Text information processing method and device, equipment and storage medium

Abstract

The invention discloses a text information processing method and device, equipment and a storage medium. Evaluation text information is obtained; extracting defect text information according to the evaluation text information; and determining target processing information according to the defect text information. The present invention aims to improve the accuracy of determining processing information.

Description

Technical Field

[0001] The present invention relates to the technical field of text information processing, and particularly relates to a text information processing method, apparatus, device, and storage medium. Background Art

[0002] With the improvement of living standards, more and more products or consumption places for people to consume appear, such as cinemas, supermarkets, televisions, and computers. In order to adapt to the market, merchants need to understand the disadvantages of their own products or consumption places and make timely rectification and treatment of these disadvantages. At present, the rectification and treatment of objects such as products and consumption places are mostly based on the initiative of the managers of these objects during operation, resulting in relatively one-sided processing of the target objects and low accuracy of the processing information for the target objects. Summary of the Invention

[0003] Embodiments of the present invention provide a text information processing method, apparatus, device, and storage medium, aiming to improve the accuracy of determining processing information.

[0004] In a first aspect, embodiments of the present invention provide a text information processing method, and the text information processing method includes:

[0005] Obtain evaluation text information;

[0006] Extract defective text information according to the evaluation text information;

[0007] Determine target processing information according to the defective text information.

[0008] Optionally, the extracting defective text information according to the evaluation text information includes:

[0009] Perform encoding processing on the evaluation text information to obtain a first text feature;

[0010] Reconstruct the first text feature to obtain a second text feature;

[0011] Determine the defective text information according to the second text feature.

[0012] Optionally, the reconstructing the first text feature to obtain a second text feature includes:

[0013] Perform associated prediction processing on the first text feature to obtain text association information;

[0014] Reconstruct the first text feature based on soft mask information and the text association information to obtain the second text feature.

[0015] Optionally, the determining the defective text information according to the second text feature includes:

[0016] Process the second text feature based on the attention module to obtain the target attention score corresponding to the second text feature;

[0017] Determine the defective text information based on the target attention score and the second text feature.

[0018] Optionally, the processing of the second text feature based on the attention module to obtain the target attention score corresponding to the second text feature includes:

[0019] Perform a linear projection on the second text feature based on the attention module to obtain the query information, key information, and value information corresponding to the second text feature;

[0020] Determine the input traffic information and output traffic information according to the query information and the key information;

[0021] Determine the target output traffic information based on the input traffic information and the external interaction information, and determine the target input traffic information based on the output traffic information and the external interaction information;

[0022] Determine the target attention score according to the target output traffic information and the target input traffic information.

[0023] Optionally, the determining of the target attention score according to the target output traffic information and the target input traffic information includes:

[0024] Perform a competitive operation based on the target output traffic information to obtain the first operation information;

[0025] Perform an aggregation operation based on the first operation information, the query information, and the key information to obtain the second operation information;

[0026] Perform a distribution operation based on the target input traffic information to obtain the third operation information, and obtain the target attention score according to the second operation information and the third operation information.

[0027] Optionally, the determining of the target processing information according to the defective text information includes:

[0028] Perform convolution processing on the defective text information based on multiple convolutional layers with different granularities to obtain multiple third text features;

[0029] After splicing the third text features, perform pooling processing to obtain the fourth text feature;

[0030] Fuse the defective text information and the fourth text feature, and output the target processing information based on the fusion result.

[0031] Optionally, after determining the target processing information according to the defective text information, it further includes:

[0032] Obtain the processing type of the target processing information;

[0033] If the processing type belongs to the target processing type, based on the target processing information, perform the control operation corresponding to the target processing information on the target object corresponding to the evaluation text information;

[0034] If the processing type does not belong to the target processing type, based on the target processing information, output auxiliary rectification prompt information for the target object corresponding to the evaluation text information.

[0035] In a second aspect, an embodiment of the present invention provides a text information processing device, where the text information processing device includes:

[0036] An acquisition module, configured to acquire evaluation text information for a target object;

[0037] An extraction module, configured to extract defective text information of the target object according to the evaluation text information;

[0038] A determination module, configured to determine target processing information of the target object according to the defective text information.

[0039] Preferably, the extraction module extracts defective text information according to the evaluation text information, including:

[0040] Perform encoding processing on the evaluation text information to obtain a first text feature;

[0041] Reconstruct the first text feature to obtain a second text feature;

[0042] Determine the defective text information according to the second text feature;

[0043] Preferably, the extraction module reconstructs the first text feature to obtain a second text feature, including:

[0044] Perform association prediction processing on the first text feature to obtain text association information;

[0045] Reconstruct the first text feature based on the soft mask information and the text association information to obtain the second text feature;

[0046] Preferably, the extraction module determines the defective text information according to the second text feature, including:

[0047] Process the second text feature based on the attention module to obtain the target attention score corresponding to the second text feature;

[0048] Determine the defective text information based on the target attention score and the second text feature;

[0049] Preferably, the extraction module processes the second text feature based on the attention module to obtain the target attention score corresponding to the second text feature, including:

[0050] Perform a linear projection on the second text feature based on the attention module to obtain query information, key information, and value information corresponding to the second text feature;

[0051] Determine input traffic information and output traffic information according to the query information and the key information;

[0052] Determine target output traffic information based on the input traffic information and external interaction information, and determine target input traffic information based on the output traffic information and the external interaction information;

[0053] Determine the target attention score according to the target output traffic information and the target input traffic information;

[0054] Preferably, the extraction module determines the target attention score according to the target output traffic information and the target input traffic information, including:

[0055] Perform a competition operation based on the target output traffic information to obtain first operation information;

[0056] Perform an aggregation operation based on the first operation information, the query information, and the key information to obtain second operation information;

[0057] Perform a distribution operation based on the target input traffic information to obtain third operation information, and obtain the target attention score according to the second operation information and the third operation information.

[0058] Preferably, the determination module determines target processing information according to the defective text information, including:

[0059] Perform convolution processing on the defective text information based on multiple convolutional layers with different granularities to obtain multiple third text features;

[0060] After splicing the third text features, perform pooling processing to obtain a fourth text feature;

[0061] Fuse the defective text information and the fourth text feature, and output the target processing information based on the fusion result.

[0062] Preferably, after the determination module determines the target processing information according to the defective text information, it further includes:

[0063] Obtain the processing type of the target processing information;

[0064] If the processing type belongs to the target processing type, based on the target processing information, perform the control operation corresponding to the target processing information on the target object corresponding to the evaluation text information;

[0065] If the processing type does not belong to the target processing type, based on the target processing information, output auxiliary rectification prompt information for the target object corresponding to the evaluation text information.

[0066] In a third aspect, an embodiment of the present invention further provides a text information processing device, including a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, the processor is caused to execute the steps of any one of the text information processing methods provided by the embodiments of the present invention.

[0067] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium includes a computer program. When the computer program runs on an electronic device, the computer program is used to cause the electronic device to execute the steps of any one of the text information processing methods provided by the embodiments of the present invention.

[0068] The present invention first obtains evaluation text information; extracts defective text information according to the evaluation text information; and determines target processing information according to the defective text information. By obtaining the evaluation text information for the target object in this way, the evaluation text information can reflect the public's views, so that the defective text information that can accurately reflect the defects can be extracted from it. Furthermore, the target processing information that can improve the defects can be accurately analyzed according to the defective text information, thereby improving the accuracy of confirming the processing information. Description of the Drawings

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those skilled in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0070] Figure 1 It is a flowchart of an embodiment of the text information processing method provided in the embodiment of the present invention;

[0071] Figure 2 It is a schematic flowchart of another embodiment of the text information processing method provided in the embodiments of the present invention;

[0072] Figure 3 It is a schematic flowchart of yet another embodiment of the text information processing method provided in the embodiments of the present invention;

[0073] Figure 4 It is a first schematic structural diagram of an evaluation text information analysis model provided in the embodiments of the present invention;

[0074] Figure 5 It is a second schematic structural diagram of an evaluation text information analysis model provided in the embodiments of the present invention;

[0075] Figure 6 It is a third schematic structural diagram of an evaluation text information analysis model provided in the embodiments of the present invention;

[0076] Figure 7 It is a fourth schematic structural diagram of an evaluation text information analysis model provided in the embodiments of the present invention;

[0077] Figure 8 It is a schematic structural diagram of a text information processing device provided in the embodiments of the present invention;

[0078] Figure 9 It is a schematic structural diagram of a text information processing device provided in the embodiments of the present invention. Detailed implementation manners

[0079] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention. At the same time, in the description of the embodiments of the present invention, terms such as "first" and "second" are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more features. In the description of the embodiments of the present invention, the meaning of "a plurality" is two or more, unless otherwise specifically defined.

[0080] The embodiments of the present invention provide a text information processing method, device, text information processing device, and computer-readable storage medium.

[0081] Specifically, this embodiment will be described from the perspective of a text information processing device, which can be specifically integrated in a text information processing device. That is, the text information processing method of this embodiment of the present invention can be executed by the text information processing device. Optionally, the text information processing device can be a terminal device such as a mobile phone or a computer, or a server.

[0082] The following will be described in detail with reference to the accompanying drawings. In this embodiment, the execution subject is a text information processing device as an example. It should be noted that the description order of the following embodiments does not limit the preferred order of the embodiments. Although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from that shown in the drawings.

[0083] Currently, the rectification and treatment of objects such as products and consumption places are mostly based on the active discovery by the managers of these objects during the operation process, resulting in relatively one-sided treatment of the target objects and inaccurate treatment information obtained for the target objects.

[0084] To solve the above problems, the present invention discloses a text information processing method. Please refer to Figure 1 The specific process of this text information processing method can be as follows in steps S10 to S40, where:

[0085] Step S10, obtain evaluation text information;

[0086] In this embodiment, the evaluation text information is a text evaluation published for the target object to be processed. The target object is the object to be processed, and these objects can be objects such as products, software, and consumption places that can be used or processed. An evaluation system can be provided for the target object, and the group that can contact or use the target object can publish evaluations of the target object through the evaluation system. For example, users and customers of the target object. The form of the evaluation of the target object is not limited. Among them, it includes evaluating the target object in text form, and then evaluation text information for the target object can be obtained. The evaluation text information contains the views of users on the target object, and also includes information that can characterize the defects of the target object, such as "the food is not delicious", "the environment is noisy", "dirty and messy", etc. Therefore, it is necessary to extract defect text information that can characterize the defects of the target object from the evaluation text information.

[0087] Step S20, extract defect text information according to the evaluation text information;

[0088] In this embodiment, the evaluation text information can reflect the public's views on the target object, including the description information of the defects of the target object. By analyzing the evaluation text information through a model, the description information of the defects of the target object is extracted, and then the defect text information is obtained based on the description information of the defects of the target object. If there are multiple pieces of evaluation text information, the description information of the defects of the target object in each piece of evaluation text information can be summarized, and the description information of the same defect can be classified into one category to sort out non-repetitive defect text information. Optionally, according to the number of description information corresponding to the same defect in the defect text information, a small number of views can be filtered out. For example, the defect text information can be arranged according to the number of description information corresponding to the same defect, and the defect text with a ranking number greater than the preset number can be deleted, or the defect text information with the number of description information corresponding to the same defect less than the preset threshold can be deleted.

[0089] Optionally, the target evaluation text information is screened out from the evaluation text information, and the defect text information is extracted based on the target evaluation text information. The evaluation text information is the view information on the target object, and there are different opinions on the target object in the evaluation text information, that is, the amount of description information about defects in the evaluation text information is different. The greater the amount of description information about defects in the evaluation text information, the more beneficial it is to extract the defect text information. Therefore, the target evaluation text information can be screened out from the evaluation text information according to the amount of description information about defects in the evaluation text information, and the evaluation text information is updated based on the target evaluation text information, and the non-target evaluation text information in the evaluation text information is deleted to extract the defect text information, that is, the evaluation text information used to extract the defect text information can be the screened target evaluation text information, so as to extract the defect text information more accurately and efficiently.

[0090] Optionally, the amount of descriptive information about the defect contained in the evaluation text information can be determined according to the type of the evaluation text information. For example, the evaluation can be a good review, a medium review, or a bad review. Among them, the bad review can most reflect the defects of the target object, and the amount of descriptive information about the defect contained in the bad review is greater than that in the good review or the medium review. Therefore, based on the bad review, it is easier, more accurate, and more efficient to obtain the defect text information of the target object. Thus, after obtaining the evaluation text information about the target object, according to the type of the evaluation text information, the amount of descriptive information about the defect contained in the evaluation text information can be determined, and then the evaluation text information that does not meet the requirements of the descriptive information about the defect, such as other than the bad review, etc., can be filtered out, and the evaluation text information that meets the requirements of the descriptive information about the defect, such as the bad review, etc., can be used as the target evaluation text information. The type of the evaluation text information can be determined by other evaluation information associated with the evaluation text information. For example, the score information received synchronously during user evaluation, such as star selection: "five-star good review", score input: "9.8 points", etc., or the emotion expression information, such as "like OR dislike", the emoticons, emoji, etc. other than the evaluation text information in the evaluation input box.

[0091] Optionally, the sentiment analysis model can also be used to perform sentiment analysis on the evaluation text information to determine the sentiment tendency information corresponding to the evaluation text information. The sentiment tendency information can characterize the sentiment tendency according to the evaluation text information, that is, it can characterize the praise or criticism of the evaluation text information for the target object. Furthermore, the amount of descriptive information about the defect contained in the evaluation text information can be determined according to the sentiment tendency information. If the evaluation of the evaluation text information for the target object is more positive, the amount of descriptive information about the defect contained in the evaluation text information is less. On the contrary, if the evaluation of the evaluation text information for the target object is more negative, the amount of descriptive information about the defect contained in the evaluation text information is more.

[0092] If the amount of descriptive information about the defect in the evaluation text information is less than the preset amount of information, the evaluation text information will be deleted. If the amount of descriptive information about the defect in the evaluation text information is greater than or equal to the preset amount of information, the evaluation text information will be used as the target evaluation text information.

[0093] In some embodiments, the evaluation information can also be directly obtained according to the type of the evaluation to first screen the evaluation information, and then the evaluation text information can be obtained from the evaluation information. For example, obtaining the bad review information, processing the bad review information, and extracting the defect text information, etc. Similar methods will not be elaborated in this embodiment.

[0094] Step S30, determine the target processing information of the target object according to the defect text information.

[0095] In this embodiment, the defective text information can characterize problems, defects, etc. existing in the target object. According to the defective text information, the target processing information of the target object can be correspondingly determined. For the target object, if there are defects, there will be corresponding processing methods. The defective text information can reflect the defects of the target object, and the processing method for the target object can be correspondingly described as the target processing information. Furthermore, an association can be established between the defective text information and the target processing information, and this association can be integrated into a preset comparison table, a preset comparison function, or a preset target processing information generation model. If accurate defective text information is obtained based on the evaluation text information, the accurate target processing information for the target object can be determined according to the above association. Based on the processing method described by the target processing information, the defects of the target object described by the defective text information can be rectified.

[0096] Further, after step S30, it further includes:

[0097] Obtain the processing type of the target processing information;

[0098] If the processing type belongs to the target processing type, based on the target processing information, perform the control operation corresponding to the target processing information on the target object corresponding to the evaluation text information;

[0099] If the processing type does not belong to the target processing type, based on the target processing information, output auxiliary rectification prompt information for the target object corresponding to the evaluation text information.

[0100] In this embodiment, to determine whether the processing type of the target processing information belongs to the target processing type, the target processing type can be the target processing type of the processing operation corresponding to the target processing information that can control the target object through software. For example, for the problem that the lights in the cinema are too bright, the corresponding target processing information is to reduce the light brightness, and the light can be controlled by software to reduce the brightness, so this target processing information belongs to the target processing type; while for the problem that the lights in the cinema are broken, the corresponding target processing information is to replace the lamp tube, and this processing operation currently needs to be implemented manually and cannot be achieved through software control, so this target processing information does not belong to the target processing type. If the processing type belongs to the target processing type, the control operation corresponding to the target processing information can be performed on the target object corresponding to the evaluation text information according to the target processing information, and the target object can be automatically processed to achieve the effect of automatically rectifying the target object. If the processing type does not belong to the target processing type, the auxiliary rectification prompt information for the target object corresponding to the evaluation text information can be output according to the target processing information to notify the management staff to process and rectify the target object according to the auxiliary rectification prompt information, thereby improving the processing efficiency for the target object and facilitating the rectification of the target object.

[0101] In the technical solution disclosed in this embodiment, evaluation text information is obtained; defect text information is extracted according to the evaluation text information; and target processing information is determined according to the defect text information. In this way, the evaluation text information for the target object is obtained, and the evaluation text information can reflect the public's views, so that the defect text information that can accurately reflect the defects can be extracted from it. Furthermore, the target processing information that can improve the defects can be accurately analyzed according to the defect text information, thereby improving the accuracy of confirming the processing information.

[0102] Optionally, referring to Figure 2 , based on any of the above embodiments, in another embodiment of the text information processing method of the present invention, the step S20 includes:

[0103] Step S21: Process the evaluation text information into first text features based on an encoding module;

[0104] In this embodiment, the evaluation text information can be processed based on an evaluation text information analysis model. Referring to Figure 4 , to extract the defect text information of the target object. The evaluation text information analysis model includes an encoding module and a soft mask layer.

[0105] After the evaluation text information is input into the evaluation text information analysis model, the evaluation text information can be preprocessed first. The preprocessing can include word segmentation processing and splicing processing. After that, the vocabulary types in the evaluation text information can be distinguished first, and non-entity vocabulary can be screened out. If there are multiple pieces of evaluation text information, the evaluation text information can also be spliced to obtain a text sequence, which can be expressed as: Among them, [CLS] is the marker at the beginning of the sequence corresponding to the evaluation text information in the text sequence, and [SEP] is the separator between the sequences corresponding to each piece of evaluation text information in the text sequence.

[0106] The processed text sequence is input into the encoding module. The soft encoding module can include a semantic encoding layer (Embedding Layer), and the input text is vectorized using pre-trained word embeddings and position encodings to obtain a number of first text features. Each text feature corresponds to a part of the text in the evaluation text information, such as a character or a word. The first text feature can be expressed as h t .

[0107] Step S22: Reconstruct the first text features to obtain second text features;

[0108] In this embodiment, after the encoding module, the first text feature in vector representation is obtained, which is more convenient for processing the first text feature. The first text feature corresponds to part of the information in the evaluation text information, and the amount of information related to the defective text information carried in different first text features varies. Therefore, it is necessary to reconstruct the first text feature to enhance the presence representation of the first text feature with a large amount of information, or reduce the presence representation of the first text feature with a small amount of information, so as to more accurately analyze the defective text information.

[0109] Step S23: Determine the defective text information according to the second text feature.

[0110] In this embodiment, after the reconstruction process, the first text feature with less irrelevant information carried will be weakened to obtain the second text feature. The first text feature with more relevant information carried will obtain the second text feature after the reconstruction process, and the relevant information it carries will be more obvious. Further, based on the second text feature for continued processing, the propagation of irrelevant information in the model can be reduced, and the propagation of relevant information in the model can be highlighted. Furthermore, through the processing of the evaluation text information analysis model, more accurate defective text information can be obtained.

[0111] Among the technical features disclosed in this embodiment, the evaluation text information is encoded to obtain the first text feature; the first text feature is reconstructed to obtain the second text feature; and the defective text information is determined according to the second text feature. In this embodiment, based on the encoding module, the evaluation text information is processed into the first text feature. After the first text feature undergoes the reconstruction process, it is possible to obtain more accurate defective text information mainly based on the second text feature with richer relevant information, and then more accurate target processing information of the target object can be determined, thereby further improving the accuracy of processing the target object.

[0112] Further, perform an association prediction process on the first text feature to obtain text association information;

[0113] Reconstruct the first text feature based on the soft mask information and the text association information to obtain the second text feature.

[0114] In this embodiment, the evaluation text information analysis model may further include a reconstruction module, and the evaluation text information analysis model is reconstructed based on the reconstruction module. The reconstruction module may include a prediction network. Based on the prediction network, correlation prediction processing is performed on the first text feature, and the correlation degree between the first text feature and the defective text information can be predicted. The correlation degree can be used as text correlation information and is represented by the prediction probability output by the prediction network. The correlation degree can characterize the amount of relevant information carried in the first text feature. The greater the correlation degree, the more relevant information there is, and reconstruction processing is required to enhance such first text features; conversely, the less relevant information there is, and the more irrelevant information there is, and reconstruction processing is required to weaken such first text features.

[0115] Specifically, target key information can be set for the defective text information to be obtained. The defective text information is information that can characterize the defects of the target object. Therefore, the target key information can also be related to the target object, that is, the target key information can be keywords set based on the defects to be identified for the target object. For example, when the target object is a cinema and the cinema needs to be rectified, in some evaluation text information about the cinema, it contains both movie reviews and evaluations of the cinema. Obviously, the movie review part will affect the results of the model. Therefore, keywords such as "seats", "screen", "service", etc. can be set as the target key information. On the contrary, when the target object is a movie and the movie needs to be rectified, keywords such as "plot", "actor", "character", etc. can be set as the target key information.

[0116] The correlation probability between the first text feature and the target key information is predicted through the prediction network. For example, the correlation probability that the text information corresponding to the first text feature belongs to the target key information is predicted to characterize the correlation degree between the first text feature and the defective text information to be determined. This correlation probability can be used as text correlation information. The prediction network can be a two-layer feedforward neural network (FNN) with the activation function GLEU. The prediction network includes a softmax layer, and the correlation probability that the corresponding text information belongs to the target key information is calculated through the softmax function t of the corresponding text information belonging to the target key information.

[0117] The first text feature is input into the above prediction network, and the correlation probability between the first text feature and the target key information is predicted. The specific formula can be as follows:

[0118] p t = softmax(FNN(h t ))

[0119] where p t is the correlation probability between the input information h t and the target key information.

[0120] After obtaining the correlation probability between the first text feature and the target key information, the first text feature with a high correlation probability with the target key information contains more relevant information and less irrelevant information, and has a higher value for determining the defective text information of the target object. The first text feature with a high correlation probability with the target key information can be enhanced. The first text feature with a low correlation probability with the target key information contains more irrelevant information and less relevant information, and has a low value for determining the defective text information of the target object. The first text feature with a high correlation probability with the target key information can be weakened. Based on this, the first text feature is reconstructed to obtain a second text feature with a higher value for determining the defective text information.

[0121] Specifically, the reconstruction module further includes a soft mask layer. The preset soft mask information in the soft mask layer is obtained. The soft mask information can be a preset soft mask vector. Based on the soft mask information, the first text feature with a low correlation probability with the target key information is masked, and conversely, the first text feature with a high correlation probability with the target key information is retained, so as to realize the adjustment of the first text feature. This adjustment can be a scaling process. Specifically, it can be implemented according to the following formula:

[0122] h s = h t × p t + v m × (1 - p t )

[0123] where v m is the soft mask information, and the soft mask information can be the soft mask vector [0, 0, 0,..., 0]. p t is the correlation probability corresponding to the first text feature h t . The larger the predicted correlation probability p t between the first text feature and the target key information, the closer the adjusted first text feature is to the original first text feature h t ; conversely, the smaller p t , the closer the scaled first text feature is to the soft mask information v m . In this way, the first text feature with irrelevant information is masked, thereby weakening the first text feature with a large amount of irrelevant information and a small amount of relevant information, and reducing error propagation.

[0124] In the technical solution disclosed in this embodiment, after performing correlation prediction processing on the first text feature, the text correlation information corresponding to the first text feature is obtained. The text correlation information can represent the relevant information amount of the first text feature. Then, the soft mask information and the text correlation information are used to reconstruct the first text feature to obtain a second text feature with better relevant information amount, thereby improving the accuracy of determining the defective text information.

[0125] Further, determining the defective text information according to the second text feature includes:

[0126] Processing the second text feature based on an attention module to obtain a target attention score corresponding to the second text feature;

[0127] Determining the defective text information based on the target attention score and the second text feature.

[0128] In this embodiment, the evaluation text information analysis model may further include an attention module im-Transformer. Refer to Figure 5 , this attention module can be connected after the soft mask layer. After the soft mask layer processes the second text feature, it outputs the processed result to the attention module. After the attention module processes the second text feature, it can obtain the target attention score of the second text feature. The target attention score can pay more attention to important second text features. Based on the target attention score, linear processing is performed on the second text feature to increase the weight of the more worthy-of-attention second text feature and reduce the weight of the second text feature that is not worthy of attention. The linearly processed second text feature can determine more accurate defective text information, and further can determine more accurate target processing information of the target object, thereby further improving the accuracy of processing the target object.

[0129] Further, processing the second text feature based on the attention module to obtain a target attention score corresponding to the second text feature includes:

[0130] Performing linear projection on the second text feature based on the attention module to obtain query information, key information, and value information corresponding to the second text feature;

[0131] Determining input traffic information and output traffic information according to the query information and the key information;

[0132] Determining target output traffic information based on the input traffic information and external interaction information, and determining target input traffic information based on the output traffic information and the external interaction information;

[0133] Determining the target attention score according to the target output traffic information and the target input traffic information.

[0134] In this embodiment, the basic structure of the attention module can be a Transformer block with the basic structure of GPT-2. Specifically, the core part of the attention module can be composed of 48 improved Transformers, and each layer contains 48 attention heads. Among them, the traditional self-attention mechanism can be replaced by a streaming attention mechanism to further adjust the second text feature and capture more distal interactions between long sequence features.

[0135] When there are multiple evaluation text information or the number of text characters in the evaluation text information is large, the number of second text features will also be relatively large. Furthermore, the data volume of the second text features in the evaluation text information analysis model belongs to a long sequence. For an input of length n, the computational complexity of the self-attention mechanism is O(n 2 ), so the self-attention mechanism is not suitable for processing long sequences. Therefore, the attention module in this embodiment adopts the following streaming attention mechanism, which can reduce the computational complexity of long sequences to O(n), thereby effectively accelerating the speed of model training and reducing the model complexity. From the perspective of a flow network, this attention mechanism introduces flow conservation into the calculation of the attention scores of each second text feature, so as to capture more distal interactions between long sequence features and better extract global semantic information.

[0136] Specifically, referring to Figure 6 , based on the attention module, a linear projection is performed on the second text feature to obtain the query information Qurey (Q), key information Key (K), and value information VaLue (V) corresponding to the second text feature. These information can be represented by vectors. Among them, Q = W q X, K = W k X, V = W v X, W q , W k and W v are all weight matrices.

[0137] Furthermore, the query information Qurey (Q) and key information Key (K) can also be normalized for subsequent calculations. The processing method can be to perform a non-negative projection on the vectors Q and K based on the following formula:

[0138]

[0139]

[0140] In the attention module, a functional relationship can be established between the query information Query (Q) and the key information Key (K). The output result of the above function is input into the softmax for operation to calculate the weights. The values (V) are weighted using the above weights to obtain the attention scores of the second feature information. Inside the attention module, the query information Query (Q) and the key information Key (K) have information interaction with the value information Value (V). The nodes where the query information Query (Q) and the key information Key (K) are located are the source ends, while the node where the value information Value (V) is located is the sink end, and the information flow direction is from the source end to the sink end.

[0141] According to the conservation principle, the input information flow is equal to the output information flow. When the information flow side is restricted, a competition or allocation mechanism is introduced. First, based on the query information and the key information, determine the input traffic information corresponding to the sink end of the second text feature inside the attention module and the output traffic information corresponding to the source end of the second text feature inside the attention module. Among them, the input traffic information of the sink end is And the output traffic information of the source end is

[0142] According to the conservation principle, the output traffic information is equal to the input traffic information. Then, the amount of information exchanged between the attention module and the external network, that is, the external interaction information, can be set to 1 to form an inflow conservation. In this way, based on the input traffic information and the external interaction information of the attention module, the target output traffic information of each source end can be determined. The target output traffic information O ′ Can be obtained based on the following formula:

[0143]

[0144] Similarly, based on the output traffic information and the external interaction information of the attention module, determine the target input traffic information of the sink end. The target output traffic information I ′ Can be obtained based on the following formula:

[0145]

[0146] Since the external interaction information of the attention module is fixed, it will inevitably cause competition. Based on the competition mechanism and the allocation mechanism, process the target output traffic information and the target input traffic information from the sink end to the source end to determine the target attention scores corresponding to the second text feature.

[0147] Furthermore, determining the target attention scores according to the target output traffic information and the target input traffic information includes:

[0148] Perform a competition operation based on the target output traffic information to obtain the first operation information;

[0149] Perform an aggregation operation based on the first operation information, the query information, and the key information to obtain second operation information;

[0150] Perform a distribution operation based on the target input traffic information to obtain third operation information, and obtain the target attention score according to the second operation information and the third operation information.

[0151] In this embodiment, after determining the target input traffic information of the sink end and the target output traffic information of the source end corresponding to the second text feature in the attention module, a competition operation, an aggregation operation, and a distribution operation are introduced in the attention module. Refer to Figure 6 . Specifically, perform a competition operation on the target output traffic information of the source end in the attention module to obtain first operation information. The specific formula is as follows:

[0152] V ′ = softmax(O ′ ) × V

[0153] where softmax() is a function with competitive ability, and V ′ is the first operation information.

[0154] After the competition operation, perform an aggregation operation on the first operation information, the query information, and the key information to obtain second operation information. The specific formula is as follows:

[0155]

[0156] where A is the second operation information obtained after the aggregation operation.

[0157] After the aggregation operation, perform a distribution operation on the target input traffic information of the sink end in the attention module to obtain third operation information. Obtain the target attention score according to the third operation information and the second operation information. The specific formula is as follows:

[0158] R = sigmoid(I ′ ) × A

[0159] where sigmoid() is a function with distribution ability, and R is the target attention score.

[0160] When determining the target output traffic information and the target input traffic information based on the external interaction information between the attention module and other modules in the model, it is equivalent to fixing the resources of the information transmission channel, enhancing the internal competition of the attention module, and thus paying more attention to the important second text features. Such an attention mechanism is more suitable for long sequences and large models, and can obtain a target attention score that more accurately focuses on the second text features. Linear processing is performed on the second text features based on the target attention score, and global features can be obtained based on the result of the linear processing of the second text features. The defective text information is determined according to all the global features. For example, all the global features are used as the defective text information, or text information that can describe the defects of the target object is generated based on the global features. Thereby, the accuracy of the defective text information can be improved.

[0161] Optionally, referring to Figure 3 , based on any of the above embodiments, in another embodiment of the text information processing method of the present invention, after step S30, the method further includes:

[0162] S31. Respectively perform convolution processing on the defective text information based on multiple convolutional layers with different granularities to obtain multiple third text features;

[0163] In this embodiment, the evaluation text information analysis model may further include multiple convolutional layers with different granularities (CNN, 12*12, CNN, 24*24, and CNN, 48*48) and a pooling layer, and the pooling layer may be an adaptive maxpooling layer. Referring to Figure 7 . After performing layer normalization (Layer Norm) processing on the defective text information, it is respectively input into multiple convolutional layers with different granularities, and the multiple convolutional layers with different granularities respectively perform convolution processing on the defective text information to obtain multiple third text features. Due to the different convolutional granularities, the amount of information contained in them is different, so there are different degrees or different dimensions of local information amounts among the extracted third text features.

[0164] S32. After splicing the third text features, perform pooling processing to obtain a fourth text feature;

[0165] In this embodiment, the third text features output by multiple convolutional layers with different granularities are spliced and input into the pooling layer for pooling processing to obtain a local feature with more diverse local information amounts as the fourth text feature. The pooling layer can also automatically calculate the size and stride of the pooling window according to the input information, that is, the feature after splicing the third text features, and can avoid problems such as over-compression and information loss.

[0166] S33. Fuse the defective text information and the fourth text feature, and output the target processing information based on the fusion result.

[0167] In this embodiment, the defective text information is the input information of the module composed of the convolutional layer and the pooling layer, and the fourth text feature is the output information of the module composed of the convolutional layer and the pooling layer. Based on the defective text information and the fourth text feature, residual connection can be performed, that is, the defective text information and the fourth text feature are fused, so as to fuse the global information amount in the defective text information with the local information amount in the fourth text feature, obtain richer semantic information, and thus more high-quality target processing information can be queried or generated.

[0168] In this embodiment, the defective text information is respectively subjected to convolutional processing based on multiple convolutional layers with different granularities to obtain multiple third text features; after the third text features are concatenated, pooling processing is performed to obtain a fourth text feature; the defective text information and the fourth text feature are fused, and the target processing information is output based on the fusion result. By using multiple convolutional layers and pooling layers with different granularities to enrich the transmitted local information amount, a fourth text feature is obtained, which is fused with the defective text information with global information amount to obtain richer semantic information describing the defects of the target object, so that more accurate and high-quality target processing information can be found, further improving the accuracy of the processing information for the target object.

[0169] This embodiment also provides a text information processing device, which can be specifically integrated in a text information processing device. For example, as Figure 8 shown, the text information processing device may include:

[0170] An acquisition module 1001, configured to acquire evaluation text information for a target object;

[0171] An extraction module 1002, configured to extract defective text information of the target object according to the evaluation text information;

[0172] A determination module 1003, configured to determine target processing information of the target object according to the defective text information.

[0173] Preferably, the extraction module 1002 extracts defective text information according to the evaluation text information, including:

[0174] Performing encoding processing on the evaluation text information to obtain a first text feature;

[0175] Performing reconstruction on the first text feature to obtain a second text feature;

[0176] Determining the defective text information according to the second text feature;

[0177] Preferably, the extraction module 1002 reconstructs the first text feature to obtain a second text feature, including:

[0178] Performing an association prediction process on the first text feature to obtain text association information;

[0179] Reconstructing the first text feature based on the soft mask information and the text association information to obtain the second text feature;

[0180] Preferably, the extraction module 1002 determines the defective text information according to the second text feature, including:

[0181] Processing the second text feature based on the attention module to obtain a target attention score corresponding to the second text feature;

[0182] Determining the defective text information based on the target attention score and the second text feature;

[0183] Preferably, the extraction module 1002 processes the second text feature based on the attention module to obtain a target attention score corresponding to the second text feature, including:

[0184] Performing a linear projection on the second text feature based on the attention module to obtain query information, key information, and value information corresponding to the second text feature;

[0185] Determining input traffic information and output traffic information according to the query information and the key information;

[0186] Determining target output traffic information based on the input traffic information and external interaction information, and determining target input traffic information based on the output traffic information and the external interaction information;

[0187] Determining the target attention score according to the target output traffic information and the target input traffic information;

[0188] Preferably, the extraction module 1002 determines the target attention score according to the target output traffic information and the target input traffic information, including:

[0189] Performing a competition operation based on the target output traffic information to obtain first operation information;

[0190] Performing an aggregation operation based on the first operation information, the query information, and the key information to obtain second operation information;

[0191] Performing a distribution operation based on the target input traffic information, and obtaining the target attention score according to the second operation information and the third operation information.

[0192] Preferably, the determining module 1003 determines target processing information according to the defect text information, including:

[0193] Performing convolution processing on the defect text information based on a plurality of convolution layers with different granularities respectively to obtain a plurality of third text features;

[0194] After splicing the third text features, performing pooling processing to obtain a fourth text feature;

[0195] Fusing the defect text information and the fourth text feature, and outputting the target processing information based on the fusion result.

[0196] Preferably, after the determining module 1003 determines the target processing information of the target object according to the defect text information, it further includes:

[0197] Obtaining the processing type of the target processing information;

[0198] If the processing type belongs to the target processing type, then based on the target processing information, performing the control operation corresponding to the target processing information on the target object corresponding to the evaluation text information;

[0199] If the processing type does not belong to the target processing type, then based on the target processing information, outputting auxiliary rectification prompt information for the target object corresponding to the evaluation text information.

[0200] In this embodiment, the evaluation text information is first obtained; according to the evaluation text information, the defect text information is extracted; according to the defect text information, the target processing information is determined. By obtaining the evaluation text information for the target object in this way, the evaluation text information can reflect the public's views, so that the defect text information that can accurately reflect the defects can be extracted from it, and then the target processing information that can improve the defects can be accurately analyzed according to the defect text information, thereby improving the accuracy of confirming the processing information.

[0201] As Figure 9 shown, Figure 9 is a schematic structural diagram of a text information processing device provided by an embodiment of the present invention. The text information processing device 1100 includes a processor 1101 with one or more processing cores, a memory 1102 with one or more computer-readable storage media, and a computer program stored on the memory 1102 and executable on the processor. Among them, the processor 1101 is electrically connected to the memory 1102. Those skilled in the art can understand that the structural diagram of the text information processing device shown in the figure does not constitute a limitation on the text information processing device, and it may include more or fewer components than shown, or combine some components, or arrange different components.

[0202] The processor 1101 is the control center of the text information processing device 1100, connecting various parts of the entire text information processing device 1100 through various interfaces and circuits. By running or loading software programs and / or units stored in the memory 1102, and by invoking the data stored in the memory 1102, it executes various functions of the text information processing device 1100 and processes data, thereby monitoring the text information processing device 1100 as a whole. The processor 1101 can be a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NP), etc., and can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention.

[0203] In the embodiments of the present invention, the processor 1101 in the text information processing device 1100 will, according to the following steps, load the instructions corresponding to the processes of one or more application programs into the memory 1102, and the processor 1101 will run the application programs stored in the memory 1102 to implement various functions, such as:

[0204] Obtain evaluation text information;

[0205] Extract defective text information according to the evaluation text information;

[0206] Determine target processing information according to the defective text information.

[0207] For the specific implementation of each of the above operations, reference can be made to the previous embodiments, which will not be elaborated here.

[0208] Optionally, as Figure 9 shown, the text information processing device 1100 further includes: a touch display screen 1103, a radio frequency circuit 1104, an audio circuit 1105, an input unit 1106, and a power supply 1107. Among them, the processor 1101 is electrically connected to the touch display screen 1103, the radio frequency circuit 1104, the audio circuit 1105, the input unit 1106, and the power supply 1107 respectively. Those skilled in the art can understand that Figure 9 the structure of the text information processing device shown in

[0209] The touch display screen 1103 can be used to display a graphical user interface and receive operation instructions generated by a user acting on the graphical user interface. The touch display screen 1103 may include a display panel and a touch panel. Among them, the display panel can be used to display information input by the user or information provided to the user, as well as various graphical user interfaces of the text information processing device. These graphical user interfaces can be composed of graphics, text, icons, videos, and any combination thereof. Optionally, the display panel can be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. The touch panel can be used to collect touch operations of the user on or near it (such as operations of the user using any suitable object or accessory such as a finger or a stylus on or near the touch panel), determine the corresponding operation instructions, and execute the corresponding program for the operation instructions. Optionally, the touch panel can include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the touch orientation of the user, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into contact coordinates, and then sends it to the processor 1101, and can receive and execute the commands sent by the processor 1101. The touch panel can cover the display panel. After the touch panel detects a touch operation on or near it, it is transmitted to the processor 1101 to determine the type of touch event. Subsequently, the processor 1101 provides a corresponding visual output on the display panel according to the type of touch event. In the embodiment of the present invention, the touch panel and the display panel can be integrated into the touch display screen 1103 to implement input and output functions. However, in some embodiments, the touch panel and the touch panel can be implemented as two independent components to implement input and output functions. That is, the touch display screen 1103 can also be used as a part of the input unit 1106 to implement the input function.

[0210] The radio frequency circuit 1104 can be used to transmit and receive radio frequency signals to establish wireless communication with a network device or other text information processing devices through wireless communication, and transmit and receive signals with the network device or other text information processing devices.

[0211] The audio circuit 1105 can be used to provide an audio interface between the user and the text information processing device through a speaker and a microphone. The audio circuit 1105 can transmit the electrical signal converted from the received audio data to the speaker, which converts it into a sound signal for output; on the other hand, the microphone converts the collected sound signal into an electrical signal, which is received by the audio circuit 1105 and then converted into audio data. After the audio data is output and processed by the processor 1101, it is sent through the radio frequency circuit 1104 to, for example, another text information processing device, or the audio data is output to the memory 1102 for further processing. The audio circuit 1105 may also include an earphone jack to provide communication between the peripheral earphone and the text information processing device.

[0212] The input unit 1106 can be used to receive input digital, character information or the user's first text feature (such as fingerprint, iris, facial information, etc.), and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0213] The power supply 1107 is used to supply power to each component of the text information processing device 1100. Optionally, the power supply 1107 can be logically connected to the processor 1101 through a power management system, so as to implement functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 1107 may also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0214] Although Figure 9 not shown in the figure, the text information processing device 1100 may also include a camera, a sensor, a Wi-Fi module, a Bluetooth module, etc., which will not be elaborated here.

[0215] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0216] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions or by controlling relevant hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0217] For this reason, an embodiment of the present invention provides a computer-readable storage medium, in which multiple computer programs are stored. These computer programs can be loaded by a processor to execute any one of the text information processing methods provided by the embodiments of the present invention. These computer programs can execute the steps of the following text information processing method:

[0218] Obtain the evaluation text information;

[0219] Extract defective text information according to the evaluation text information;

[0220] Determine target processing information according to the defective text information.

[0221] For the specific implementation of each of the above operations, reference may be made to the foregoing embodiments, which will not be elaborated herein.

[0222] Among them, the computer-readable storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc.

[0223] Since the computer program stored in the computer-readable storage medium can execute any text information processing method provided by the embodiments of the present invention, the beneficial effects achievable by any text information processing method provided by the embodiments of the present invention can be realized. For details, reference may be made to the foregoing embodiments, which will not be elaborated herein.

[0224] In the above embodiments of the text information processing device, computer-readable storage medium, text information processing equipment, and computer program product, the descriptions of the respective embodiments have their own focuses. For parts not elaborated in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes and the beneficial effects that can be brought about by the above-described text information processing device, computer-readable storage medium, computer program product, text information processing equipment, and their corresponding units can refer to the description of the text information processing method in the above embodiments, which will not be elaborated herein specifically.

[0225] The above has introduced in detail a text information processing method, a text information processing device, a text information processing equipment, a computer-readable storage medium, and a computer program product provided by the embodiments of the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, based on the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A text information processing method, characterized in that, The described text information processing method includes: Obtain evaluation text information; Extract defective text information according to the evaluation text information; Determine target processing information according to the defective text information.

2. The text information processing method according to claim 1, wherein The extracting defective text information according to the evaluation text information includes: Perform encoding processing on the evaluation text information to obtain a first text feature; Reconstruct the first text feature to obtain a second text feature; Determine the defective text information according to the second text feature.

3. The text information processing method according to claim 2, wherein The reconstructing the first text feature to obtain a second text feature includes: Perform correlation prediction processing on the first text feature to obtain text correlation information; Reconstruct the first text feature based on soft mask information and the text correlation information to obtain the second text feature.

4. The text information processing method according to claim 2, wherein The determining the defective text information according to the second text feature includes: Process the second text feature based on an attention module to obtain a target attention score corresponding to the second text feature; Determine the defective text information based on the target attention score and the second text feature.

5. The text information processing method according to claim 4, wherein The processing the second text feature based on an attention module to obtain a target attention score corresponding to the second text feature includes: Perform linear projection on the second text feature based on the attention module to obtain query information, key information, and value information corresponding to the second text feature; Determine input traffic information and output traffic information according to the query information and the key information; Determine target output traffic information based on the input traffic information and external interaction information, and determine target input traffic information based on the output traffic information and the external interaction information; Determine the target attention score according to the target output traffic information and the target input traffic information.

6. The text information processing method according to claim 5, wherein, The determining the target attention score according to the target output traffic information and the target input traffic information includes: Perform a competition operation based on the target output traffic information to obtain first operation information; Perform an aggregation operation based on the first operation information, the query information, and the key information to obtain second operation information; Perform a distribution operation based on the target input traffic information to obtain third operation information, and obtain the target attention score according to the second operation information and the third operation information.

7. The text information processing method according to claim 1, wherein The determining target processing information according to the defective text information includes: Perform convolution processing on the defective text information based on multiple convolutional layers with different granularities to obtain multiple third text features; After splicing the third text features, perform pooling processing to obtain a fourth text feature; Fuse the defective text information and the fourth text feature, and output the target processing information based on the fusion result.

8. The text information processing method according to claim 1, characterized in that, After determining the target processing information according to the defective text information, it further includes: Obtain the processing type of the target processing information; If the processing type belongs to the target processing type, then perform a control operation corresponding to the target processing information on the target object corresponding to the evaluation text information based on the target processing information; If the processing type does not belong to the target processing type, based on the target processing information, auxiliary rectification prompt information for the target object corresponding to the evaluation text information is output.

9. A text information processing device, characterized in that, The text information processing device includes: An acquisition module, configured to acquire evaluation text information for a target object; An extraction module, configured to extract defective text information of the target object according to the evaluation text information; A determination module, configured to determine target processing information of the target object according to the defective text information; Preferably, the extraction module extracts defective text information according to the evaluation text information, including: Performing encoding processing on the evaluation text information to obtain a first text feature; Reconstructing the first text feature to obtain a second text feature; Determining the defective text information according to the second text feature; Preferably, the extraction module reconstructs the first text feature to obtain a second text feature, including: Performing correlation prediction processing on the first text feature to obtain text correlation information; Reconstructing the first text feature based on soft mask information and the text correlation information to obtain the second text feature; Preferably, the extraction module determines the defective text information according to the second text feature, including: Processing the second text feature based on an attention module to obtain a target attention score corresponding to the second text feature; Determining the defective text information based on the target attention score and the second text feature; Preferably, the extraction module processes the second text feature based on an attention module to obtain a target attention score corresponding to the second text feature, including: Performing linear projection on the second text feature based on the attention module to obtain query information, key information, and value information corresponding to the second text feature; Determining input traffic information and output traffic information according to the query information and the key information; Determining target output traffic information based on the input traffic information and external interaction information, and determining target input traffic information based on the output traffic information and the external interaction information; Determining the target attention score according to the target output traffic information and the target input traffic information; Preferably, the extraction module determines the target attention score according to the target output traffic information and the target input traffic information, including: Performing a competition operation based on the target output traffic information to obtain first operation information; Performing an aggregation operation based on the first operation information, the query information, and the key information to obtain second operation information; Performing a distribution operation based on the target input traffic information to obtain third operation information, and obtaining the target attention score according to the second operation information and the third operation information; Preferably, the determination module determines target processing information according to the defective text information, including: Performing convolution processing on the defective text information based on multiple convolutional layers with different granularities to obtain multiple third text features; After splicing the third text features, performing pooling processing to obtain a fourth text feature; Fuse the defective text information and the fourth text feature, and output the target processing information based on the fusion result; Preferably, after the determination module determines the target processing information according to the defective text information, it further includes: Obtain the processing type of the target processing information; If the processing type belongs to the target processing type, perform the control operation corresponding to the target processing information on the target object corresponding to the evaluation text information based on the target processing information; If the processing type does not belong to the target processing type, output auxiliary rectification prompt information for the target object corresponding to the evaluation text information based on the target processing information.

10. A text information processing device, characterized in that, It includes a processor and a memory, and the memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the text information processing method according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a computer program. When the computer program runs on an electronic device, the computer program is used to cause the electronic device to execute the steps of the text information processing method according to any one of claims 1-8.

Images