Data information classification method and device, computer equipment and storage medium

By extracting feature words from comment text and generating feature vectors by combining contextual semantics and sentence order, and then using a bidirectional neural network model for classification, the problem of low accuracy in sentiment analysis in existing technologies is solved, achieving more accurate sentiment analysis and resource allocation.

CN119004197APending Publication Date: 2024-11-22TCL TECHNOLOGY GROUP CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310582509.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-22
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing data classification methods mainly rely on analyzing relevant evaluation phrases appearing in the comment text individually, resulting in poor accuracy in sentiment analysis and ignoring the correlation between sentences in the comment text, which affects the accuracy of merchant resource allocation.

Method used

By extracting feature word information from the comment text, combining the semantic relationship between the feature words and the sentence order, a target feature vector is generated. Then, a bidirectional neural network model is used to fuse the semantic information, and finally, the vector is input into a trained classification model for sentiment classification.

Benefits of technology

This improves the accuracy of sentiment classification, ensures the accuracy of merchant and user reviews, thereby optimizing platform resource allocation and reducing the impact of malicious reviews.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004197A_ABST
    Figure CN119004197A_ABST
Patent Text Reader

Abstract

The invention provides a data information classification method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining to-be-classified text information of a target object, and extracting feature word information in the to-be-classified text information; determining a target feature vector based on the feature word information, a context semantic association relationship of the feature word information and a statement sequence relationship of the to-be-classified text information; and determining a classification result of the to-be-classified text information based on the target feature vector. According to the embodiment of the invention, the text feature vector fully fusing the semantic information of the context information is obtained based on the feature words related to the evaluation dimension and the context information thereof, and the association relationship between the context sentences, so that the classification of the to-be-classified text information can be completed more accurately.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of text processing technology, and in particular to a data information classification method, device, computer equipment and storage medium. Background Art

[0002] In many merchant platforms, there is usually a user evaluation system to provide improvement directions for merchants to optimize their stores. For example, taking the common food delivery platform as an example, users usually evaluate merchants from evaluation dimensions such as food, delivery, and packaging. Therefore, it is key to classify user comments and extract their emotional tendencies in each evaluation dimension.

[0003] However, current data information classification methods are mainly based on analyzing relevant evaluation phrases appearing in the review text separately, resulting in poor accuracy of the sentiment tendency finally extracted. Summary of the invention

[0004] Based on this, it is necessary to provide a data information classification method, device, computer equipment and storage medium to address the above technical problems, so as to solve the problem that the existing data information classification method has poor classification effect, resulting in poor accuracy of the obtained emotional tendency.

[0005] In a first aspect, the present application provides a data information classification method, comprising:

[0006] Acquire the to-be-classified text information of the target object, and extract the characteristic word information in the to-be-classified text information;

[0007] Determining a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified;

[0008] A classification result of the text information to be classified is determined based on the target feature vector.

[0009] As a feasible embodiment of the present application, determining the target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified includes:

[0010] Generate a sentence feature vector of a target sentence corresponding to the feature word information based on the feature word information and the contextual semantic association relationship of the feature word information;

[0011] Based on the sentence feature vector and the sentence sequence relationship of the text information to be classified, a target feature vector is determined.

[0012] As a feasible embodiment of the present application, the generating of a sentence feature vector of a target sentence corresponding to the feature word information based on the feature word information and the contextual semantic association relationship of the feature word information includes:

[0013] Determining a forward output vector of the feature word information based on the feature word information and the preceding context semantic information in the preceding and following context semantic association relationship of the feature word information;

[0014] Determining a reverse output vector of the feature word information based on the feature word information and contextual semantic information in a contextual semantic association relationship of the feature word information;

[0015] Based on the forward output vector and the reverse output vector, a sentence feature vector of the target sentence corresponding to the feature word information is generated.

[0016] As a feasible embodiment of the present application, determining the forward output vector of the feature word information based on the feature word information and the preceding context semantic information in the preceding and following context semantic association relationship of the feature word information includes:

[0017] Merging the word feature vector of the feature word information with a preset indicator vector to obtain a first input word vector;

[0018] Obtaining a second input word vector corresponding to the preceding context semantic information in the preceding and following context semantic association relationship of the feature word information;

[0019] The first input word vector and the second output word vector are spliced ​​and input into a preset first bidirectional neural network to obtain a forward output vector of the feature word information.

[0020] As a feasible embodiment of the present application, generating a sentence feature vector of a target sentence corresponding to the feature word information based on the forward output vector and the reverse output vector includes:

[0021] Acquire normalized position information of the feature word information in the target sentence corresponding to the feature word information;

[0022] The forward output vector and the reverse output vector are fused based on the weight corresponding to the normalized position information to obtain a sentence feature vector of the target sentence.

[0023] As a feasible embodiment of the present application, determining a target feature vector based on the sentence feature vector and the sentence sequence relationship of the text information to be classified includes:

[0024] Merging the word feature vector of the feature word information in the target sentence with the sentence feature vector of the target sentence to obtain an updated feature vector of the target sentence;

[0025] Inputting the updated feature vector of the target sentence into a preset second bidirectional neural network based on the sentence sequence relationship of the text information to be classified, to obtain a forward output vector of the target sentence and a reverse output vector of the target sentence;

[0026] A target feature vector of the target comment text is generated based on the forward output vector of the target sentence and the reverse output vector of the target sentence.

[0027] As a feasible embodiment of the present application, the determining the classification result of the to-be-classified text information based on the target feature vector includes:

[0028] The target feature vector is input into a pre-trained classification model to obtain a classification result of the text information to be classified, wherein the classification model is trained through the following steps:

[0029] Obtaining a sample feature vector of a sample comment text and a classification label of the sample comment text;

[0030] Inputting the sample feature vector into a preset initial classification model to obtain a predicted classification result of the sample feature vector;

[0031] Calculating a model loss value corresponding to the initial classification model based on the number of each category label in the classification label and the difference between the predicted classification result and the classification label;

[0032] Updating the initial classification model based on the model loss value to obtain an updated classification model;

[0033] Until the updated model loss value obtained by inputting the sample feature vector into the updated classification model meets the preset training conditions, the currently obtained updated classification model is set as the pre-trained classification model.

[0034] As a feasible embodiment of the present application, the step of obtaining the to-be-classified text information of the target object includes:

[0035] Acquire first text information to be classified of a first target object, wherein the first text information to be classified is provided by a second target object;

[0036] After the step of determining the classification result of the text information to be classified based on the target feature vector, the method further includes:

[0037] Based on the classification result of the text information to be classified and the evaluation weight of the second target object, a platform resource allocation result of the first target object is determined, wherein the evaluation weight of the second target object is set based on the classification result of the text information to be classified of the second target object.

[0038] In a second aspect, the present application provides a data information classification device, comprising:

[0039] An acquisition module is used to acquire the to-be-classified text information of the target object and extract the characteristic word information in the to-be-classified text information;

[0040] A generating module, configured to determine a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified;

[0041] A classification module is used to determine a classification result of the text information to be classified based on the target feature vector.

[0042] In a third aspect, the present application further provides a computer device, the computer device comprising:

[0043] one or more processors;

[0044] Memory; and

[0045] One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the processor in accordance with any of the data information classification methods provided above.

[0046] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is loaded by a processor to execute any of the data information classification methods described above.

[0047] The data information classification method provided in the embodiment of the present application first extracts the feature word information related to the evaluation dimension after obtaining the text information to be classified, and then generates a target feature vector of the text information to be classified based on the contextual semantic association relationship of the feature word information and the sentence order relationship of the text information to be classified. The target feature vector fully integrates the semantic information of the feature words related to each evaluation dimension and their contextual information. Therefore, based on the target feature vector, the sentiment analysis of the text information to be classified can be completed more accurately, and a more accurate sentiment tendency classification result can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 A schematic diagram of a step flow of a data information classification method provided in an embodiment of the present application;

[0050] Figure 2 A schematic diagram of a process flow of generating a feature vector of a text to be classified based on a feature word provided in an embodiment of the present application;

[0051] Figure 3 A schematic diagram of a flow chart of steps for extracting a sentence feature vector of a target sentence based on bidirectional extraction provided in an embodiment of the present application;

[0052] Figure 4 A schematic diagram of a process flow of introducing an indicator into a vector representation provided in an embodiment of the present application;

[0053] Figure 5 A schematic flow chart of a step of generating a feature vector of a sentence based on position information of a feature word in a sentence provided in an embodiment of the present application;

[0054] Figure 6 A schematic diagram of a process flow for fusing a word vector representation with a feature vector of a target sentence provided in an embodiment of the present application;

[0055] Figure 7 A schematic diagram of a step flow of training a classification model for classification provided in an embodiment of the present application;

[0056] Figure 8 A schematic diagram of the steps of another data information classification method provided in an embodiment of the present application;

[0057] Fig. 9 A schematic diagram of the structure of a data information classification device provided in an embodiment of the present application;

[0058] Fig.10 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0060] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0061] In the description of the present application, the word "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present application.

[0062] To facilitate understanding of the data information classification method provided by the embodiment of the present application, the application scenario of the data information classification method provided by the embodiment of the present application is first described. Specifically, the data information classification method provided by the embodiment of the present application is usually run on a platform where merchants provide commodity sales services. For example, the more common ones are takeaway platforms, clothing sales platforms, etc. For ease of description, the subsequent description is taken as an example of the takeaway platform. Of course, it is also feasible to apply the data information classification method to other platforms with user evaluation systems. The embodiment of the present application is not repeated here. Among them, the platform where the data information classification method runs includes the comment text of each merchant, which is usually provided by different users and contains the user's emotional tendency to the merchant in each evaluation dimension. For example, taking the takeaway platform as an example, the merchant's evaluation dimensions mainly include food, packaging, delivery, etc., and the user's emotional tendency can be roughly divided into three dimensions: positive, general, and negative. Of course, the evaluation dimensions and emotional tendencies provided by the above embodiments are only a feasible implementation scheme, which can be specifically set based on actual needs.

[0063] However, in general, existing evaluation classification methods are mainly based on the analysis of dimension-related words in the comment text. For example, taking the comment text "delivery too fast, delivery in advance, food is cold" as an example, existing evaluation classification methods usually extract the two evaluation dimension-related words "delivery too fast" and "food is cold" and analyze them separately, thus believing that the user's emotional tendency towards the merchant in the delivery evaluation dimension is positive, but the emotional tendency towards the merchant in the product evaluation dimension is negative, while ignoring the impact of the correlation between the sentences in the comment text on the emotional polarity, resulting in the final classification result being inaccurate, affecting the subsequent allocation of related resources for merchants, such as merchants with more positive user emotional tendencies can get better resource allocation to retain users. In addition, in actual application scenarios, there is still the problem of some users discrediting merchants to damage the rights and interests of merchants.

[0064] Precisely to solve the above-mentioned problems, the embodiments of the present application provide a data information classification method, device, computer equipment and storage medium that can be run on a platform for mutual evaluation between merchants and users. The data information classification method can be used to implement sentiment tendency classification of users' comments on merchants, thereby obtaining merchants' evaluation scores, and can also be used to implement sentiment tendency classification of merchants' comments on users, thereby obtaining users' evaluation scores, and in the subsequent platform resource allocation process, corresponding services are provided to merchants and users based on the evaluation scores to enhance the experience of merchants and users. Specifically, the data information classification method is usually installed in the data information classification device in the form of a computer program, and the data information classification device is provided in the computer device in the form of a processor. The data information classification device in the computer device executes the data information classification method provided in the embodiments of the present application by running the computer program corresponding to the data information classification method. Figure 1 As shown, Figure 1 A schematic flow chart of a data information classification method provided in an embodiment of the present application, specifically, comprising steps S110 to S130:

[0065] S110, obtaining text information to be classified of a target object, and extracting feature word information in the target review text.

[0066] In the embodiments of the present application, based on the description of the relevant application scenarios provided above, it can be known that the target object usually refers to a certain merchant, store or user on the platform. For example, taking a merchant as an example, the text information to be classified at this time is the comment text information of the merchant provided by the platform users within a certain period in the past, such as within a historical month, and when the target object is a user, the text information to be classified at this time is the comment text information of the user provided by the platform merchant within a historical month.

[0067] Furthermore, the characteristic word information in the target review text generally refers to the evaluation dimension related words in the target review text, that is, some words in the target review text related to the preset evaluation dimension. For example, taking merchants as an example, the evaluation dimension related words at this time generally refer to the associated words related to the three evaluation dimensions of food, packaging, and delivery described in the target review text. For users, at this time, the evaluation dimension related words generally refer to the associated words related to the evaluation dimensions such as honesty, punctuality for taking takeout, and communication attitude in the target review text. Of course, the evaluation dimension provided above is only an implementation scheme, which can be set based on actual needs. When the set evaluation dimension related words are saved in a preset dictionary, the data information classification device can query the preset dictionary and extract the corresponding characteristic word information from the target review text.

[0068] In addition, considering that in general, a review text r consists of m sentences, that is, r = (s1, s2, ..., s m ), and each sentence consists of T words Composition, that is, i =(w i1 , w i2 ,…,w it ), therefore, after extracting the corresponding feature word from the target comment text, the target sentence s where the feature word is located i can be split into context sequence s v =(w i1 ,…,w i(l-1) , w i(r+1) ,…,w iT ), and the feature word sequence s τ =(w il ,…,w ir ). It should be noted that, usually, the feature word sequence s τ The evaluation dimension related words of different evaluation dimensions are included in the t-th feature word, which is the evaluation dimension related word w. On this basis, the indicator vector can be further set to distinguish the related words of each dimension for model recognition. For example, assuming there are three dimensions, the t-th feature word, that is, the evaluation dimension related word w it It is related to the second dimension. At this time, 1 can form a new word vector representation w it ′=[w it , 0, 1, 0], where [0, 1, 0] in the word vector representation can be understood as an indicator vector. The value of the dimension in the vector is 1, which indicates which evaluation dimension the corresponding feature word is related to. For example, [1, 0, 0] means that the feature word is related to the first evaluation dimension, and [0, 0, 1] means that the feature word is related to the third evaluation dimension.

[0069] Of course, it should be noted that in order to display the review text on a computer, usually, each word in the text will be converted into a corresponding vector representation. Specifically, there are many forms of word vector representation, such as the more common embedded vector representation, global vector representation, etc. As an optional implementation scheme, the embodiment of the present application uses the global vector representation of the word for processing, that is, the word w iT The corresponding word vector is x iT , at this time, the target sentence s where the feature word is located i The vector representation x i =(x i1 , c i2 , …, x iT ), which includes the vector representation of the feature words And the vector representation of the remaining context

[0070] S120, determining a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified.

[0071] In the embodiment of the present application, after obtaining the feature word information, the context vector representation describing the semantic association relationship between the feature word and its context, and the sentence order relationship in the text to be classified, the data information classification device will further fully consider the semantic relationship between words and sentences, and make full use of the contextual semantic association relationship of the feature word information and the sentence order relationship of the text information to be classified to perform corresponding processing on the vector representation of the feature word, thereby obtaining a target feature vector that integrates the contextual semantic information. The target feature and vector can better reflect the overall semantic information of the classified text. Specifically, as a feasible embodiment of the present application, the specific implementation scheme for determining the target feature vector can be referred to in the subsequent Figure 2 and related contents of its explanation.

[0072] S130: Determine a classification result of the text information to be classified based on the target feature vector.

[0073] In the embodiment of the present application, after the target feature vector of the target comment text is extracted, the classification result of the target comment text can be obtained by inputting the target feature vector into the classification model obtained by pre-training. Specifically, based on the application scenario provided above, it can be known that the classification result here is usually the classification result of emotional tendency. For example, as a feasible implementation scheme, the classification results include positive, pertinent and negative. Of course, it is feasible to set other classification results based on actual needs, such as excellent, good, passing, bad reviews, etc., and this application does not limit this.

[0074] It should be noted that the classification model here usually needs to be trained in advance using training samples, but considering that in the actual application process, due to the imbalance of sample data categories, the model training is difficult to converge. For example, taking the distribution dimension as an example, in the actual application process, the sample data with positive classification results may account for more than 95%, while the sample data with neutral and negative results may account for less than 5%, resulting in poor classification effect of the trained classification model. Therefore, as a feasible embodiment of the present application, an improved model training method is proposed. Specifically, the specific implementation scheme of the classification model training can be referred to in the subsequent Figure 7 and its explanatory contents.

[0075] Furthermore, based on the aforementioned classification results, the evaluation results of the target object on the platform can be further obtained. For example, when the classification result of a comment text is positive, the score of the comment text can be set to 5 points. Conversely, when the classification result of a comment text is negative, the score of the comment text can be set to 0 points. Therefore, by averaging the scores of all comment texts of the target object, the evaluation scores of the target object in each evaluation dimension can be obtained. Specifically, the higher the evaluation score, the better the performance of the corresponding evaluation dimension of the target object, such as punctual delivery and better packaging. Therefore, better resources can be provided to the target object on the platform. For example, when the target object is a merchant, the discount rate of its push and subsidy can be increased, or when the target object is a user, the discount rate of its red envelope can be increased, etc. Of course, the above is only a feasible implementation plan. The specific resource allocation strategy can be set based on the actual needs of the platform, and this application does not limit this.

[0076] In addition, in combination with the application scenarios provided above, it can be known that the data information classification provided by the embodiments of the present application can be run on a platform for mutual evaluation between users and merchants, that is, the merchant's evaluation score can be obtained based on the user's comments, and the user's evaluation score can be obtained based on the merchant's comments. However, considering that in the actual application process, there are problems of malicious comments from some users or merchants. As a feasible embodiment of the present application, after obtaining the sentiment tendency classification result of the corresponding evaluation dimension based on the comment text of the first target object provided by the second target object, the weight will be set based on the evaluation score of the second target object, where the evaluation score of the second target object is obtained by sentiment classification based on the comment text provided by other first target objects. Specifically, the implementation plan can be found in the subsequent Figure 7 and its explanatory contents.

[0077] The data information classification method provided in the embodiment of the present application first extracts the feature word information related to the evaluation dimension after obtaining the text information to be classified, and then generates a target feature vector of the text information to be classified based on the contextual semantic association relationship of the feature word information and the sentence order relationship of the text information to be classified. The target feature vector fully integrates the semantic information of the feature words related to each evaluation dimension and their contextual information. Therefore, based on the target feature vector, the sentiment analysis of the text information to be classified can be completed more accurately, and a more accurate sentiment tendency classification result can be obtained.

[0078] like Figure 2 As shown, Figure 2 A schematic flow chart of steps for generating a feature vector of a text to be classified based on a feature word provided in an embodiment of the present application, specifically, comprising steps S210 to S220:

[0079] S210: Generate a sentence feature vector of a target sentence corresponding to the feature word information based on the feature word information and a contextual semantic association relationship between the feature word information and the ...

[0080] In the embodiment of the present application, based on the above related description, it can be known that the feature word information is a vector The semantic association between the feature word information and the context is represented by vector exists in the form of, and by vector And vector By performing joint processing, the feature words related to the evaluation dimension and their context information can be integrated at the word level, thereby obtaining a feature vector v that integrates the evaluation dimension and context semantics. it Specifically, usually, the above-mentioned method of extracting feature vectors can be obtained by processing in the form of a neural network model. For example, as a common implementation scheme, the data information classification device can use a trained deep neural network (DNN) or a convolutional neural network model (CNN) to represent the vector of the feature word. and vector representation of context information The output vector of the intermediate layer is processed and used as the sentence feature vector of the target sentence.

[0081] It should be noted that different neural network models can be used to extract the feature vector of the target sentence from different angles, but the conventional method of inputting the sentence vector into the network model in a sequential manner can only partially capture the positive semantic information of the sentence. Therefore, in order to more comprehensively capture the semantic information of the sentence, as a feasible embodiment of the present application, a solution for extracting the feature vector of the sentence based on a bidirectional neural network model is proposed. The specific solution can be found in the subsequent Figure 3 and its explanatory contents.

[0082] S230: Determine a target feature vector based on the sentence feature vector and the sentence sequence relationship of the text information to be classified.

[0083] In the embodiment of the present application, the sentence feature vector v of each target sentence is obtained as described above. it Finally, considering that the text to be classified is composed of several sentences and the context sentences are also interrelated, this will also affect the sentiment tendency of the text to be classified in a specific evaluation dimension. Therefore, as a feasible embodiment of the present application, the semantic information can be integrated at the sentence level based on the order relationship of each target sentence in the text to be classified and the feature vector of the target sentence, so as to obtain the target feature vector of the text to be classified. Specifically, similar to the aforementioned step S210, the target feature vector of the text to be classified can also be obtained by processing in the form of a neural network model, such as the aforementioned deep neural network (DNN) or convolutional neural network model (CNN). Of course, in order to more comprehensively capture the semantic information of the text to be classified, as a feasible embodiment of the present application, a sentence-level bidirectional neural network model can be used to extract the target feature vector of the text to be classified. Of course, in order to further reflect the evaluation dimension reflected by each sentence, before processing the sentence feature vector of the target sentence, the word feature vector of the feature word in the target sentence will be fused with the sentence feature vector of the target sentence. For the specific implementation scheme, please refer to the subsequent Figure 6 and its explanatory contents.

[0084] like Figure 3 As shown, Figure 3 A schematic flow chart of a step of extracting a sentence feature vector of a target sentence based on bidirectional extraction is provided in an embodiment of the present application, specifically, comprising steps S310 to S330:

[0085] S310: Determine a forward output vector of the feature word information based on the feature word information and contextual semantic information in a contextual semantic association relationship of the feature word information.

[0086] There are many bidirectional neural network models. For example, as a feasible implementation scheme, the bidirectional neural network model provided in this application is a bidirectional gated recurrent unit (Bi-GRU). Specifically, the bidirectional gated recurrent unit is composed of two gated recurrent units. When the target sentence s i The vector representation x i =(x i1 , x i2 , …, x iT) are sequentially input into the bidirectional gated recurrent model, a gated recurrent model of the bidirectional gated recurrent model will represent the vector x i The forward processing is performed in sequence, thereby using the feature word information and the previous context semantic information of the context semantic association relationship of the feature word information to obtain the forward output vector of the feature word information, wherein the specific processing formula based on the gated loop model is as follows:

[0087]

[0088] Among them, x it That is, the target sentence vector representation x i The tth component in the positive direction can also be understood as the feature word vector input into the gated recurrent model for processing at the tth processing moment. describes the output vector of the hidden layer in the gated recurrent model at time t, It describes the output vector of the hidden layer in the gated recurrent model at time t-1. represents the bidirectional gated recurrent model used for forward processing in the bidirectional gated recurrent model. When the target sentence vector represents x i The tth component x in the positive direction it When it is a feature word, the forward output vector of the feature word is the output of the gated recurrent model.

[0089] S320: Determine a reverse output vector of the feature word information based on the feature word information and contextual semantic information in a contextual semantic association relationship of the feature word information.

[0090] In the embodiment of the present application, in the bidirectional gated recurrent model Bi-GRU, in addition to the aforementioned method for representing the vector x i The gated recurrent model for forward processing also includes the vector representation x i A gated loop model for negative processing, specifically, the specific processing formula of the gated loop model is as follows:

[0091]

[0092] Specifically, for the parameter description of the above formula, reference may be made to the aforementioned step S210, wherein: represents the bidirectional gated recurrent model used for reverse processing in the bidirectional gated recurrent model. When the target sentence vector represents x i The tth component x in the positive direction it When it is a feature word, the reverse output vector of the feature word is the output of the gated recurrent model.

[0093] Of course, it should be noted that no matter whether the vector represents x i The forward processing is still the vector representation x i The reverse processing of , all need to rely on the indication of the feature words, so, in the input vector representation x i In the process of i Update and add corresponding indicators. Specifically, the following will take the forward processing as an example for explanation. For the specific implementation scheme, please refer to the subsequent Figure 4 and its explanatory contents.

[0094] S330: Generate a sentence feature vector of the target sentence corresponding to the feature word information based on the forward output vector and the reverse output vector.

[0095] In the embodiment of the present application, the forward output vector corresponding to the feature word is obtained above And the reverse output vector After that, further and Fusion can obtain the fusion vector of feature words. in The fusion vectors based on each feature word in the target sentence can be further concatenated to obtain the sentence feature vector v of the target sentence. it .

[0096] In addition, considering that the position of the feature word in the sentence is different, the cumulative effect of the words before and after the feature word in the sentence will change its influence on the sentiment tendency. For example, when the feature word is located in the second half of the sentence, the cumulative effect of the semantic information of the preceding context of the feature word exceeds the influence of the semantic information of the following context on the sentiment tendency. In other words, the position of the feature word affects the weight of the context. Therefore, the context position weight conversion can be introduced to complete the fusion of the feature word fusion vector and obtain the sentence feature vector v of the target sentence. it For specific implementation solutions, please refer to the subsequent Figure 5 and its explanatory contents.

[0097] like Figure 4 As shown, Figure 4 A schematic flow chart of a step of introducing an indicator into a vector representation provided in an embodiment of the present application, specifically, comprising steps S410 to S430:

[0098] S410: Merge the word feature vector of the feature word information with a preset indicator vector to obtain a first input word vector.

[0099] In the embodiment of the present application, the vector x is represented by a bidirectional gated loop model. iThe gated recurrent model for forward processing is used as an example to explain. In fact, in the bidirectional gated recurrent model, the vector x is represented i The gated recurrent model for negative processing will also add the indicator vector, that is, in practical applications, the vector representing x i Before inputting into the bidirectional gated recurrent model, the vector x is represented i The corresponding components of are added to the corresponding indicator vector to obtain a new vector representation x i ′, and input into the bidirectional gated recurrent model for processing.

[0100] Specifically, for the vector representation x i The feature vector of the feature word It can be combined with a preset indicator vector Concatenate to get the first input word vector

[0101] S420, obtaining a second input word vector corresponding to the preceding context semantic information in the preceding and following context semantic association relationship of the feature word information;

[0102] In the embodiment of the present application, based on the contextual semantic association relationship of the feature word information, the second input word vector corresponding to the contextual semantic information can be directly obtained. No need to add an extra indicator vector.

[0103] S430, concatenating the first input word vector and the second output word vector and inputting them into a preset first bidirectional neural network to obtain a forward output vector of the feature word information.

[0104] In the embodiment of the present application, based on the first input word vector obtained above and the second input word vector We can further get the target sentence s i The updated vector represents x i ′ =x i ′ 1,…,x i ′ T ), and the updated vector represents x i ′ Input into a bidirectional neural network, such as a bidirectional gated recurrent unit Bi-GRU, then a gated recurrent unit in the bidirectional gated recurrent unit will perform an update on the vector representation x based on the processing formula provided in the aforementioned step 210 i ′ Processing is performed to obtain the forward output vector of the feature word Similarly, another gated recurrent unit in the bidirectional gated recurrent unit will process the updated vector representation x based on the processing formula provided in the aforementioned step 220. i ′ Processing is performed to obtain the reverse output vector of the feature word

[0105] like Figure 5 As shown, Figure 5 A schematic flow chart of steps for generating a sentence feature vector based on position information of a feature word in a sentence provided in an embodiment of the present application, specifically, comprising steps S510 to S520:

[0106] S510, obtaining the normalized position information of the feature word in the target sentence corresponding to the feature word. In the embodiment of the present application, the normalized position information of the feature word can be obtained by To describe, among them, Among them, the vector representation of the target sentence x i The tth component in is the feature word, and K is the length of the target sentence, so,

[0107] S520, fusing the forward output vector and the reverse output vector based on the weight corresponding to the normalized position information to obtain a sentence feature vector of the target sentence.

[0108] In the embodiment of the present application, considering the influence of the normalized position of the feature word on the semantics, a solution is proposed for fusing the forward output vector and the reverse output vector based on the weight corresponding to the normalized position information to obtain the sentence feature vector of the target sentence, wherein, specifically, the calculation formula of the normalized position information and the weight is as follows:

[0109]

[0110] in, is the normalized position information obtained above, K is the position center of the feature word, which can be obtained by calculating the average of the normalized values ​​of the starting position l and the ending position r of the feature word in the target sentence, that is, is the normalized value of the starting position l, is the normalized value of the end position r.

[0111] Based on the weights obtained above, the sentence feature vector v of the target sentence can be obtained by further combining the forward output vector and the reverse output vector. it , the specific calculation formula is as follows:

[0112]

[0113] like Figure 6 As shown, Figure 6 A schematic flow chart of a step of fusing a word vector representation with a feature vector of a target sentence provided in an embodiment of the present application, specifically, comprising steps S610 to S630:

[0114] S610: Merge the word feature vector of the feature word information in the target sentence with the sentence feature vector of the target sentence to obtain an updated feature vector of the target sentence.

[0115] In the embodiment of the present application, the sentence feature vector v of the target sentence is obtained as described above. it On this basis, we further it Word feature vector with feature words Fusion can obtain the updated feature vector of the target sentence After all target sentences in the target review text are processed in this way, the following sequence can be generated based on the position of each target sentence in the review text, that is, the order of each target sentence in the target review text:

[0116] S620: Input the updated feature vector of the target sentence into a preset second bidirectional neural network based on the sentence sequence relationship of the text information to be classified, and obtain a forward output vector of the target sentence and a reverse output vector of the target sentence.

[0117] In the present application embodiment, Figure 2 Similar to the steps provided, when the updated feature vector of the target sentence is updated based on the order relationship of the target sentence in the target comment text, that is, the sequence When input into the second bidirectional neural network, for example, a bidirectional gated recurrent unit (Bi-GRU) based on multi-sentence level implementation is used for processing, the forward output vector of each target sentence can also be obtained. And the reverse output vector in:

[0118]

[0119]

[0120] in, Representation sequence The tth component in and They are respectively the gated recurrent unit for forward processing and the gated recurrent unit for reverse processing in the bidirectional gated recurrent unit. For the description of other parameters, please refer to the above Figure 2 .

[0121] S630: Generate a target feature vector of the target comment text based on the forward output vector of the target sentence and the reverse output vector of the target sentence.

[0122] In the embodiment of the present application, after obtaining the forward output vector and the reverse output vector of the target sentence, the target feature vector of the target comment text can be obtained by splicing and fusing the two.

[0123] like Figure 7 As shown, Figure 7 A schematic flow chart of a step of training a classification model for classification provided in an embodiment of the present application, specifically, comprising steps S710 to S760:

[0124] S710, obtaining a sample feature vector of a sample comment text and a classification label of the sample comment text.

[0125] In the embodiments of the present application, the sample feature vector of the sample comment text refers to the feature vector obtained by processing the sample comment text through steps similar to those mentioned above, and the classification label of the sample comment text is the sentiment tendency classification result obtained in advance based on manual annotation, for example, positive, neutral, negative, etc. Of course, based on the influence of practical factors, the classification labels of the sample comment text usually have an uneven sample quantity, that is, usually, most of the sample comment texts have a classification label of positive, while only a small number of sample comment texts have a classification label of neutral or negative.

[0126] S720: Input the sample feature vector into a preset initial classification model to obtain a predicted classification result of the sample feature vector.

[0127] In the embodiment of the present application, the initial classification model refers to a network model with a certain structure but in which all parameters are initialized values. The sample feature vector is input into the preset initial classification model for processing, and the predicted classification results of each sample feature vector can be obtained. It should be noted that the obtained predicted classification result p(y) is usually present in the form of a probability distribution, which describes the probability value of the sample feature vector corresponding to different classification results y.

[0128] In addition, since the parameters in the initial classification model are all initialized values, the predicted classification results obtained are often quite different from the classification labels obtained by manual annotation. The training of the model is based on the difference to continuously update and optimize the parameters in the classification model, so that when the sample feature vector is input into the classification model, the predicted classification results obtained can fit the classification labels well.

[0129] S730, calculating a model loss value corresponding to the initial classification model based on the number of each category label in the classification label and the difference between the predicted classification result and the classification label.

[0130] In the embodiment of the present application, different from the conventional implementation scheme of calculating the model loss value based only on the difference between the predicted classification result and the classification label, the embodiment of the present application proposes a new model loss function. Specifically, the model loss value is improved on the basis of the multi-classification cross entropy loss value, and the training effect of different categories is adjusted by using the adjustment factor γ corresponding to the number of each category label in the classification label. Specifically, the specific calculation formula of the model loss value is as follows:

[0131] L=-(M(p)·y(1-p) γ log(p)+M(1-p)(1-y)p γ log(1-p)

[0132] S740, updating the initial classification model based on the model loss value to obtain an updated classification model.

[0133] In an embodiment of the present application, based on the model loss value obtained above, the parameters in the initial classification model are further updated based on the back propagation algorithm to obtain an updated classification model.

[0134] S750, determining whether the updated model loss value obtained by inputting the sample feature vector into the updated classification model satisfies the preset training condition. If yes, execute step S660, if no, execute other steps.

[0135] In an embodiment of the present application, when the updated model loss value obtained by inputting the sample feature vector into the updated classification model is less than the preset threshold for several consecutive times, it indicates that the updated classification model obtained at this time can well complete the processing of the sample feature vector and output a predicted classification result similar to the manual annotation result. Therefore, the model can be regarded as having been trained. Otherwise, it is necessary to further use the obtained model loss value to update the parameters in the classification model again until the model training is completed.

[0136] S760, setting the currently updated classification model as the pre-trained classification model.

[0137] In an embodiment of the present application, when the training is completed, the currently obtained classification model is set as the pre-trained classification model, which can be used for subsequent classification processing of the feature vector of the target comment text to obtain the sentiment classification result of the target comment text.

[0138] like Figure 8 As shown, Figure 8 A schematic flow chart of another method for classifying data information provided in an embodiment of the present application, specifically, comprising steps S810 to S840:

[0139] S810: Acquire first text information to be classified of a first target object, and extract feature word information from the first text information to be classified.

[0140] In the embodiment of the present application, based on the application scenario provided above, it can be known that the first text information to be classified of the first target object is usually provided by the second target user, for example, the comment text of the merchant is usually provided by the customer, and the comment text of the customer is usually provided by the merchant. The extraction of the characteristic words in the first text information to be classified can refer to the relevant description of the above step S110.

[0141] S820: Determine a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified.

[0142] S830: Determine a classification result of the text information to be classified based on the target feature vector.

[0143] In the embodiment of the present application, the specific implementation process of the aforementioned steps S820 to S840 can refer to the aforementioned steps S120 to S130 and Figure 2 to Figure 7 The relevant description of the embodiments of the present application will not be repeated here.

[0144] S840: Determine a platform resource allocation result for the first target object based on the classification result of the text information to be classified and the evaluation weight of the second target object.

[0145] Different from the aforementioned classification result of only the target comment text, the solution provided in the embodiment of the present application will further obtain the platform resource allocation result of the first target object based on the classification result of the target comment text. For example, the resource allocation can be the discount of the push and subsidy provided above, or the discount of the red envelope, etc. Of course, in order to avoid malicious evaluations between users and merchants, after the classification result of the text information to be classified of the first target object provided by the second target object, it will also be adjusted based on the evaluation weight of the second target object, wherein the evaluation weight of the second target object is set based on the classification result of the text information to be classified of the second target object, and the classification result of the text information to be classified of the second target object can also be obtained based on the data information classification method provided in the embodiment of the present application.

[0146] Specifically, to facilitate understanding of the above embodiments, take the comment texts of three different users A, B, and C obtained by a certain merchant a as an example. At this time, each user will also be evaluated by other merchants. For example, merchants b, c, and d have evaluated user A. If it is determined through classification processing of the comment texts of merchants b, c, and d on users that the sentiment inclination of merchants b, c, and d towards user A in each evaluation dimension is positive, then when determining the platform resource allocation result of merchant a, the sentiment inclination in the target comment text provided by A will be considered with a higher weight. On the contrary, if it is determined through classification processing of the comment texts of merchants b, c, and d on users that the sentiment inclination of merchants b, c, and d towards user A in each evaluation dimension is negative, then user A may have malicious evaluation behavior. At this time, when determining the platform resource allocation result of merchant a, the sentiment inclination in the target comment text provided by A will be considered with a lower weight, thereby ensuring the rights and interests of the merchant. Of course, in the above process, it is also feasible to interchange merchants and users, and the embodiments of the present application will not be repeated here.

[0147] In order to better implement the data information classification method provided in the embodiment of the present application, on the basis of the data information classification method provided in the embodiment of the present application, the embodiment of the present application also provides a data information classification device, such as Fig. 9 As shown, the data information classification device 900 includes:

[0148] The acquisition module 910 is used to acquire the text information to be classified of the target object and extract the characteristic word information in the text information to be classified;

[0149] A generating module 920, configured to determine a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified;

[0150] The classification module 930 is used to determine the classification result of the text information to be classified based on the target feature vector.

[0151] In some embodiments of the present application, the generation module 920 is also used to generate a sentence feature vector of the target sentence corresponding to the feature word information based on the feature word information and the contextual semantic association relationship of the feature word information; and determine the target feature vector based on the sentence feature vector and the sentence sequence relationship of the text information to be classified.

[0152] In some embodiments of the present application, the generation module 920 is also used to determine the forward output vector of the feature word information based on the feature word information and the preceding and following semantic information in the contextual semantic association relationship of the feature word information; determine the reverse output vector of the feature word information based on the feature word information and the following semantic information in the contextual semantic association relationship of the feature word information; and generate a sentence feature vector of the target sentence corresponding to the feature word information based on the forward output vector and the reverse output vector.

[0153] In some embodiments of the present application, the generation module 920 is also used to fuse the word feature vector of the feature word information with a preset indicator vector to obtain a first input word vector; obtain a second input word vector corresponding to the context semantic information in the context semantic association relationship of the feature word information; and concatenate the first input word vector and the second output word vector into a preset first bidirectional neural network to obtain a forward output vector of the feature word information.

[0154] In some embodiments of the present application, the generation module 920 is also used to obtain the normalized position information of the feature word information in the target sentence corresponding to the feature word information; based on the weight corresponding to the normalized position information, the forward output vector and the reverse output vector are merged to obtain the sentence feature vector of the target sentence.

[0155] In some embodiments of the present application, the generation module 920 is also used to merge the word feature vector of the feature word information in the target sentence with the sentence feature vector of the target sentence to obtain an updated feature vector of the target sentence; based on the sentence sequence relationship of the text information to be classified, the updated feature vector of the target sentence is input into a preset second bidirectional neural network to obtain a forward output vector of the target sentence and a reverse output vector of the target sentence; and based on the forward output vector of the target sentence and the reverse output vector of the target sentence, a target feature vector of the target comment text is generated.

[0156] In some embodiments of the present application, the classification module 93 is also used to input the target feature vector into a pre-trained classification model to obtain a classification result of the text information to be classified, and obtain a sample feature vector of a sample comment text, and a classification label of the sample comment text; input the sample feature vector into a preset initial classification model to obtain a predicted classification result of the sample feature vector; based on the number of category labels in the classification label and the difference between the predicted classification result and the classification label, calculate the model loss value corresponding to the initial classification model; update the initial classification model based on the model loss value to obtain an updated classification model; until the updated model loss value obtained by inputting the sample feature vector into the updated classification model meets the preset training conditions, the currently obtained updated classification model is set as the pre-trained classification model.

[0157] In some embodiments of the present application, the acquisition module 910 is also used to acquire first text information to be classified of the first target object, wherein the first text information to be classified is provided by the second target object; the classification module 930 is also used to determine the platform resource allocation result of the first target object based on the classification result of the text information to be classified and the evaluation weight of the second target object, wherein the evaluation weight of the second target object is set based on the classification result of the text information to be classified of the second target object.

[0158] For the specific definition of the data information classification device, please refer to the definition of the data information classification method above, which will not be repeated here. Each module in the above-mentioned data information classification device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0159] In some embodiments of the present application, the data information classification device 900 can be implemented in the form of a computer program. The computer program can be implemented in Fig.10 The computer device shown in the figure is run. The memory of the computer device can store various program modules constituting the data information classification device 900, for example, Fig. 9 The acquisition module 910, the generation module 920 and the classification module 930 are shown. The computer program composed of various program modules enables the processor to execute the steps in the data information classification method of each embodiment of the present application described in this specification.

[0160] For example, Fig.10 The computer device shown can be Fig. 9The acquisition module 910 in the data information classification device 900 shown executes step S110. The computer device can execute step S120 through the generation module 920. The computer device can execute step S130 through the classification module 930. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external computer device through a network connection. When the computer program is executed by the processor, a data information classification method is implemented.

[0161] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0162] In some embodiments of the present application, a computer device is provided, comprising one or more processors; a memory; and one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the following steps:

[0163] Acquire the to-be-classified text information of the target object, and extract the characteristic word information in the to-be-classified text information;

[0164] Determining a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified;

[0165] A classification result of the text information to be classified is determined based on the target feature vector.

[0166] In some embodiments of the present application, a computer-readable storage medium is provided, which stores a computer program. The computer program is loaded by a processor, so that the processor performs the following steps:

[0167] Acquire the to-be-classified text information of the target object, and extract the characteristic word information in the to-be-classified text information;

[0168] Determining a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified;

[0169] A classification result of the text information to be classified is determined based on the target feature vector.

[0170] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0171] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0172] The above is a detailed introduction to a data information classification method, device, computer equipment and storage medium provided in the embodiments of the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A data information classification method, characterized in that: include: Acquire the to-be-classified text information of the target object, and extract the characteristic word information in the to-be-classified text information; Determining a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified; A classification result of the text information to be classified is determined based on the target feature vector.

2. The data information classification method according to claim 1, characterized in that: The step of determining a target feature vector based on the feature word information, the contextual semantic association of the feature word information, and the sentence sequence of the text information to be classified includes: Generate a sentence feature vector of a target sentence corresponding to the feature word information based on the feature word information and the contextual semantic association relationship of the feature word information; Based on the sentence feature vector and the sentence sequence relationship of the text information to be classified, a target feature vector is determined.

3. The data information classification method according to claim 2, characterized in that: The step of generating a sentence feature vector of a target sentence corresponding to the feature word information based on the feature word information and the contextual semantic association relationship of the feature word information includes: Determining a forward output vector of the feature word information based on the feature word information and the preceding context semantic information in the preceding and following context semantic association relationship of the feature word information; Determining a reverse output vector of the feature word information based on the feature word information and contextual semantic information in a contextual semantic association relationship of the feature word information; Based on the forward output vector and the reverse output vector, a sentence feature vector of the target sentence corresponding to the feature word information is generated.

4. The data information classification method according to claim 3, characterized in that: The step of determining the forward output vector of the feature word information based on the feature word information and the preceding context semantic information in the preceding and following context semantic association relationship of the feature word information comprises: Merging the word feature vector of the feature word information with a preset indicator vector to obtain a first input word vector; Obtaining a second input word vector corresponding to the preceding context semantic information in the preceding and following context semantic association relationship of the feature word information; The first input word vector and the second output word vector are spliced ​​and input into a preset first bidirectional neural network to obtain a forward output vector of the feature word information.

5. The data information classification method according to claim 3, characterized in that: The generating, based on the forward output vector and the reverse output vector, a sentence feature vector of the target sentence corresponding to the feature word information includes: Acquire normalized position information of the feature word information in the target sentence corresponding to the feature word information; The forward output vector and the reverse output vector are fused based on the weight corresponding to the normalized position information to obtain a sentence feature vector of the target sentence.

6. The data information classification method according to claim 2, characterized in that: The step of determining a target feature vector based on the sentence feature vector and the sentence sequence relationship of the text information to be classified includes: Merging the word feature vector of the feature word information in the target sentence with the sentence feature vector of the target sentence to obtain an updated feature vector of the target sentence; Inputting the updated feature vector of the target sentence into a preset second bidirectional neural network based on the sentence sequence relationship of the text information to be classified, to obtain a forward output vector of the target sentence and a reverse output vector of the target sentence; A target feature vector of the target comment text is generated based on the forward output vector of the target sentence and the reverse output vector of the target sentence.

7. The data information classification method according to claim 1, characterized in that: The determining the classification result of the text information to be classified based on the target feature vector includes: The target feature vector is input into a pre-trained classification model to obtain a classification result of the text information to be classified, wherein the classification model is trained through the following steps: Obtaining a sample feature vector of a sample comment text and a classification label of the sample comment text; Inputting the sample feature vector into a preset initial classification model to obtain a predicted classification result of the sample feature vector; Calculating a model loss value corresponding to the initial classification model based on the number of each category label in the classification label and the difference between the predicted classification result and the classification label; Updating the initial classification model based on the model loss value to obtain an updated classification model; Until the updated model loss value obtained by inputting the sample feature vector into the updated classification model meets the preset training conditions, the currently obtained updated classification model is set as the pre-trained classification model.

8. The data information classification method according to any one of claims 1 to 7, characterized in that: The step of obtaining the to-be-classified text information of the target object includes: Acquire first text information to be classified of a first target object, wherein the first text information to be classified is provided by a second target object; After the step of determining the classification result of the text information to be classified based on the target feature vector, the method further includes: Based on the classification result of the text information to be classified and the evaluation weight of the second target object, a platform resource allocation result of the first target object is determined, wherein the evaluation weight of the second target object is set based on the classification result of the text information to be classified of the second target object.

9. A data information classification device, characterized in that: include: An acquisition module is used to acquire the to-be-classified text information of the target object and extract the characteristic word information in the to-be-classified text information; A generating module, configured to determine a target feature vector based on the feature word information, the contextual semantic association relationship of the feature word information, and the sentence sequence relationship of the text information to be classified; A classification module is used to determine a classification result of the text information to be classified based on the target feature vector.

10. A computer device, characterized in that: The computer device comprises: one or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the processor to implement the data information classification method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is loaded by a processor to execute the data information classification method according to any one of claims 1 to 8.