based on JSON (JavaScript Object Notation); The above category classification results are: Includes classified information on restaurants, travel, tickets, daily necessities, fashion, miscellaneous goods, food, beauty, and home appliances. The above classification information includes location, price, business hours, rating, purchasing tier, and phone number. How to manage content.

4. In claim 3, The above content management method is, The control unit further includes a step of receiving the integerized label encoding data from an external server that provides a service corresponding to the content data using an interface; How to manage content.

5. In claim 1, The above content management method is, The control unit further includes a step of setting the weight by further using the user's age, height, gender, weight, place of residence, the time at which the content data was collected, the type of service provided by the external server that collected the content data, the location corresponding to the content data, the ratio of the training data and the test data, the amount of the training data, and the number of the plurality of classes; The control unit further includes a step of setting a weight for each of the plurality of classes based on the following mathematical expression 1 using the frequency of the plurality of classes included in the training data; [Mathematical Formula 1] Above W i represents the weight corresponding to the i-th class, n represents the sum of the frequencies of multiple classes, and n i represents the frequency of the ith class, and I i represents the predefined importance of the i-th class, and D i represents the data capacity of the i-th class, and P i is characterized by representing the data acquisition cycle of the i-th class, How to manage content.

Images