User comment information push method, computer device and readable storage medium
By using the classification model of the convolutional layer, capsule network and activation function layer, user comment data is classified, which solves the problem of low efficiency in collecting user comment information and realizes efficient iterative update of marketing information.
Patent Information
- Application Number
- CN202410763355.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-13
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-06-13
AI Technical Summary
In the prior art, users' recommendation data collection efficiency for product marketing messages is low, and they need to be classified one by one, affecting the iterative update of marketing information.
By determining the target product information and historical user comment data sets, a sample of user comment information is generated, and a classification model of the convolutional layer, capsule network and activation function layer is used to classify user comment data, generate accurate user comment classification information, and complete the It is pushed to the product terminal.
It improves the efficiency of collecting user comment information, simplifies iterative updates of marketing information, and can accurately and efficiently generate user comment classification information.
Smart Images

Figure CN118332126B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computers, and in particular to a method for pushing user comment information, a computer device and a readable storage medium. Background Art
[0002] In online marketing, ordinary users, who are the main audience, are the target objects that advertisers urgently need to gain data insights from. Whether the marketing information of a product is acceptable to users is crucial. At present, the method often used to collect user suggestions for product marketing messages is to collect them through questionnaires. However, the above method usually has the following technical problems: the efficiency of collecting suggestion data is low, and the suggestion data needs to be classified one by one, which affects the iterative update of marketing information. Summary of the invention
[0003] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.
[0004] Some embodiments of the present disclosure propose a method for pushing user comment information, a computer device, and a computer-readable storage medium to solve the technical problems mentioned in the above background technology section.
[0005] In a first aspect, some embodiments of the present disclosure provide a method for pushing user comment information, the method comprising: determining a product identifier corresponding to target product information; determining target text information corresponding to the target product information, and obtaining a historical user comment data set corresponding to the target text information; for each user comment data in the user comment data set, combining the user comment data, the product identifier and the target text information into a user comment information sample; performing sample annotation processing on each user comment information sample to generate a user comment information training sample set; selecting a target user comment information training sample from the user comment information training sample set; determining an initial user comment information classification model, wherein the initial user comment information classification model comprises: a target convolutional layer, a capsule network and an activation function layer; inputting the user comment information sample included in the target user comment information training sample into the target convolutional layer; , obtain a comment convolution information set; input the above comment convolution information set into the above capsule network to obtain comment capsule matrix information; input the above comment capsule matrix information into the above activation function layer to obtain initial user comment classification information; determine the loss value between the above initial user comment classification information and the corresponding sample label; in response to determining that the above loss value is less than or equal to a preset threshold, determine the initial user comment information classification model as a trained user comment information classification model; in response to receiving each user comment data of the current text information, input each user comment data into the above user comment information classification model to obtain a user comment classification result set, and according to the above user comment classification result set, push the user comment data to the product terminal corresponding to the above target product information, wherein the current text information is the latest text information corresponding to the above target product information, and one user comment data corresponds to one user comment classification result.
[0006] In a second aspect, the present disclosure further provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the method described in any implementation of the first aspect is implemented.
[0007] In a third aspect, the present disclosure further provides a computer-readable storage medium, on which a computer program is stored, wherein when the computer program is executed by a processor, the method described in any implementation manner of the first aspect is implemented.
[0008] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: classification information corresponding to user comment information can be accurately and efficiently generated. The collection efficiency of suggestion comment data is improved, and it is convenient to iterate and update marketing information. First, determine the product identification corresponding to the target product information; determine the target text information corresponding to the above target product information, and obtain the historical user comment data set corresponding to the above target text information; for each user comment data in the above user comment data set, combine the above user comment data, the above product identification and the above target text information into a user comment information sample. In this way, it is convenient to determine the relationship between the product and the user comment data. Secondly, perform sample annotation processing on each user comment information sample to generate a user comment information training sample set; select the target user comment information training sample from the above user comment information training sample set; determine the initial user comment information classification model, wherein the above initial user comment information classification model includes: a target convolution layer, a capsule network and an activation function layer. In this way, it is convenient to use the annotated samples to train the initial user comment information classification model. Thus, the user comment data can be classified by the trained user comment information classification model. Next, the user comment information samples included in the target user comment information training samples are input into the target convolution layer to obtain a comment convolution information set; the comment convolution information set is input into the capsule network to obtain comment capsule matrix information; the comment capsule matrix information is input into the activation function layer to obtain initial user comment classification information; and the loss value between the initial user comment classification information and the corresponding sample label is determined. Thus, through the target convolution layer, the feature information of the comment information can be fully obtained. Through the capsule network, multi-level semantic information can be generated to improve the generalization ability of the model. Through the activation function layer, accurate mapping of feature information can be achieved. Then, in response to determining that the loss value is less than or equal to the preset threshold, the initial user comment information classification model is determined as the trained user comment information classification model. Thus, using the convolution layer, capsule network and activation function layer, the classification information of the comment data can be accurately and efficiently generated. Finally, in response to receiving each user comment data of the current text information, the above-mentioned each user comment data is input into the above-mentioned user comment information classification model to obtain a user comment classification result set, and according to the above-mentioned user comment classification result set, the user comment data is pushed to the product terminal corresponding to the above-mentioned target product information, wherein the current text information is the latest text information corresponding to the above-mentioned target product information, and one user comment data corresponds to one user comment classification result. In this way, the classification information corresponding to the user comment information can be accurately and efficiently generated. The collection efficiency of suggestion comment data is improved, and it is convenient to iteratively update the marketing information. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar figure labels represent the same or similar elements. It should be understood that the drawings are schematic and that the components and elements are not necessarily drawn to scale.
[0010] Figure 1 is a flow chart of some embodiments of the method for pushing user comment information according to the present disclosure;
[0011] Figure 2 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION
[0012] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.
[0013] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.
[0014] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.
[0015] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".
[0016] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.
[0017] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.
[0018] Figure 1 1 is a flow chart of some embodiments of the method for pushing user comment information according to the present disclosure. 100 of some embodiments of the method for pushing user comment information according to the present disclosure is shown. The method for pushing user comment information includes the following steps:
[0019] Step 101, determining the product identification corresponding to the target product information.
[0020] In some embodiments, the execution subject (eg, a computer) of the user review information push method may determine the product identification corresponding to the target product information. The target product information may refer to brand information of a certain brand. The product may refer to a brand. The product identification may refer to a brand identification.
[0021] Step 102: determine the target text information corresponding to the target product information, and obtain a historical user review data set corresponding to the target text information.
[0022] In some embodiments, the execution subject may determine the target text information corresponding to the target product information, and obtain a historical user comment data set corresponding to the target text information. The target text information may refer to marketing information about the target product information, such as article information / advertisement / video information. One historical user comment data corresponds to one user. The historical user comment data is text data that comments on the target text information.
[0023] Step 103: for each user comment data in the user comment data set, the user comment data, the product identifier and the target text information are combined into a user comment information sample.
[0024] In some embodiments, the execution entity may combine the user comment data, the product identification and the target text information into a user comment information sample for each user comment data in the user comment data set. The user comment data, the product identification and the target text information may be combined into a user comment information sample.
[0025] Step 104: perform sample labeling processing on each user comment information sample to generate a user comment information training sample set.
[0026] In some embodiments, the execution entity may perform sample annotation processing on each user comment information sample to generate a user comment information training sample set.
[0027] In an actual application scenario, the above execution entity can perform sample labeling processing on each user comment information sample through the following steps:
[0028] In the first step, data quality screening and labeling are performed on each user comment information sample to obtain a sample set of filtered user comment information. Among them, data quality screening can be to eliminate irrelevant or noisy data to ensure the accuracy and relevance of the screening results. In practice, the quality of the data and its relevance to the corresponding product can be judged by understanding the data content in the data and using preset rules and criteria. Comment data containing special characters can be excluded. Labeling can be the process of generating sample labels for each user comment information sample.
[0029] The second step is to perform text standardization on the sample set of screened user comment information to obtain a standardized sample set of user comment information. Text standardization may include: removing duplicate data, processing special characters, removing punctuation marks, and supplementing missing data.
[0030] The third step is to perform word embedding processing on each user comment data in the above-mentioned standardized user comment information sample set to obtain the user comment information sample set after word embedding processing as the user comment information training sample set. For example, a low-dimensional embedding layer can be used to perform word embedding processing on each user comment data in the above-mentioned standardized user comment information sample set to obtain the user comment information sample set after word embedding processing as the user comment information training sample set.
[0031] Step 105: Select target user comment information training samples from the above user comment information training sample set.
[0032] In some embodiments, the execution subject may select a target user comment information training sample from the user comment information training sample set, or may randomly select a user comment information training sample from the user comment information training sample set as the target user comment information training sample.
[0033] Step 106: Determine an initial user comment information classification model.
[0034] In some embodiments, the execution subject may determine an initial user comment information classification model. The initial user comment information classification model includes a target convolution layer, a capsule network, and an activation function layer. The network structure of the initial user comment information classification model and the initial weights of each network layer may be determined. The target convolution layer includes a full convolution network layer sequence, and each full convolution network layer includes a full convolution network sequence. The target convolution layer may be a multi-layer convolution layer. Each convolution layer includes at least one convolution network connected in series. There is at least one convolution network connected in parallel between the convolution networks of each layer. The activation layer may be a function layer for feature information mapping. The user comment information classification model may be a neural network model for classifying user comment information. The full convolution network layer may be a layer of a full convolution network (FCN). For example, the full convolution network layer may include a plurality of fully convolution networks arranged in sequence. Each full convolution network layer has a corresponding dilation rate. Each full convolution network in the full convolution network sequence may be arranged in parallel. The activation function layer may be a function for mapping comment feature information.
[0035] Step 107: input the user comment information samples included in the target user comment information training samples into the target convolution layer to obtain a comment convolution information set.
[0036] In some embodiments, the execution subject may input the user comment information sample included in the target user comment information training sample into the target convolution layer to obtain a comment convolution information set. The comment convolution information may be text feature extraction information corresponding to the user comment information sample.
[0037] In an actual application scenario, the execution subject may input the user comment information samples included in the target user comment information training samples into the target convolution layer to obtain the comment convolution information set:
[0038] 1. Perform word embedding processing on the above user comment information samples to obtain word embedded user comment information samples. A low-dimensional embedding layer can be used to perform word embedding processing on the above user comment information samples to obtain word embedded user comment information samples.
[0039] 2. Input the word embedding vector sequence included in the above word embedding user comment information sample into the full convolutional network layer at the initial position in the full convolutional network layer sequence included in the target convolutional layer to generate an initial full convolutional vector sequence
[0040] 3. Determine the next fully convolutional network layer corresponding to the fully convolutional network layer at the above initial position as the target fully convolutional network layer.
[0041] 4. For the target fully convolutional network layer, perform the following processing steps:
[0042] First, in response to determining that the target full convolutional network layer is the full convolutional network layer at the termination position in the above-mentioned full convolutional network layer sequence, the initial full convolutional vector sequence is input into the target full convolutional network layer to generate a termination full convolutional vector sequence. First, for each convolutional network layer included in the target full convolutional network layer, the following processing steps are performed: determine the time step corresponding to the above-mentioned convolutional network layer as the target time step; determine the initial full convolutional vector corresponding to the above-mentioned target time step in the above-mentioned initial full convolutional vector sequence as the current full convolutional vector; in response to determining that the above-mentioned target time step has a corresponding previous time step, determine the initial full convolutional vector corresponding to the above-mentioned previous time step as the previous full convolutional vector; input the above-mentioned current full convolutional vector and the above-mentioned previous full convolutional vector into the above-mentioned convolutional network layer to generate a termination full convolutional vector. Second, determine each termination full convolutional vector as a termination full convolutional vector sequence.
[0043] Then, a convolution information set is generated according to the initial full convolution vector sequence and the termination full convolution vector sequence. First, the initial full convolution vector sequence can be determined as the initial convolution information. Then, the termination full convolution vector sequence can be determined as the termination convolution information. Finally, the initial convolution information and the termination convolution information are combined to generate a comment convolution information set.
[0044] Step 108: input the comment convolution information set into the capsule network to obtain comment capsule matrix information.
[0045] In some embodiments, the execution entity may input the comment convolution information set into the capsule network to obtain comment capsule matrix information.
[0046] In an actual application scenario, first, the execution subject may perform the following processing steps for each comment convolution information in the comment convolution information set:
[0047] 1. Multiply each full convolution vector in the full convolution vector sequence corresponding to the comment convolution information by the corresponding weight matrix to generate a first weight multiplication matrix, thereby obtaining a first weight multiplication matrix sequence. Each full convolution vector has a corresponding weight matrix.
[0048] 2. Multiply each first weight multiplication matrix in the above first weight multiplication matrix sequence by the corresponding coupling coefficient to generate a second weight multiplication matrix, thereby obtaining a second weight multiplication matrix sequence. The coupling coefficient may be a coupling coefficient in a capsule network.
[0049] 3. Add each second weight multiplication matrix in the above second weight multiplication matrix sequence to obtain an addition matrix.
[0050] Then, the obtained sum matrices are merged into the review capsule matrix information.
[0051] Step 109: Input the comment capsule matrix information into the activation function layer to obtain initial user comment classification information. The activation function layer may be a Softmax function.
[0052] In some embodiments, the execution entity may input the comment capsule matrix information into the activation function layer to obtain initial user comment classification information.
[0053] Step 110, determining the loss value between the above-mentioned initial user comment classification information and the corresponding sample label.
[0054] In some embodiments, the execution subject may determine the loss value between the initial user comment classification information and the corresponding sample label. For example, the loss value between the initial user comment classification information and the corresponding sample label may be determined by a preset loss function. The loss function may be a hinge loss function or a cross entropy loss function.
[0055] Step 111, in response to determining that the above loss value is less than or equal to a preset threshold, determining the initial user comment information classification model as the trained user comment information classification model.
[0056] In some embodiments, the execution entity may determine the initial user comment information classification model as a trained user comment information classification model in response to determining that the loss value is less than or equal to a preset threshold.
[0057] Step 112, in response to receiving each user comment data of the current text information, input the above each user comment data into the above user comment information classification model to obtain a user comment classification result set, and according to the above user comment classification result set, push the user comment data to the product terminal corresponding to the above target product information.
[0058] In some embodiments, the execution subject may, in response to receiving each user comment data of the current text information, input each user comment data into the user comment information classification model to obtain a user comment classification result set, and push the user comment data to the product terminal corresponding to the target product information according to the user comment classification result set. The current text information is the latest text information corresponding to the target product information, and one user comment data corresponds to one user comment classification result. The user comment classification result may indicate whether it belongs to the suggestion user comment. The suggestion user comment may indicate the comment data proposed by the user for product improvement / experience.
[0059] In an actual application scenario, the execution subject may perform the following processing steps for each user comment data in the above user comment data:
[0060] First, the user comment data is word-embedded to generate word-embedded comment information, wherein the word-embedded comment information includes: a word-embedded vector sequence.
[0061] Second, the word embedding comment information is input into the target convolution layer included in the user comment information classification model to obtain a convolution information set. The target convolution layer includes: a full convolution network layer sequence, and each full convolution network layer includes: a full convolution network sequence.
[0062] The above-mentioned word embedding comment information is input into the target convolution layer included in the above-mentioned user comment information classification model to obtain a convolution information set, including:
[0063] In the first step, the above word embedding vector sequence is input into the full convolutional network layer at the initial position in the above full convolutional network layer sequence to generate an initial full convolutional vector sequence.
[0064] The second step is to determine the next fully convolutional network layer in the above fully convolutional network layer sequence corresponding to the above initial fully convolutional network layer as the candidate fully convolutional network layer.
[0065] In the third step, in response to determining that the candidate full convolutional network layer is the full convolutional network layer at the end position in the above-mentioned full convolutional network layer sequence, for each full convolutional network corresponding to the candidate full convolutional network layer, perform the following processing steps:
[0066] 1. Determine the time step corresponding to the above fully convolutional network as the target time step.
[0067] 2. Determine the initial full convolution vector in the above-mentioned initial full convolution vector sequence corresponding to the above-mentioned target time step as the current full convolution vector.
[0068] 3. In response to determining that the target time step has a corresponding previous time step, determine an initial full convolution vector corresponding to the previous time step as a previous full convolution vector.
[0069] 4. Input the above current full convolution vector and the above previous full convolution vector into the above full convolution network to generate a termination full convolution vector.
[0070] The fourth step is to generate a convolution information set based on the initial full convolution vector sequence and the termination full convolution vector sequence.
[0071] Third, the above-mentioned convolution information set is input into the above-mentioned capsule network to obtain capsule matrix information. First, for each convolution information in the above-mentioned convolution information set, the following processing steps are performed: each full convolution vector in the full convolution vector sequence corresponding to the convolution information is multiplied by the corresponding weight matrix to generate a first weight multiplication matrix to obtain a first weight multiplication matrix sequence; each first weight multiplication matrix in the above-mentioned first weight multiplication matrix sequence is multiplied by the corresponding coupling coefficient to generate a second weight multiplication matrix to obtain a second weight multiplication matrix sequence; each second weight multiplication matrix in the above-mentioned second weight multiplication matrix sequence is added to obtain an addition matrix. Then, the obtained addition matrices are merged into capsule matrix information. The coupling coefficient can be a coupling coefficient in a capsule network. Among them, each full convolution vector has a corresponding weight matrix.
[0072] Fourth, the capsule matrix information is input into the activation function layer to obtain the user comment classification result. The activation function layer can be a Softmax function.
[0073] In an actual application scenario, the execution subject can push the user review data to the product terminal corresponding to the target product information through the following steps:
[0074] In the first step, the user comment classification results that meet the preset conditions in the above user comment classification result set are determined as target user comment classification results to obtain a target user comment classification result group. The preset condition may be that the user comment classification results represent suggestion user comments.
[0075] The second step is to push each user comment data corresponding to the above target user comment classification result group to the product terminal corresponding to the above target product information.
[0076] Therefore, through the target convolution layer, the text feature information of the comment information can be fully obtained. Through the capsule network, multi-level semantic information can be generated to improve the generalization ability of the model. Through the activation layer, accurate mapping of feature information can be achieved. Therefore, the classification information of the comment information can be accurately and efficiently generated. Therefore, user comment data can be accurately pushed to the product terminal.
[0077] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present disclosure. The computer device may be a terminal.
[0078] like Figure 2 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.
[0079] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any user comment information pushing method.
[0080] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.
[0081] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any user comment information pushing method.
[0082] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 2 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present disclosure, and does not constitute a limitation on the computer device to which the scheme of the present disclosure 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.
[0083] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0084] In one embodiment, the processor is used to run a computer program stored in a memory to implement the following steps: determining a product identifier corresponding to target product information; determining target text information corresponding to the target product information, and obtaining a historical user review data set corresponding to the target text information; for each user review data in the user review data set, combining the user review data, the product identifier and the target text information into a user review information sample; performing sample annotation processing on each user review information sample to generate a user review information training sample set; selecting a target user review information training sample from the user review information training sample set; determining an initial user review information classification model, wherein the initial user review information classification model includes: a target convolutional layer, a capsule network and an activation function layer; inputting the user review information sample included in the target user review information training sample into the target convolutional layer; In the convolution layer, a comment convolution information set is obtained; the comment convolution information set is input into the capsule network to obtain comment capsule matrix information; the comment capsule matrix information is input into the activation function layer to obtain initial user comment classification information; the loss value between the initial user comment classification information and the corresponding sample label is determined; in response to determining that the loss value is less than or equal to a preset threshold, the initial user comment information classification model is determined as a trained user comment information classification model; in response to receiving each user comment data of the current text information, each user comment data is input into the user comment information classification model to obtain a user comment classification result set, and according to the user comment classification result set, the user comment data is pushed to the product terminal corresponding to the target product information, wherein the current text information is the latest text information corresponding to the target product information, and one user comment data corresponds to one user comment classification result.
[0085] The embodiments of the present disclosure also provide a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions. The method implemented when the program instructions are executed can refer to the various embodiments of the user comment information push method of the present disclosure.
[0086] The computer-readable storage medium may be an internal storage unit of the computer device in the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., provided on the computer device.
[0087] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.
[0088] The serial numbers of the embodiments of the present disclosure are only for description and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific implementation mode of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present disclosure, and these modifications or replacements should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be based on the protection scope of the claims.
Claims
1. A method for pushing user comment information, the method comprising: Determine the product identification corresponding to the target product information; Determine target text information corresponding to the target product information, and obtain a historical user review data set corresponding to the target text information; For each user comment data in the user comment data set, combining the user comment data, the product identifier and the target text information into a user comment information sample; Perform sample labeling processing on each user comment information sample to generate a user comment information training sample set; Selecting a target user comment information training sample from the user comment information training sample set; Determine an initial user comment information classification model, wherein the initial user comment information classification model includes: a target convolution layer, a capsule network and an activation function layer; Inputting the user comment information samples included in the target user comment information training samples into the target convolution layer to obtain a comment convolution information set; Inputting the comment convolution information set into the capsule network to obtain comment capsule matrix information; Inputting the comment capsule matrix information into the activation function layer to obtain initial user comment classification information; Determine a loss value between the initial user comment classification information and the corresponding sample label; In response to determining that the loss value is less than or equal to a preset threshold, determining the initial user comment information classification model as a trained user comment information classification model; In response to receiving each user comment data of the current text information, inputting each user comment data into the user comment information classification model to obtain a user comment classification result set, and pushing the user comment data to the product terminal corresponding to the target product information according to the user comment classification result set, wherein the current text information is the latest text information corresponding to the target product information, and one user comment data corresponds to one user comment classification result; For each user comment data in the user comment data, the following processing steps are performed: Performing word embedding processing on the user comment data to generate word embedding comment information, wherein the word embedding comment information includes: a word embedding vector sequence; Inputting the word embedding comment information into the target convolution layer included in the user comment information classification model to obtain a convolution information set, wherein the target convolution layer includes: a full convolution network layer sequence, and each full convolution network layer includes: a full convolution network sequence; Inputting the convolution information set into the capsule network to obtain capsule matrix information; Inputting the capsule matrix information into the activation function layer to obtain the user comment classification result; The word embedding comment information is input into the target convolution layer included in the user comment information classification model to obtain a convolution information set, including: Inputting the word embedding vector sequence into the fully convolutional network layer at an initial position in the fully convolutional network layer sequence to generate an initial fully convolutional vector sequence; Determine a next fully convolutional network layer in the fully convolutional network layer sequence corresponding to the fully convolutional network layer at the initial position as a candidate fully convolutional network layer; In response to determining that the candidate fully convolutional network layer is the fully convolutional network layer at the terminating position in the sequence of fully convolutional network layers, for each fully convolutional network corresponding to the candidate fully convolutional network layer, performing the following processing steps: Determine a time step corresponding to the full convolutional network as a target time step; Determine an initial full convolution vector corresponding to the target time step in the initial full convolution vector sequence as the current full convolution vector; In response to determining that the target time step has a corresponding previous time step, determining an initial full convolution vector corresponding to the previous time step as a previous full convolution vector; Inputting the current full convolution vector and the previous full convolution vector into the full convolution network to generate a termination full convolution vector; Generate a convolution information set according to the initial full convolution vector sequence and the termination full convolution vector sequence; The method is characterized in that the step of pushing the user review data to a product terminal corresponding to the target product information according to the user review classification result set comprises: Determine the user comment classification results that meet the preset conditions in the user comment classification result set as the target user comment classification results, and obtain a target user comment classification result group; Pushing each user comment data corresponding to the target user comment classification result group to the product terminal corresponding to the target product information; The sample labeling process is performed on each user comment information sample to generate a user comment information training sample set, including: Perform data quality screening and labeling on each user comment information sample to obtain a filtered user comment information sample set; Performing text standardization processing on the sample set of filtered user comment information to obtain a standardized sample set of user comment information; Word embedding processing is performed on each user comment data in the standardized user comment information sample set to obtain a user comment information sample set after word embedding processing as a user comment information training sample set.
2. A computer device, wherein: The computer device comprises a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein the computer program implements the steps of the method according to claim 1 when executed by the processor.
3. A computer-readable storage medium, wherein: The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the method according to claim 1 are implemented.
