A method and apparatus for recognizing a text object

Abstract

The application discloses a text object recognition method and device, which is used to improve the efficiency and accuracy of type identification of objects included in network attack events or dynamic events. The method comprises the following steps: in response to a recognition instruction, extracting a plurality of objects included in the description text of a network event; respectively extracting features of the plurality of objects to determine a feature matrix of each object; wherein each row vector in the feature matrix of any object is used to represent a feature of any object; performing feature re-extraction on each column vector of the feature matrix of any object to generate an intermediate matrix corresponding to each column vector, and performing feature enhancement on each intermediate matrix to generate a target vector corresponding to each intermediate matrix; inputting a target matrix composed of the generated plurality of target vectors into a pre-trained type prediction model to determine the type to which any object belongs.

Description

Technical Field

[0001] This application relates to the field of network security technology, and in particular to a method and apparatus for recognizing text objects. Background Technology

[0002] With the rapid development of internet technology, network security plays a crucial role. Currently, determining whether a dynamic event occurring on the network constitutes an attack typically involves comparing entities within the dynamic event with entities in known attack events, or comparing entities within the dynamic event with security vulnerabilities in a vulnerability database. For example, if a vulnerability database identifies a specific address as an attack source, and the source address in a dynamic event is that address, then the dynamic event can be identified as an attack.

[0003] It is evident that rapidly identifying entities within dynamic and attack events, and determining the types of those entities, is crucial for the security detection of dynamic events. Since the descriptive text of attack or dynamic events is often composed of long and complex sentences, existing technologies propose using rule extraction methods to identify multiple objects within the text and determine their types in order to extract the text objects used to describe the event entities. However, with the development of the internet, the forms of descriptive text have become increasingly diverse. Therefore, using fixed rule extraction methods is not suitable for the current rapid iteration of object type identification in descriptive text. Summary of the Invention

[0004] The exemplary embodiments of this application provide a method and apparatus for recognizing text objects, which can improve the efficiency and accuracy of type recognition of objects included in network attack events or dynamic events.

[0005] Firstly, this application proposes a method for recognizing text objects, including:

[0006] In response to the identification command, extract multiple objects included in the description text of the network event;

[0007] Feature extraction is performed on the multiple objects respectively to determine the feature matrix of each object; wherein each row vector in the feature matrix of any object is used to represent a feature of any object.

[0008] For each column vector of the feature matrix of any object, feature re-extraction is performed to generate an intermediate matrix corresponding to each column vector, and feature enhancement is performed on each intermediate matrix to generate a target vector corresponding to each intermediate matrix;

[0009] The target matrix, composed of multiple generated target vectors, is input into a pre-trained type prediction model to determine the type of any object.

[0010] In some embodiments, the step of re-extracting features from each column vector of the feature matrix of any object to generate an intermediate matrix corresponding to each column vector includes:

[0011] Obtain multiple pre-set convolutional kernels;

[0012] The convolution calculation is performed on any column vector using the multiple convolution kernels in sequence to obtain the intermediate matrix corresponding to any column vector.

[0013] In some embodiments, the step of performing feature enhancement on each intermediate matrix to generate a target vector corresponding to each intermediate matrix includes:

[0014] Using a 1*1 convolution kernel, convolution calculations are performed sequentially on multiple row vectors included in any intermediate matrix;

[0015] The calculation results corresponding to the multiple row vectors are combined to form the target vector corresponding to any intermediate matrix.

[0016] In some embodiments, the step of extracting features from the plurality of objects respectively and determining the feature matrix of each object includes:

[0017] The object is input into a pre-trained feature extraction model, which outputs the feature matrix of the object.

[0018] In some embodiments, the feature extraction model includes a surface feature extraction module, a syntactic feature extraction module, and a semantic feature extraction module; the step of inputting any object into a pre-trained feature extraction model and outputting a feature matrix of any object includes:

[0019] The surface feature extraction module is used to extract the surface features of any object and output the surface feature vector of any object.

[0020] The syntactic feature extraction module is used to extract the syntactic features of any object and output the syntactic feature vector of any object;

[0021] The semantic feature extraction module is used to extract the semantic features of any object and output the semantic feature vector of any object;

[0022] The feature matrix of any object is determined based on the surface feature vector, the syntactic feature vector, and the semantic feature vector.

[0023] Secondly, this application proposes a text object recognition device, the device comprising:

[0024] The feature processing unit is configured to execute:

[0025] In response to the identification command, extract multiple objects included in the description text of the network event;

[0026] Feature extraction is performed on the multiple objects respectively to determine the feature matrix of each object; wherein each row vector in the feature matrix of any object is used to represent a feature of any object.

[0027] For each column vector of the feature matrix of any object, feature re-extraction is performed to generate an intermediate matrix corresponding to each column vector, and feature enhancement is performed on each intermediate matrix to generate a target vector corresponding to each intermediate matrix;

[0028] The type prediction unit is used to input a target matrix composed of multiple generated target vectors into a pre-trained type prediction model to determine the type to which any object belongs.

[0029] In some embodiments, the feature processing unit is specifically used for:

[0030] Obtain multiple pre-set convolutional kernels;

[0031] The convolution calculation is performed on any column vector using the multiple convolution kernels in sequence to obtain the intermediate matrix corresponding to any column vector.

[0032] In some embodiments, the feature processing unit is specifically used for:

[0033] Using a 1*1 convolution kernel, convolution calculations are performed sequentially on multiple row vectors included in any intermediate matrix;

[0034] The calculation results corresponding to the multiple row vectors are combined to form the target vector corresponding to any intermediate matrix.

[0035] In some embodiments, the feature processing unit is specifically used for:

[0036] The object is input into a pre-trained feature extraction model, which outputs the feature matrix of the object.

[0037] In some embodiments, the feature extraction model includes a surface feature extraction module, a syntactic feature extraction module, and a semantic feature extraction module; the feature processing unit is specifically used for:

[0038] The surface feature extraction module is used to extract the surface features of any object and output the surface feature vector of any object.

[0039] The syntactic feature extraction module is used to extract the syntactic features of any object and output the syntactic feature vector of any object;

[0040] The semantic feature extraction module is used to extract the semantic features of any object and output the semantic feature vector of any object;

[0041] The feature matrix of any object is determined based on the surface feature vector, the syntactic feature vector, and the semantic feature vector.

[0042] Thirdly, an electronic device is provided, comprising a controller and a memory. The memory stores computer-executable instructions, and the controller executes the computer-executable instructions in the memory to perform operational steps of any possible implementation of the method of the first aspect using hardware resources in the controller.

[0043] Fourthly, a computer-readable storage medium is provided, which stores instructions that, when executed on a computer, cause the computer to perform the methods described above.

[0044] This application proposes multi-dimensional feature extraction for each object in the descriptive text of network events, obtaining a feature matrix for each object. Based on the characteristic that column vectors in the feature matrix are feature data of the same dimension, features are re-extracted from the column vectors to further enhance the extracted features. Furthermore, to facilitate type prediction, this application proposes a method to aggregate the expanded multi-channel feature data into a single channel, obtaining single-channel feature data, and using this single-channel data for type prediction. This application's solution fully extracts the object's features and predicts the object's type based on these features, effectively improving the accuracy of type identification for network event text objects. In addition, this application applies a deep learning model to the type prediction process, adapting to the current rapidly iterating needs of descriptive text type identification, improving the efficiency of type identification, and thus enabling timely identification and alerting of attack events. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A flowchart illustrating a text object recognition method provided in this application embodiment;

[0047] Figure 2 This is a schematic diagram of a feature extraction process provided in an embodiment of this application;

[0048] Figure 3This is a schematic diagram illustrating a feature re-extraction process provided in an embodiment of this application;

[0049] Figure 4 This is a schematic diagram of the structure of a text object recognition device provided in an embodiment of this application;

[0050] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0052] It should be noted that the terms "first," "second," etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data used can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0053] Threat intelligence, as a component of cybersecurity, can be used within enterprises to predict potential attacks and prevent them. It serves as an early warning system, assists in detection and response during attacks, and is used post-attack for analysis and locating the source of the attack. To ensure threat intelligence fulfills these roles, it is necessary to accurately detect entities included in the threat intelligence of known attack events, as well as entities included in the descriptive text of currently occurring dynamic events. Entity comparison is used to determine whether a current or past dynamic event is an attack. Therefore, accurately and quickly identifying entities from attack events or dynamic events becomes crucial. Entities can include attack methods, regions, attack times, etc.

[0054] Currently, whether identifying entity types from attack events or dynamic events, it's necessary to first identify the corresponding text objects from the event's descriptive text and then determine the category of each object. Since the descriptive text of current network events is mostly long and complex English sentences, related technologies propose matching pre-defined rules with the descriptive text to determine the type of successfully matched text objects. However, with the development of the internet, descriptive text is constantly changing; for example, the types of attacks included in descriptive text are increasing. Therefore, fixed rule-based matching methods are no longer suitable for existing descriptive text.

[0055] In view of this, this application proposes a method and apparatus for identifying text objects. By extracting features from the descriptive text, the type of each object in the descriptive text can be identified, adapting to the current rapidly iterating needs of descriptive text recognition. This application proposes to extract features from the objects included in the descriptive text to obtain a feature matrix for each object. Based on the fact that the column vectors in the feature matrix are feature data of the same dimension, further feature re-extraction is performed on the column vectors to deeply enhance the extracted features, thereby ensuring that the accurate object type can be predicted based on the features.

[0056] Alternatively, the solution in this application can be applied to a variety of scenarios:

[0057] In one possible scenario, the solution proposed in this application can be used to build an attack detection system that can quickly identify various types of objects from known attack events. It stores objects according to their type, allowing users to clearly understand the specific information of the current attack event and thus provide targeted solutions. Optionally, the solution input by the user can also be associated with the types of objects in the attack event and stored, so that a solution can be quickly output when the same or similar attack event occurs again.

[0058] In another possible scenario, the solution presented in this application can also be used to quickly locate and respond to attack events during routine operations and maintenance. Optionally, the solution presented in this application can be used to detect dynamic events occurring in the current network and identify various types of objects within them. Then, based on the identified types, the objects in the current dynamic events can be compared with the objects of known attack events, or the objects included in the current dynamic events can be compared with security vulnerabilities in a vulnerability database. This allows for a rapid assessment of the current threat situation, generation of alerts, and provision of corresponding countermeasures, thereby improving the efficiency of daily operations and maintenance by technical personnel.

[0059] It should be noted that the solution in this application is not limited to the two scenarios mentioned above. The two scenarios are described as examples only and are not intended to limit the solution in this application.

[0060] The following describes the solution of this application in detail, taking into account the two possible implementation scenarios mentioned above. In the embodiments of this application below, "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple. The singular expressions "a," "an," "the," "the," "this," and "this" are intended to also include expressions such as "one or more," unless the context explicitly indicates otherwise. And, unless otherwise stated.

[0061] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0062] See Figure 1 This is a flowchart of a text object recognition method provided in an embodiment of this application. Optionally, Figure 1 The method flow shown can be implemented by a server or server cluster, or it can be executed by any terminal device, such as a personal computer or other electronic device. Alternatively, the method flow can be executed by a computing platform, the computing platform's chip system, or the computing platform's processor. This application is for... Figure 1 The subject executing the method flow shown is not specifically limited. Figure 1 The method flow shown specifically includes:

[0063] 101. In response to the identification command, extract multiple objects included in the description text of the network event.

[0064] Optionally, the identification command can be input by operations and maintenance personnel, or it can be triggered periodically or non-periodically. For example, it can acquire dynamic events occurring in the network once at a set interval, or it can acquire dynamic events in real time for identification. Furthermore, multiple objects included in the description text of network events can be extracted. As an optional method, BIO annotation can be used to extract multiple objects included in the description text.

[0065] 102. Perform feature extraction on multiple objects to determine the feature matrix of each object.

[0066] Optionally, a feature extraction model can be pre-set and trained. Multiple objects are input into the feature extraction model, and the model outputs a feature matrix corresponding to each object. Each row vector in the feature matrix of any object represents a feature of that object. For example, a particular row vector in the feature matrix of any object can represent the semantic feature of that object. Different row vectors can represent different features of that object, and different column vectors can represent feature data of that object in different dimensions.

[0067] 103. Perform feature re-extraction on each column vector of the feature matrix of any object to generate an intermediate matrix corresponding to each column vector.

[0068] This application proposes to extract features from each column vector of the object's feature matrix, based on the data representation of each column vector. This deep extraction of features in each dimension yields an intermediate matrix corresponding to each column vector.

[0069] 104. Perform feature enhancement on each intermediate matrix to generate the target vector corresponding to each intermediate matrix.

[0070] According to step 103 above, feature data of multiple channels can be obtained. Further, multiple channels can be aggregated into a single channel to obtain feature data of the single channel, that is, to generate the target vector corresponding to each intermediate matrix.

[0071] 105. The target matrix composed of multiple generated target vectors is input into a pre-trained type prediction model to determine the type of any object.

[0072] Optionally, this type of prediction model can employ a deep learning model (such as the BilSTM-CRF model). When training the model, the training set can include target matrices of different types of objects and their corresponding type labels. The target matrices of different types of objects are used as the input to the model, and the corresponding type labels are used as the output to train the model.

[0073] Furthermore, when using a type prediction model, the target matrix of any object that needs to be predicted can be input into the model, so that the model can accurately output the type label of that object.

[0074] Based on the above scheme, this application proposes to extract multi-dimensional features from each object in the descriptive text of network events, obtaining a feature matrix for each object. Leveraging the characteristic that column vectors in the feature matrix are feature data of the same dimension, features are re-extracted from the column vectors to further enhance the extracted features. Furthermore, to facilitate type prediction, this application proposes to aggregate the expanded multi-channel feature data to obtain single-channel feature data, using the single-channel data for type prediction. This application's scheme fully extracts the features of the objects and predicts the object type based on these features, effectively improving the accuracy of type identification for network event text objects. In addition, this application applies a deep learning model to the type prediction process, which can adapt to the current rapidly iterating needs of descriptive text type identification, improving the efficiency of type identification and enabling timely detection and alerts for attack events.

[0075] As an optional approach, after obtaining the descriptive text of the network event to be identified, multiple objects can be extracted using BIO annotation. Here, B represents the beginning of an object, I represents the position of an object excluding the beginning, and O represents other objects (such as objects not of interest). Optionally, the descriptive text can also be preprocessed, for example, removing descriptive text whose length does not match the annotation length.

[0076] Furthermore, when extracting features from the extracted objects, a pre-trained feature extraction model (such as the RoBERTa model) can be used to extract multi-dimensional features for each object. Optionally, different network layers in the feature extraction model can be used to extract different features. For example, see [link to relevant documentation]. Figure 2 The feature extraction process shown can be implemented using a feature extraction model that includes a surface feature extraction module, a syntactic feature extraction module, and a semantic feature extraction module. When extracting features from any object, the surface feature extraction module can extract the surface features of the object and output a surface feature vector; the syntactic feature extraction module can extract the syntactic features of the object and output a syntactic feature vector; and the semantic feature extraction module can extract the semantic features of the object and output a semantic feature vector. Furthermore, these three vectors can be combined to form a feature matrix for any given object. In some embodiments, before extracting features from any object using the feature extraction model, the length of the object can be cropped according to the model's input dimensions.

[0077] It should be noted that this application does not limit the order of feature extraction or the types of features extracted. Figure 2 The feature extraction process described herein is merely an example. In one possible implementation, layers 1-4 of the feature extraction model can be used to extract surface features, serving as a surface feature extraction module; layers 5-8 can be used to extract syntactic features, serving as a syntactic feature extraction module; and layers 9-12 can be used to extract semantic features, serving as a semantic feature extraction module.

[0078] In some embodiments, after extracting features from any object to obtain its feature matrix, feature enhancement can be performed on the feature matrix. For example, semantic convolutional neural networks (SCNN) can be used to enhance features. The process of feature enhancement using SCNN is described in detail below. Optionally, features can be re-extracted from each column vector in the feature matrix to obtain multi-channel feature data, and then the multi-channel data can be aggregated into single-channel feature data to achieve feature enhancement. In one possible implementation, when re-extracting features from any column vector of the feature matrix, multiple pre-set convolution kernels can be obtained first. These kernels are then used to perform convolution calculations on the column vector to obtain the intermediate matrix corresponding to that column vector. Different convolution kernels can focus on different feature spaces. For a clearer understanding of the feature re-extraction process, see [link to relevant documentation]. Figure 3 This application exemplifies the process of obtaining an intermediate matrix from a column vector using multi-kernel convolution, using a single column vector as an example. Figure 3 In this context, (a) represents the original column vector. Figure 3 In the diagram, (b) represents the intermediate matrix obtained after expansion, where k is the number of convolution kernels used for convolution calculation.

[0079] Furthermore, to facilitate object type prediction, after re-extracting features to obtain multi-channel feature data, the multi-channel feature data can be further aggregated into single-channel feature data, thereby obtaining the target vector corresponding to each intermediate matrix. In one optional approach, a 1*1 convolution kernel can be used to perform convolution calculations on the multiple row and column vectors included in any intermediate matrix. The calculation results corresponding to each row vector are combined to form the target vector corresponding to that intermediate matrix, thus ensuring that the target vector can fully express the object's features in multiple dimensions.

[0080] Alternatively, multiple generated target vectors can be combined into a target matrix, which can then be input into a trained type prediction model (such as BilSTM-CRF) to output a label representing the type of the object.

[0081] Based on the same concept as the method described above, see [link to relevant documentation]. Figure 4 This application provides a text object recognition device 400, which implements the steps in the above method. To avoid repetition, these steps will not be described again here. The device 400 includes a feature processing unit 401 and a type prediction unit 402.

[0082] Feature processing unit 401 is configured to perform:

[0083] In response to the identification command, extract multiple objects included in the description text of the network event;

[0084] Feature extraction is performed on the multiple objects respectively to determine the feature matrix of each object; wherein each row vector in the feature matrix of any object is used to represent a feature of any object.

[0085] For each column vector of the feature matrix of any object, feature re-extraction is performed to generate an intermediate matrix corresponding to each column vector, and feature enhancement is performed on each intermediate matrix to generate a target vector corresponding to each intermediate matrix;

[0086] The type prediction unit 402 is used to input the target matrix composed of multiple generated target vectors into a pre-trained type prediction model to determine the type to which any object belongs.

[0087] In some embodiments, the feature processing unit 401 is specifically used for:

[0088] Obtain multiple pre-set convolutional kernels;

[0089] The convolution calculation is performed on any column vector using the multiple convolution kernels in sequence to obtain the intermediate matrix corresponding to any column vector.

[0090] In some embodiments, the feature processing unit 401 is specifically used for:

[0091] Using a 1*1 convolution kernel, convolution calculations are performed sequentially on multiple row vectors included in any intermediate matrix;

[0092] The calculation results corresponding to the multiple row vectors are combined to form the target vector corresponding to any intermediate matrix.

[0093] In some embodiments, the feature processing unit 401 is specifically used for:

[0094] The object is input into a pre-trained feature extraction model, which outputs the feature matrix of the object.

[0095] In some embodiments, the feature extraction model includes a surface feature extraction module, a syntactic feature extraction module, and a semantic feature extraction module; the feature processing unit 401 is specifically used for:

[0096] The surface feature extraction module is used to extract the surface features of any object and output the surface feature vector of any object.

[0097] The syntactic feature extraction module is used to extract the syntactic features of any object and output the syntactic feature vector of any object;

[0098] The semantic feature extraction module is used to extract the semantic features of any object and output the semantic feature vector of any object;

[0099] The feature matrix of any object is determined based on the surface feature vector, the syntactic feature vector, and the semantic feature vector.

[0100] Figure 5 A schematic diagram of the structure of an electronic device 500 provided in an embodiment of this application is shown. The electronic device 500 in this embodiment may further include a communication interface 503, such as a network port, through which the electronic device can transmit data.

[0101] In this embodiment, the memory 502 stores instructions that can be executed by at least one controller 501. By executing the instructions stored in the memory 502, the at least one controller 501 can perform various steps in the above-described method. For example, the controller 501 can implement the above-described... Figure 4 The functions of the feature processing unit 401 and the type prediction unit 402 in the process.

[0102] The controller 501 is the control center of the electronic device, capable of connecting various parts of the device via various interfaces and lines. It executes instructions stored in the memory 502 and retrieves data stored in the memory 502. Optionally, the controller 501 may include one or more processing units. The controller 501 may integrate an application controller and a modem controller. The application controller primarily handles the operating system and applications, while the modem controller primarily handles wireless communication. It is understood that the modem controller may not be integrated into the controller 501. In some embodiments, the controller 501 and the memory 502 may be implemented on the same chip; in other embodiments, they may be implemented on separate chips.

[0103] Controller 501 can be a general-purpose controller, such as a central processing unit (CPU), digital signal controller, application-specific integrated circuit, field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose controller can be a microcontroller or any conventional controller. The steps performed by the data statistics platform disclosed in the embodiments of this application can be directly executed by the hardware controller, or executed by a combination of hardware and software modules within the controller.

[0104] Memory 502, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory 502 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic memory, magnetic disk, optical disk, etc. Memory 502 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. In the embodiments of this application, memory 502 can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0105] By designing and programming the controller 501, for example, the code corresponding to the neural network model training method described in the foregoing embodiments can be embedded into the chip, so that the chip can execute the steps of the aforementioned neural network model training method when running. How to design and program the controller 501 is a technique known to those skilled in the art, and will not be described in detail here.

[0106] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a controller of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the controller of the computer or other programmable data processing device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0108] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0109] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0110] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0111] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for recognizing text objects, characterized in that, The method includes: In response to the identification command, extract multiple objects included in the description text of the network event; Feature extraction is performed on the multiple objects respectively to determine the feature matrix of each object; wherein, each row vector in the feature matrix of any object is used to represent one feature of the object, and different column vectors are used to represent feature data of different dimensions of the object. Based on feature data in the feature matrix where the column vectors are of the same dimension, multiple pre-set convolutional kernels are obtained; By sequentially performing convolution calculations on any column vector using the multiple convolution kernels, an intermediate matrix corresponding to any column vector is obtained; Using 1 A convolution kernel of 1 is used to perform convolution calculations on multiple row vectors included in any intermediate matrix in turn; The calculation results corresponding to the multiple row vectors are combined into the target vector corresponding to any intermediate matrix; The target matrix, composed of multiple generated target vectors, is input into a pre-trained type prediction model to determine the type of any object.

2. The method according to claim 1, characterized in that, The step of extracting features from the plurality of objects and determining the feature matrix of each object includes: The object is input into a pre-trained feature extraction model, which outputs the feature matrix of the object.

3. The method according to claim 2, characterized in that, The feature extraction model includes a surface feature extraction module, a syntactic feature extraction module, and a semantic feature extraction module; The step of inputting any object into a pre-trained feature extraction model and outputting the feature matrix of any object includes: The surface feature extraction module is used to extract the surface features of any object and output the surface feature vector of any object. The syntactic feature extraction module is used to extract the syntactic features of any object and output the syntactic feature vector of any object; The semantic feature extraction module is used to extract the semantic features of any object and output the semantic feature vector of any object; The feature matrix of any object is determined based on the surface feature vector, the syntactic feature vector, and the semantic feature vector.

4. A text object recognition device, characterized in that, The device includes: The feature processing unit is configured to execute: In response to the identification command, extract multiple objects included in the description text of the network event; Feature extraction is performed on the multiple objects respectively to determine the feature matrix of each object; wherein, each row vector in the feature matrix of any object is used to represent one feature of the object, and different column vectors are used to represent feature data of different dimensions of the object. Based on feature data in the feature matrix where the column vectors are of the same dimension, multiple pre-set convolutional kernels are obtained; By sequentially performing convolution calculations on any column vector using the multiple convolution kernels, an intermediate matrix corresponding to any column vector is obtained; Using 1 A convolution kernel of 1 is used to perform convolution calculations on multiple row vectors included in any intermediate matrix in turn; The calculation results corresponding to the multiple row vectors are combined into the target vector corresponding to any intermediate matrix; The type prediction unit is used to input a target matrix composed of multiple generated target vectors into a pre-trained type prediction model to determine the type to which any object belongs.

5. The apparatus according to claim 4, characterized in that, The feature processing unit is specifically used for: The object is input into a pre-trained feature extraction model, which outputs the feature matrix of the object.

6. The apparatus according to claim 5, characterized in that, The feature extraction model includes a surface feature extraction module, a syntactic feature extraction module, and a semantic feature extraction module; the feature processing unit is specifically used for: The surface feature extraction module is used to extract the surface features of any object and output the surface feature vector of any object. The syntactic feature extraction module is used to extract the syntactic features of any object and output the syntactic feature vector of any object; The semantic feature extraction module is used to extract the semantic features of any object and output the semantic feature vector of any object; The feature matrix of any object is determined based on the surface feature vector, the syntactic feature vector, and the semantic feature vector.

7. An electronic device, characterized in that, include: Memory and controller; Memory, used to store program instructions; A controller is configured to invoke program instructions stored in the memory and execute the method of any one of claims 1-3 according to the obtained program.

8. A computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are used to perform the method as described in any one of claims 1-3.

Images