Method and system for processing feedback information

Abstract

The application provides a feedback information processing method and system, the method comprises: obtaining target feedback information, performing semantic classification on the target feedback information through a plurality of semantic recognition models to obtain a plurality of semantic classification results, wherein each semantic classification model is trained to classify a specific semantic, and different semantic classification models correspond to different specific semantics, then determining a target product corresponding to the target feedback information based on the plurality of semantic classification results, and then performing a target operation based on the target product. The application scheme can improve the accuracy of the classification result of the target feedback information, that is, the target product corresponding to the target feedback information can be accurately determined based on the target feedback information.

Description

Technical Field

[0001] This specification relates to the field of information processing technology, and in particular to a method and system for processing feedback information. Background Technology

[0002] To improve product service quality, some systems have launched product feedback platforms. Users can submit information such as problems encountered while using the product, suggestions, and complaints to the feedback platform. After receiving this feedback, the platform can determine which product the feedback pertains to and then improve that product, thereby enhancing product service quality.

[0003] In some related technologies, after receiving feedback from users, the feedback platform has distribution personnel manually analyze and judge the feedback to determine which product the feedback is directed at. Other related technologies pre-train a classification model, which automatically classifies the feedback to determine its target product upon receipt.

[0004] However, in some application scenarios, there are many types of products, such as hundreds or thousands of products. In such scenarios, whether manual classification or automatic classification using a classification model is used, the accuracy of the classification results is not high. Feedback information often cannot be accurately classified into the corresponding products, which affects product quality improvement and reduces the user feedback experience. Summary of the Invention

[0005] This manual provides a method for processing feedback information, which can accurately determine the target product to which the feedback information is directed.

[0006] Firstly, this specification provides a method for processing feedback information, comprising: acquiring target feedback information; performing semantic classification on the target feedback information using multiple semantic classification models to obtain multiple semantic classification results, wherein each semantic classification model is trained to classify specific semantics, and different semantic classification models target different specific semantics; determining the target product to which the target feedback information is targeted based on the multiple semantic classification results; and performing a target operation based on the target product.

[0007] In some embodiments, at least some of the semantic classification models are organized in a tree structure, wherein the tree structure includes K layers, and the semantic granularity of the specific semantics targeted by the semantic classification model is progressively refined according to the order from the first layer to the Kth layer, where K is an integer greater than 1; and the semantic classification of the target feedback information by multiple semantic classification models includes: performing semantic classification of the target feedback information layer by layer according to the tree structure through the semantic classification model of each layer.

[0008] In some embodiments, at least some of the semantic classification models among the plurality of semantic classification models target different semantic dimensions for the specific semantic.

[0009] In some embodiments, the plurality of semantic classification models include: a first coarse semantic classification model and N first specialized semantic classification models, wherein the semantic classification result corresponding to the first coarse semantic classification model includes: the probability corresponding to multiple product categories under the first product classification method, and the semantic classification result corresponding to each of the first specialized semantic classification models includes: the probability corresponding to at least some products in the same product category under the first product classification method.

[0010] In some embodiments, the plurality of semantic classification models further include: a user classification model, wherein the semantic classification result corresponding to the user classification model includes: the probability corresponding to a plurality of preset user types.

[0011] In some embodiments, determining the target product for which the target feedback information is targeted based on the plurality of semantic classification results includes: determining a first semantic feature corresponding to the target feedback information based on the semantic classification results corresponding to the first coarse semantic classification model, the N first specialized semantic classification models, and the user classification model; and classifying the first semantic feature using a first product classification model to determine the target product for which the target feedback information is targeted.

[0012] In some embodiments, the plurality of semantic classification models further include: a second coarse semantic classification model and M second specialized semantic classification models, wherein the semantic classification result corresponding to the second coarse semantic classification model includes: the probability corresponding to multiple product categories under the second product classification method, and the semantic classification result corresponding to each second specialized semantic classification model includes: the probability corresponding to at least some products in the same product category under the second product classification method.

[0013] In some embodiments, determining the target product for which the target feedback information is targeted based on the plurality of semantic classification results includes: determining a first semantic feature corresponding to the target feedback information based on the semantic classification results corresponding to the first coarse semantic classification model, the N first specialized semantic classification models, and the user classification model, and classifying the first semantic feature using a first product classification model to obtain a first product classification result; determining a second semantic feature corresponding to the target feedback information based on the semantic classification results corresponding to the second coarse semantic classification model, the M second specialized semantic classification models, and the user classification model, and classifying the second semantic feature using a second product classification model to obtain a second product classification result; and determining the target product for which the target feedback information is targeted based on the first product classification result and the second product classification result.

[0014] In some embodiments, the plurality of semantic classification models further include: a feedback topic classification model, wherein the semantic classification result corresponding to the feedback topic classification model includes: the probability corresponding to a plurality of preset feedback topics; and determining the target product to which the target feedback information is targeted based on the first product classification result and the second product classification result, including: determining the target product to which the target feedback information is targeted based on the first product classification result, the second product classification result, the semantic classification result corresponding to the user classification model, and the semantic classification result corresponding to the feedback topic classification model.

[0015] In some embodiments, determining the target product for which the target feedback information is targeted, based on the first product classification result, the second product classification result, the semantic classification result corresponding to the user classification model, and the semantic classification result corresponding to the feedback topic classification model, includes: generating a first merged feature based on the first product classification result and the semantic classification result corresponding to the user classification model; generating a second merged feature based on the first product classification result and the semantic classification result corresponding to the feedback topic classification model; generating a third merged feature based on the second product classification result and the semantic classification result corresponding to the user classification model; generating a fourth merged feature based on the second product classification result and the semantic classification result corresponding to the feedback topic classification model; generating a comprehensive semantic feature corresponding to the target feedback information based on the first merged feature, the second merged feature, the third merged feature, and the fourth merged feature; and determining the target product for which the target feedback information is targeted based on the comprehensive semantic feature.

[0016] In some embodiments, generating a comprehensive semantic feature corresponding to the target feedback information based on the first merging feature, the second merging feature, the third merging feature, and the fourth merging feature includes: obtaining the weight matrix corresponding to each of the first merging feature, the second merging feature, the third merging feature, and the fourth merging feature; and performing a weighted calculation on the first merging feature, the second merging feature, the third merging feature, and the fourth merging feature based on the weight matrix to obtain the comprehensive semantic feature.

[0017] In some embodiments, determining the target product for which the target feedback information is targeted based on the comprehensive semantic features includes: classifying the comprehensive semantic features using a third product classification model to obtain the probability of each product in a preset product set; and determining the product with the highest probability as the target product for which the target feedback information is targeted.

[0018] In some embodiments, the target operation includes: pushing the target feedback information to the product manager corresponding to the target product.

[0019] In some embodiments, pushing the target feedback information to the product manager corresponding to the target product includes: obtaining a historical feedback set corresponding to the target product, the historical feedback set including at least one feedback cluster, the historical feedback information in the same feedback cluster having the same semantics, and the historical feedback information in different feedback clusters having different semantics; and determining whether there is a target feedback cluster in the historical feedback set that has the same semantics as the target feedback information, and selecting one of a first scheme and a second scheme to execute according to the determination result, wherein the first scheme includes: when the target feedback cluster does not exist in the historical feedback set, generating a new feedback cluster, adding the target feedback information to the new feedback cluster, and pushing the target feedback information to the product manager; and the second scheme includes: when the target feedback cluster does not exist in the historical feedback set, adding the target feedback information to the target feedback cluster, and if it is determined that the updated target feedback cluster meets a preset reminder condition, then pushing the target feedback information to the product manager.

[0020] In some embodiments, in the first or second scheme, pushing the target feedback information to the product manager includes: determining a target push method based on at least one of the importance level of the feedback cluster to which the target feedback information belongs and the number of cluster information in the feedback cluster to which the target feedback information belongs; and pushing the target feedback information to the product manager in the target push method.

[0021] In some embodiments, in the second scheme, determining that the updated target feedback cluster meets the preset reminder conditions includes: determining that the target feedback cluster meets the preset reminder conditions based on at least one of the number of cluster information in the updated target feedback cluster, the importance level of the target feedback cluster, and the existence duration of the target feedback cluster.

[0022] In some embodiments, after pushing the target feedback information to the product manager corresponding to the target product, the method further includes: obtaining the response information corresponding to the target feedback information; and sending the response information to the submitting user corresponding to the target feedback information.

[0023] In some embodiments, the target feedback information includes at least one of the following: text information, image information, audio information, and video information.

[0024] Secondly, this specification also provides a feedback information processing system, comprising: at least one storage medium and at least one processor, wherein the at least one storage medium stores at least one instruction set for processing feedback information; the at least one processor is communicatively connected to the at least one storage medium, wherein when the feedback information processing system is running, the at least one processor reads the at least one instruction set and executes the feedback information processing method described in any one of the first aspects according to the instructions of the at least one instruction set.

[0025] As can be seen from the above technical solutions, the feedback information processing method and system provided in this specification include: acquiring target feedback information; performing semantic classification on the target feedback information using multiple semantic recognition models to obtain multiple semantic classification results, wherein each semantic classification model is trained to classify specific semantics, and different semantic classification models target different specific semantics; then determining the target product targeted by the target feedback information based on the multiple semantic classification results; and then performing a target operation based on the target product. Since each semantic classification model is trained to classify specific semantics, each semantic classification model corresponds to a relatively simple classification task (e.g., binary or tri-class classification) and has high accuracy in the classification task; furthermore, since different semantic classification models target different specific semantics, the multiple semantic classification results obtained by multiple semantic classification models contain more semantic information, that is, a more comprehensive semantic understanding of the target feedback information. Therefore, determining the target product targeted by the target feedback information based on multiple semantic classification results can improve the accuracy of the determined target product.

[0026] The methods for processing feedback information and other functions of the system provided in this specification are partially listed in the following description. The inventive aspects of the methods for processing feedback information and the system provided in this specification can be fully explained through practice or use of the methods, apparatus, and combinations described in the detailed examples below. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A schematic diagram of an application scenario provided according to an embodiment of this specification is shown;

[0029] Figure 2 A hardware structure diagram of a computing device provided according to an embodiment of this specification is shown;

[0030] Figure 3 A schematic diagram of a product classification method and its corresponding semantic classification model provided according to an embodiment of this specification is shown;

[0031] Figure 4 A schematic diagram of another product classification method and its corresponding semantic classification model provided according to an embodiment of this specification is shown;

[0032] Figure 5 A flowchart is shown of a method for processing feedback information according to an embodiment of this specification;

[0033] Figure 6 A schematic diagram of multiple semantic classification models organized in a tree structure is shown according to an embodiment of this specification;

[0034] Figure 7 A schematic diagram illustrating the overall algorithm model and feedback information processing procedure provided according to embodiments of this specification is shown; and

[0035] Figure 8 A flowchart illustrating a method for pushing feedback information according to an embodiment of this specification is shown. Detailed Implementation

[0036] The following description provides specific application scenarios and requirements for this specification, intended to enable those skilled in the art to make and use the contents of this specification. Various partial modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but rather to the widest scope consistent with the claims.

[0037] The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not restrictive. For example, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. When used in this specification, the terms “comprising,” “including,” and / or “containing” mean that the associated integers, steps, operations, elements, and / or components are present, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components, and / or groups, or that other features, integers, steps, operations, elements, components, and / or groups may be added to the system / method.

[0038] Considering the following description, these and other features of this specification, as well as the operation and function of the related components of the structure, and the economy of assembly and manufacture of the parts, can be significantly improved. All of these form part of this specification with reference to the accompanying drawings. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0039] The flowcharts used in this specification illustrate operations implemented according to some embodiments of this specification. It should be clearly understood that the operations in the flowcharts may not be implemented in a sequential order. Instead, the operations may be implemented in reverse order or simultaneously. Furthermore, one or more additional operations may be added to the flowcharts. One or more operations may be removed from the flowcharts.

[0040] For ease of description, the terms that will appear in the following descriptions will be explained as follows:

[0041] Feedback Information: In this application, feedback information can refer to relevant information submitted by users regarding a specific product, including but not limited to: problems encountered during product use, opinions or suggestions regarding the product, and complaints about the product. It should be noted that this application does not limit the form of feedback information, which may include any one or more combinations of text, audio, image, and video formats. Furthermore, the aforementioned product can be any form of product, such as a physical product, a virtual product, a software product, or a service.

[0042] Before describing the specific embodiments in this specification, the application scenarios of this specification will be introduced as follows:

[0043] Some product providers can offer users a product feedback portal. Users can submit information such as problems encountered during product use, suggestions, and complaints to the feedback platform through this portal. Upon receiving this feedback, the feedback platform can use the feedback processing methods provided in this specification to determine the target product of the feedback, that is, to identify which product (target product) the feedback is directed at. When the product provider offers a large number of product types, this application can pre-train multiple semantic classification models. Each semantic classification model is trained to classify specific semantics, and different semantic classification models correspond to different specific semantics. After obtaining the target feedback, the target feedback can be classified using the aforementioned multiple semantic classification models to obtain multiple semantic classification results. Based on these multiple semantic classification results, the target product of the feedback can then be determined. This approach improves the accuracy of the classification results for the target feedback, i.e., accurately determining which product the feedback is directed at.

[0044] Those skilled in the art should understand that the feedback information processing method described in this specification, when applied to other use cases, is also within the scope of protection of this specification.

[0045] Figure 1 A schematic diagram of an application scenario provided according to an embodiment of this specification is shown. The product feedback system 001 (hereinafter referred to as System 001) can be applied to feedback scenarios for any product, such as feedback scenarios for physical products, virtual products, software products, service products, etc. Figure 1 As shown, system 001 may include target user 100, client 200, server 300 and network 400.

[0046] Target user 100 can be the user submitting feedback information, which can be done through client 200. Feedback information may include, but is not limited to: problems encountered while using the product, opinions or suggestions regarding the product, complaints, etc. Feedback information may include at least one of the following: text information, image information, audio information, and video information.

[0047] Client 200 may be an electronic device that provides interactive functionality to target user 100. Client 200 may provide a feedback submission interface and may respond to feedback submissions from target user 100. In some embodiments, the feedback processing method may be executed on client 200. In this case, client 200 may store data or instructions for executing the feedback processing method described herein, and may execute or be used to execute said data or instructions. In some embodiments, client 200 may include a hardware device with data processing capabilities and the necessary programs required to drive the hardware device. Figure 1 As shown, client 200 can communicate with server 300. In some embodiments, server 300 can communicate with multiple clients 200. In some embodiments, client 200 can interact with server 300 through network 400 to receive or send messages, such as sending feedback information to server 300 and receiving reply information corresponding to the feedback information sent by server 300. In some embodiments, client 200 may include mobile devices, tablets, laptops, built-in devices in motor vehicles, or similar content, or any combination thereof. In some embodiments, the mobile device may include smart home devices, smart mobile devices, virtual reality devices, augmented reality devices, or similar devices, or any combination thereof. In some embodiments, the smart home device may include smart TVs, desktop computers, or any combination thereof. In some embodiments, the smart mobile device may include smartphones, personal digital assistants, gaming devices, navigation devices, or any combination thereof. In some embodiments, the virtual reality device or augmented reality device may include virtual reality headsets, virtual reality glasses, virtual reality patches, augmented reality headsets, augmented reality glasses, augmented reality patches, or similar content, or any combination thereof. For example, the virtual reality device or the augmented reality device may include Google Glass, head-mounted displays, VR, etc. In some embodiments, the built-in equipment in the motor vehicle may include an onboard computer, an onboard television, etc.

[0048] In some embodiments, the client 200 may have one or more applications (APPs) installed. The APPs provide the target user 110 with the ability and interface to interact with the outside world via the network 400. The APPs include, but are not limited to: web browser APPs, search APPs, chat APPs, shopping APPs, video APPs, financial management APPs, instant messaging tools, email clients, social media platform software, etc. In some embodiments, the client 200 may have a target APP installed. The target APP provides an entry point for submitting feedback information to the target user 110. The target APP can respond to the input operation of feedback information and execute the feedback information processing method described in this specification.

[0049] Server 300 may be a server providing various services, such as a backend server supporting the processing of feedback information. In some embodiments, the feedback information processing method may be executed on server 300. In this case, server 300 may store data or instructions for executing the feedback information processing method described herein, and may execute or be used to execute said data or instructions. In some embodiments, server 300 may include hardware devices with data processing capabilities and necessary programs for driving the hardware devices. Server 300 may communicate with multiple clients 200 and receive feedback information sent by clients 200.

[0050] Network 400 serves as a medium to provide a communication connection between client 200 and server 300. Network 400 facilitates the exchange of information or data. For example... Figure 1 As shown, client 200 and server 300 can connect to network 400 and transmit information or data to each other through network 400. In some embodiments, network 400 can be any type of wired or wireless network, or a combination thereof. For example, network 400 may include cable networks, wired networks, fiber optic networks, telecommunications networks, intranets, the Internet, local area networks (LANs), wide area networks (WANs), wireless local area networks (WLANs), metropolitan area networks (MANs), public switched telephone networks (PSTNs), Bluetooth networks, ZigBee networks, near field communication (NFC) networks, or similar networks. In some embodiments, network 400 may include one or more network access points. For example, network 400 may include wired or wireless network access points, such as base stations or Internet switching points, through which one or more components of client 200 and server 300 can connect to network 400 to exchange data or information.

[0051] It should be understood that Figure 1The number of clients 200, servers 300, and networks 400 shown is merely illustrative. Depending on implementation needs, there can be any number of clients 200, servers 300, and networks 400.

[0052] It should be noted that the method for processing the feedback information can be executed entirely on the client 200, entirely on the server 300, or partially on the client 200 and partially on the server 300.

[0053] Figure 2 A hardware structure diagram of a computing device 600 according to an embodiment of this specification is shown. The computing device 600 can execute the feedback information processing method described in this specification. The feedback information processing method is described in other parts of this specification. When the feedback information processing method is executed on a client 200, the computing device 600 can be the client 200. When the feedback information processing method is executed on a server 300, the computing device 600 can be the server 300. When the feedback information processing method is executed partly on the client 200 and partly on the server 300, the computing device 600 can include both the client 200 and the server 300.

[0054] like Figure 2 As shown, the computing device 600 may include at least one storage medium 630 and at least one processor 620. In some embodiments, the computing device 600 may also include a communication port 650 and an internal communication bus 610. Additionally, the computing device 600 may include I / O components 660.

[0055] The internal communication bus 610 can connect different system components, including storage medium 630, processor 620 and communication port 650.

[0056] I / O component 660 supports input / output between computing device 600 and other components.

[0057] Communication port 650 is used for data communication between computing device 600 and external sources. For example, communication port 650 can be used for data communication between computing device 600 and network 400. Communication port 650 can be a wired communication port or a wireless communication port.

[0058] Storage medium 630 may include a data storage device. The data storage device may be a non-transitory storage medium or a temporary storage medium. For example, the data storage device may include one or more of a disk 632, a read-only storage medium (ROM) 634, or a random access storage medium (RAM) 636. Storage medium 630 also includes at least one instruction set stored in the data storage device. The instructions are computer program code, which may include programs, routines, objects, components, data structures, procedures, modules, etc., that execute the feedback information processing methods provided in this specification.

[0059] At least one processor 620 can be communicatively connected to at least one storage medium 630 and a communication port 650 via an internal communication bus 610. At least one processor 620 is used to execute the at least one instruction set described above. When the computing device 600 is running, at least one processor 620 reads the at least one instruction set and, according to the instructions of the at least one instruction set, executes the feedback information processing method provided in this specification. Processor 620 can execute all the steps included in the feedback information processing method. Processor 620 can be in the form of one or more processors. In some embodiments, processor 620 may include one or more hardware processors, such as a microcontroller, microprocessor, reduced instruction set computer (RISC), application-specific integrated circuit (ASIC), application-specific instruction set processor (ASIP), central processing unit (CPU), graphics processing unit (GPU), physical processing unit (PPU), microcontroller unit, digital signal processor (DSP), field-programmable gate array (FPGA), advanced RISC machine (ARM), programmable logic device (PLD), any circuit or processor capable of performing one or more functions, or any combination thereof. For illustrative purposes only, only one processor 620 is described in this specification for the computing device 600. However, it should be noted that the computing device 600 may also include multiple processors. Therefore, the operation and / or method steps disclosed in this specification may be executed by one processor as described herein, or they may be executed jointly by multiple processors. For example, if processor 620 of the computing device 600 in this specification executes steps A and B, it should be understood that steps A and B may also be executed jointly or separately by two different processors 620 (e.g., a first processor executes step A, a second processor executes step B, or the first and second processors jointly execute steps A and B).

[0060] In the aforementioned application scenarios involving product feedback, when the product provider offers a large number of product types, the relevant technologies use manual classification or a single classification model to classify the feedback information. This results in low classification accuracy, and feedback information often fails to be accurately classified into the corresponding product, thus affecting product quality improvement and reducing the user's feedback experience.

[0061] To this end, the proposed solution can pre-train multiple semantic classification models. Each semantic classification model is trained to classify specific semantics, and different semantic classification models target different specific semantics. The following examples illustrate these multiple semantic classification models.

[0062] Figure 3 A schematic diagram illustrating a product classification method and its corresponding semantic classification model according to embodiments of this specification is shown. Figure 3 As shown, assume the preset product set includes 27 products, designated T1 to T27. This application can employ a product classification method to divide these 27 products into several product categories, for example... Figure 3 The example illustrates a division into three product categories (product categories X1 to X3). It should be understood that specific product classification strategies can be employed to group all products into a single category. See also... Figure 3 Assume the classification results are as follows: Product category 1 includes products T1 to T9, product category 2 includes products T10 to T18, and product category 3 includes products T19 to T27.

[0063] In some embodiments, based on Figure 3 The product classification method shown can be pre-trained to obtain a semantic classification model A. Semantic classification model A is trained to identify specific semantics (semantics used to distinguish different product categories) from feedback information and classify the feedback information into product categories X1 to X3 based on the identification results of these specific semantics. The semantic classification result corresponding to semantic classification model A can be denoted as E(A). E(A) can be represented by a 3-dimensional vector, which includes the classification probability corresponding to each of product categories X1 to X3.

[0064] Furthermore, based on Figure 3 The product categorization method shown takes into account that some products within the same category may be similar. User feedback for these similar products often has subtle semantic differences, and there may be semantic overlap and intersection between the feedback for these similar products. Therefore, for... Figure 3 For each product category and group of products that are difficult to distinguish, a specialized semantic classification model can be pre-trained.

[0065] For example, the semantics of feedback information for products T1 and T4 under product category X1 are not easily distinguishable. Therefore, a dedicated semantic classification model F1 can be trained separately for products T1 and T4. The semantic classification model F1 is trained to identify specific semantics from the feedback information (used to distinguish the semantics of products T1 and T4), and classify the feedback information into products T1 and T4 based on the identification results of these specific semantics. The semantic classification result corresponding to the semantic classification model F1 can be denoted as E(F1). E(F1) can be represented by a 2-dimensional vector, which includes the classification probabilities corresponding to products T1 and T4 respectively.

[0066] For example, the semantics of feedback information for products T7, T8, and T9 under product category X1 are not easily distinguishable. Therefore, a dedicated semantic classification model F2 can be trained separately for products T7, T8, and T9. The semantic classification model F2 is trained to identify specific semantics from the feedback information (used to distinguish the semantics of products T7, T8, and T9), and classify the feedback information into products T7, T8, and T9 based on the identification results of these specific semantics. The semantic classification result corresponding to the semantic classification model F2 can be denoted as E(F2). E(F2) can be represented by a 3-dimensional vector, which includes the classification probabilities corresponding to each of products T7, T8, and T9.

[0067] For example, the semantics of feedback information for products T2 and T5 under product category X1 are not easily distinguishable. Therefore, a dedicated semantic classification model F3 can be trained separately for products T2 and T5. Semantic classification model F3 is trained to identify specific semantics from the feedback information (used to distinguish the semantics of products T2 and T5), and classify the feedback information into products T2 and T5 based on the identification results of these specific semantics. The semantic classification result corresponding to semantic classification model F3 can be denoted as E(F3). E(F3) can be represented by a 2-dimensional vector, which includes the classification probabilities corresponding to products T2 and T5 respectively.

[0068] For example, the semantics of feedback information for products T3 and T6 under product category X1 are not easily distinguishable. Therefore, a dedicated semantic classification model F4 can be trained separately for products T3 and T6. Semantic classification model F4 is trained to identify specific semantics from the feedback information (used to distinguish the semantics of products T3 and T6), and classify the feedback information into products T3 and T6 based on the identification results of these specific semantics. The semantic classification result corresponding to semantic classification model F4 can be denoted as E(F4). E(F4) can be represented by a 2-dimensional vector, which includes the classification probabilities for each product T3 and T6.

[0069] It should be noted that different semantic classification models within the same product category can be used to classify overlapping products. For example, assuming that the semantics of the feedback information for products T1 and T3 under product type X1 are not easily distinguishable, a separate semantic classification model F5 can be trained for products T1 and T3. Figure 3 (Not shown in the image), its training method is similar to that of F1 to F4, and will not be elaborated here. Additionally, Figure 3 This paper only demonstrates a semantic classification model for product category 1. For product categories X2 and X3, multiple semantic classification models can be pre-trained using a similar approach, which will not be elaborated upon here. Furthermore, this application does not limit the number of dedicated semantic classification models. Figure 3 This example only uses four specialized semantic classification models, F1 to F4, as illustrations. In practical applications, more semantic classification models can be trained to address the semantic differences between feedback information from different products within each product category.

[0070] What is understandable is that Figure 3 The semantic classification model A and semantic classification models F1 to F4 in this paper differ in the granularity of the specific semantics they target during classification. The semantic granularity of the specific semantics targeted by semantic classification model A is larger (or coarser) than that targeted by semantic classification models F1 to F4. When classifying feedback information, the semantic classification models described above can be executed in order of granularity from coarse to fine. Specifically, the feedback information is first classified using semantic classification model A. Based on the classification results, the product category to which the product targeted by the feedback information belongs can be roughly determined. For example, assuming the determined product category is product category X1, the feedback information is then classified using semantic classification models F1 to F4. Based on the classification results, the product targeted by the feedback information can be more specifically determined within product category X1. It should be understood that the classification results of semantic classification models F1 to F4 should not be regarded as the final classification results. In some embodiments, the classification results of multiple semantic classification models can be comprehensively considered to finally determine the target product targeted by the feedback information. In this application, semantic classification model A can also be called a coarse semantic classification model, and semantic classification models F1 to F4 can also be called specialized semantic classification models.

[0071] It should be understood that there are usually multiple ways to classify products. In some embodiments, for the same preset set of products, different product induction approaches (i.e., product classification methods) can be used to classify the products, resulting in multiple classification results. Furthermore, according to... Figure 3 The idea behind the pre-trained model shown is to pre-train a set of semantic classification models for each classification result.

[0072] Figure 4A schematic diagram illustrating another product classification method and its corresponding semantic classification model provided according to embodiments of this specification is shown. For example... Figure 4 As shown, the method adopted is the same as Figure 3 Another approach to product categorization is to divide the 27 products in the pre-defined product set into three product categories, Y1 to Y3. Based on... Figure 4 The product classification method shown can be pre-trained to obtain a semantic classification model B. Semantic classification model B is trained to identify specific semantics (semantics used to distinguish different product categories) from feedback information and classify the feedback information into product categories Y1 to Y3 based on the recognition results of these specific semantics. The semantic classification result corresponding to semantic classification model B can be denoted as E(B). E(B) can be represented by a 3-dimensional vector, which includes the classification probability corresponding to each of product categories Y1 to Y3.

[0073] Furthermore, regarding Figure 4 For each product category and group of products that are difficult to distinguish, a dedicated semantic classification model can be pre-trained. For example, product category Y1 includes nine products: T1, T2, T3, T11, T22, T13, T7, T17, and T27. The semantics of the feedback information for products T1 and T11 are difficult to distinguish. Therefore, a dedicated semantic classification model G1 can be trained separately for products T1 and T11. Semantic classification model G1 is trained to identify specific semantics from the feedback information (used to distinguish the semantics of products T1 and T11) and classify the feedback information into products T1 and T11 based on the identification results of these specific semantics. The semantic classification result corresponding to semantic classification model G1 can be denoted as E(G1). E(G1) can be represented by a 2-dimensional vector, including the classification probabilities corresponding to products T1 and T11 respectively. Similarly, more semantic classification models G2, G3, G4, etc., can be trained, and their corresponding semantic classification results can be denoted as E(G2), E(G3), E(G4), etc., respectively. Specific ideas and Figure 3 Similarly, this will not be elaborated upon here.

[0074] The above Figure 3 and Figure 4 The semantic classification models shown all classify feedback information based on product features. In other words, Figure 3 and Figure 4 The semantic classification models shown (Model A, Model B, Models F1 to F4, Models G1 to G4) identify specific semantics related to product features from the feedback information to classify the feedback information. To further utilize the more dimensions of information in the feedback, in addition to the product feature dimensions mentioned above, the feedback information can be classified from other dimensions as well.

[0075] In some embodiments, feedback information can be classified from the dimension of user characteristics. Here, "user" refers to the user who submitted the feedback information. Specifically, a classification method for user types can be predefined, for example, dividing users into five types, user type 1 to user type 5. A user classification model C is pre-trained based on this user classification method. The user classification model C is trained to identify specific semantics related to the user who submitted the feedback information from the feedback information, and classify the feedback information into the five user types based on these specific semantics. The semantic classification result corresponding to the user classification model C can be denoted as E(C). E(C) can be represented by a 5-dimensional vector, including the classification probabilities corresponding to user types 1 to 5.

[0076] In some embodiments, feedback information can also be classified from the dimension of feedback topic. Here, the feedback topic refers to the subject of the feedback information content. Specifically, a preset number of feedback topics can be defined according to a preset feedback topic classification method, such as feedback topic 1 to feedback topic 3. A feedback topic classification model D is pre-trained based on this feedback topic classification method. The feedback topic classification model D is trained to identify specific semantics related to the feedback topic from the feedback information and classify the feedback information into the three feedback topics based on these specific semantics. The semantic classification result corresponding to the feedback topic classification model D can be denoted as E(D). E(D) can be represented by a 3-dimensional vector, including the classification probabilities corresponding to each of feedback topics 1 to 3.

[0077] It should be noted that, in addition to the user characteristic dimension and feedback topic dimension mentioned above, this application can also classify feedback information from more other dimensions and pre-train more semantic classification models. It should be understood that by pre-training semantic classification models from multiple dimensions (such as product characteristic dimension, user characteristic dimension, feedback topic dimension, etc.), this application enables multiple semantic classification models to classify feedback information from multiple dimensions, thereby achieving a more accurate semantic understanding of the feedback information. Furthermore, based on the semantic classification results of the feedback information across multiple dimensions, determining the target product addressed by the feedback information helps improve the accuracy of the ultimately identified target product.

[0078] This application does not impose specific limitations on the specific structure and training process of the pre-trained semantic classification models (e.g., models A, B, C, D, F1 to F4, G1 to G4, etc.). In some embodiments, the aforementioned semantic classification models can employ deep models commonly used in Natural Language Processing (NLP), such as the Transformer-based Bidirectional Encoder Representation from Transformers (BERT) model or the Convolutional Neural Network (CNN) model. The model that achieves the highest accuracy on the current specialized classification task can be selected experimentally.

[0079] In some embodiments, when the feedback information contains information other than text, such as image or video information, a specialized semantic classification model for images / videos can be pre-trained to understand the semantics in the image / video for specific semantic purposes and convert it into a classification probability vector output. The above model can be implemented using structures such as the VGG model or Residual Network (ResNet).

[0080] Based on the multiple semantic classification models obtained through pre-training mentioned above, the following section combines... Figure 5 This paper describes the method for processing the feedback information provided in this application.

[0081] Figure 5 A flowchart of a feedback information processing method P100 according to an embodiment of this specification is shown. As previously illustrated, the computing device 600 can execute the feedback information processing method P100 described herein. Specifically, the processor 620 can read an instruction set stored in its local storage medium and then execute the feedback information processing method P100 described herein according to the instructions in the instruction set. Figure 5 As shown, method P100 may include:

[0082] S110: Obtain target feedback information.

[0083] In some embodiments, the target user 100 may submit feedback information on the computing device 600, and the computing device 600 may obtain the target feedback information based on the submission. In some embodiments, the computing device 600 may also receive target feedback information sent by other devices (e.g., client 200). The content of the feedback information includes, but is not limited to: problems encountered while using the product, opinions or suggestions regarding the product, complaints, etc. In some embodiments, the feedback information may include at least one of the following: text information, image information, audio information, and video information.

[0084] S120: The target feedback information is semantically classified using multiple semantic classification models to obtain multiple semantic classification results. Each semantic classification model is trained to classify specific semantics, and different semantic classification models target different specific semantics.

[0085] Because each semantic classification model is trained to classify specific semantics, it is highly targeted in classifying feedback information, primarily focusing on a single semantic meaning rather than classifying based on comprehensive semantics. This results in high classification accuracy for each model within its specific semantic focus. Furthermore, since each model is trained for a specific semantic meaning, it corresponds to a relatively simple classification task, classifying within a smaller range of semantic labels. For example, each model may be a binary or tri-classification model. Therefore, the semantic classification results obtained by classifying feedback information using each model are more accurate. Compared to related technologies that use a single classification model to classify across all products, each semantic classification model in this application has classification expertise in its specific semantic meaning. Therefore, the semantic classification results obtained by classifying feedback information using each model are more accurate.

[0086] Furthermore, among the aforementioned semantic classification models, different models target different specific semantics. Therefore, by using multiple semantic classification models to semantically classify the target feedback information, the resulting semantic classification results contain more semantic information, leading to a more comprehensive semantic understanding of the target feedback information. Consequently, determining the target product corresponding to the target feedback information based on multiple semantic classification results can improve the accuracy of the identified target product.

[0087] In some embodiments, at least some of the semantic classification models are organized in a tree structure, wherein the tree structure includes K layers, and the semantic granularity of the specific semantic being targeted by the semantic classification model is progressively refined in order from the first layer to the Kth layer, where K is an integer greater than 1.

[0088] Figure 6 A schematic diagram is shown of multiple semantic classification models organized in a tree structure, according to embodiments of this specification. For example... Figure 6 As shown in the illustration, taking K=2 as an example, multiple semantic classification models can include: a first coarse semantic classification model and N first specialized semantic classification models. The first coarse semantic classification model is located at the first layer of the tree structure, and the N first specialized semantic classification models are located at the second layer. Both the first coarse semantic classification model and the N first specialized semantic classification models are configured to classify feedback information based on the dimension of the product features targeted by the feedback information. The difference between them lies in the semantic granularity of the specific semantics they target. The semantic granularity of the specific semantics targeted by the first coarse semantic classification model is coarser than that targeted by the N first specialized semantic classification models.

[0089] In some embodiments, when performing semantic classification on the target feedback information, the semantic classification of the target feedback information can be performed layer by layer according to a tree structure, using the semantic classification model at each layer. Figure 6 Taking the tree structure shown as an example, the target feedback information can first be semantically classified using a first coarse semantic classification model, and then semantically classified using N first specialized semantic classification models. In some embodiments, the aforementioned first coarse semantic classification model can be... Figure 3 The semantic classification result E(A) corresponding to the first coarse semantic classification model shown in Model A includes: the first product classification method (i.e. Figure 3 The product classification method shown represents the probabilities of multiple product categories (product categories X1 to X3) under that classification method. The above N first-specific semantic classification models can correspond to... Figure 3 The models are F1, F2, F3, F4, ..., FN. The semantic classification result corresponding to each first specialized semantic classification model includes the probability of at least some products in the same product category under the first product classification method. For example, the semantic classification result E(F1) corresponding to model F1 includes the probability of products T1 and T4 in product category X1.

[0090] It should be noted that in some embodiments, multiple tree structures may exist in the above-mentioned semantic classification models, as shown in the following text. Figure 7 Model A and models F1 to FN form one tree structure, and model B and models G1 to GM form another tree structure. In this case, for each tree structure, execution needs to proceed layer by layer from level 1 to level K. Different semantic classification models within each level can be executed sequentially or in parallel. In some embodiments, among the multiple semantic classification models mentioned above, there may also be semantic classification models that do not belong to any arbitrary tree structure, for example... Figure 7 Models C and D are used in this context. In this case, these semantic classification models can be executed sequentially or in parallel with the tree structure.

[0091] This application improves both the accuracy and overall efficiency of semantic classification by classifying target feedback information from coarse to fine semantic granularity. Furthermore, it organizes multiple pre-trained semantic classification models into a hierarchical tree structure, combining the strengths of each model in specific semantic areas to achieve significantly higher accuracy than single-model classification in complex business classification tasks. Moreover, with the addition of new products, new semantic classification models can be added to the existing algorithm without adjusting existing models, making the overall algorithm highly scalable. In the design and training phases of the multiple semantic classification models, expert experience can be fully incorporated, directly targeting specific semantic classification challenges and training the models accordingly to ensure the overall algorithm's classification performance.

[0092] In some embodiments, at least some of the semantic classification models in the plurality of semantic classification models target different semantic dimensions for the specific semantics. For example, in some embodiments, in addition to a first coarse semantic classification model (e.g., model A described above), N first specialized semantic classification models (e.g., models F1 to FN described above), the plurality of semantic classification models may also include a user classification model (e.g., model C described above). The semantic classification result corresponding to the user classification model includes probabilities corresponding to multiple preset user types. For example, the semantic classification result E(C) corresponding to model C includes probabilities corresponding to user type 1 to user type 5. In this case, the specific semantics corresponding to the first coarse semantic classification model and the N first specialized semantic classification models are product feature dimensions, and the specific semantics corresponding to the user classification model are user feature dimensions. Therefore, the plurality of semantic classification models cover both product feature dimensions and user feature dimensions.

[0093] Furthermore, in some embodiments, the plurality of semantic classification models may further include a feedback topic classification model (e.g., model D described above). The semantic classification result corresponding to the feedback topic classification model includes the probabilities corresponding to multiple preset feedback topics. For example, the semantic classification result E(D) corresponding to model D includes the probabilities corresponding to feedback topics 1 to 3. The specific semantics corresponding to the above feedback topic classification model is the content topic dimension of the feedback information. In this case, the plurality of semantic classification models cover the product feature dimension, the user feature dimension, and the topic dimension of the feedback content.

[0094] In the above embodiments, when at least some of the semantic classification models in the multiple semantic classification models target different semantic dimensions for a specific semantic, it indicates that when the multiple semantic classification models perform semantic classification on the target feedback information, they cover semantics of multiple semantic dimensions, enrich the semantic information identified from the target feedback information, and further improve the comprehensiveness of the semantic understanding of the target feedback information.

[0095] In some embodiments, the plurality of semantic classification models may include, in addition to a first coarse semantic classification model and N first specialized semantic classification models, a second coarse semantic classification model and M second specialized semantic classification models. The second coarse semantic classification model is similar in principle to the first coarse semantic classification model, but they differ in that they are based on different product classification methods. For example, the aforementioned second coarse semantic classification model could be... Figure 4 The semantic classification result E(B) corresponding to the second coarse semantic classification model shown in Model B includes: the second product classification method (i.e. Figure 4 The probabilities corresponding to multiple product categories (product categories Y1 to Y3) under the product classification method shown above. The above M second-specific semantic classification models can correspond to... Figure 4 The models are G1, G2, G3, G4, ..., GN. The semantic classification result corresponding to each second specialized semantic classification model includes the probability of at least some products in the same product category under the second product classification method. For example, the semantic classification result E(G1) corresponding to model G1 includes the probability of products T1 and T11 in product category Y1.

[0096] In some embodiments, the second coarse semantic classification model and the M second specialized semantic classification models may also adopt the following approach: Figure 6 The tree structure shown is used for organization. Accordingly, when performing semantic classification on the target feedback information, the target feedback information can first be semantically classified using a second coarse semantic classification model, and then semantically classified using N first specialized semantic classification models.

[0097] The above embodiments improve the accuracy of semantic understanding of target feedback information by using corresponding semantic classification models for each of the two different product classification methods, thereby helping to improve the accuracy of the final classification results.

[0098] S130: Based on the multiple semantic classification results, determine the target product to which the target feedback information is directed.

[0099] The multiple semantic classification results obtained from the above multiple semantic classification models characterize the semantics expressed by the target feedback information. Therefore, based on the multiple semantic classification results, the target product targeted by the target feedback information can be determined in the preset product set.

[0100] Figure 7 A schematic diagram illustrating the overall algorithm model and feedback information processing procedure provided according to embodiments of this specification is shown. Figure 7 As shown, in some embodiments, when multiple semantic classification models include: a first coarse semantic classification model (model A), N first specialized semantic classification models (models F1 to FN), and a user classification model (model C), assuming the semantic classification result corresponding to the first coarse semantic classification model is denoted as E(A), the semantic classification results corresponding to the N first specialized semantic classification models are denoted as E(F1), E(F2), ..., E(FN), and the semantic classification result corresponding to the user classification model is denoted as E(C), the target product to which the target feedback information is directed can be determined based on the above multiple semantic classification results in the following manner:

[0101] (1) Based on E(A), E(F1), E(F2), ..., E(FN) and E(C), determine the first semantic feature corresponding to the target feedback information.

[0102] In some embodiments, the first semantic feature is denoted as E( ′ The first semantic feature can be generated by concatenating the dimensions of E(A), E(F1), E(F2), ..., E(FN), and E(C), as follows:

[0103]

[0104] (2) The first semantic feature E(A) is analyzed using the first product classification model. ′ The feedback information is categorized to determine the target product to which it is directed.

[0105] The first product classification model is trained to classify products within a preset set, specifically by analyzing the first semantic feature E(A). ′ The model processes the data to determine the probability that the target feedback information applies to each product in the preset product set. Assuming the preset product set contains S products, the first product classification model processes the first semantic feature E(A) to determine the probability that the target feedback information applies to each product in the preset product set. ′ By classifying the products, we can obtain the probabilities corresponding to S products. Let E(I) be the output of the first product classification model, which is a probability vector of dimension S. Then, based on E(I), we can determine the target product for which the target feedback information is directed; for example, the product with the highest probability in E(I) can be taken as the target product.

[0106] It should be noted that this application does not limit the specific structure of the first product classification model. For example, the aforementioned first product classification model can adopt a tree model. Tree models have high computational efficiency and can meet the needs of scenarios with high real-time requirements for feedback information processing. Alternatively, the aforementioned first product classification model can also adopt a model based on artificial neural networks, or any model capable of performing classification functions.

[0107] The above classification process utilizes the semantic classification results of Model A, Models F1 to FN, and Model C, resulting in a more comprehensive semantic understanding of the target feedback information. Therefore, it can improve the accuracy of the final classification result (i.e., the identified target product).

[0108] See also Figure 7 In some embodiments, when multiple semantic classification models include: a first coarse semantic classification model (Model A), N first specialized semantic classification models (Models F1 to FN), a second coarse semantic classification model (Model B), M second specialized semantic classification models (Models G1 to GM), a user classification model (Model C), and a feedback topic classification model (Model D), assuming the semantic classification result corresponding to the first coarse semantic classification model is denoted as E(A), the semantic classification results corresponding to the N first specialized semantic classification models are denoted as E(F1), E(F2), ..., E(FN), the semantic classification result corresponding to the second coarse semantic classification model is denoted as E(B), the semantic classification results corresponding to the M second specialized semantic classification models are denoted as E(G1), E(G2), ..., E(GN), the semantic classification result corresponding to the user classification model is denoted as E(C), and the semantic classification result corresponding to the feedback topic classification model is denoted as E(D), the target product to which the target feedback information is directed can be determined based on the above multiple semantic classification results in the following manner:

[0109] (1) Based on E(A), E(F1), E(F2), ..., E(FN) and E(C), determine the first semantic feature corresponding to the target feedback information.

[0110] In some embodiments, the first semantic feature is denoted as E( ′ The first semantic feature can be generated by concatenating the dimensions of E(A), E(F1), E(F2), ..., E(FN), and E(C), as follows:

[0111]

[0112] (2) The first semantic features are classified using the first product classification model to obtain the first product classification result.

[0113] It should be understood that the execution method of step (2) can be found in the description of the relevant content above, and will not be repeated here. The difference from the previous text is that the output E(I) of the first product classification model is not used as the final classification result, but as the first product classification result.

[0114] (3) Based on E(B), E(G1), E(G2), ..., E(GM) and E(C), determine the second semantic features corresponding to the target feedback information.

[0115] In some embodiments, the second semantic feature is denoted as E( ′ The second semantic feature can be generated by concatenating the dimensions of E(B), E(G1), E(G2), ..., E(GM), and E(C), as follows:

[0116]

[0117] (4) The second semantic features are classified using the second product classification model to obtain the second product classification result.

[0118] It should be understood that the second product classification model is similar to the first product classification model, and the classification process of the second product classification model for the second semantic feature is also similar to the classification process of the first product classification model for the first semantic feature. Assuming the preset product set includes S products, then the second product classification model classifies the second semantic feature E(B) as follows: ′ By classifying the products, we can obtain the probabilities of S products. Let E(J) be the output of the second product classification model. Then, E(J) is a probability vector of dimension S.

[0119] In addition, the execution order of steps (1) and (2) above can be interchanged with that of steps (3) and (4), or steps (1) and (2) above can be executed in parallel with steps (3) and (4).

[0120] (5) Based on the first product classification result and the second product classification result, determine the target product to which the target feedback information is directed.

[0121] Continue to refer to Figure 7 A decision model is connected to the output of the first product classification model and the second product classification model. The decision model can comprehensively consider and make decisions on the first product classification result E(I) and the second product classification result E(J) to determine the target product (i.e. the final classification result) to which the target feedback information is directed.

[0122] In some embodiments, such as Figure 7As shown, the semantic classification results E(C) of model C and E(D) of model D can also be input into the decision model. In this way, the decision model can comprehensively consider E(I), E(J), E(C), and E(D) to determine the target product to which the target feedback information is directed. This application, by also inputting E(C) and E(D) into the decision model, enables the decision-making process to integrate more dimensions of semantic information, thereby improving decision accuracy.

[0123] In some embodiments, the decision-making process of the decision-making model may be as follows:

[0124] (1) Generate the first merge feature A(1) based on E(I) and E(C), generate the second merge feature A(2) based on E(I) and E(D), generate the third merge feature A(1) based on E(J) and E(C), and generate the fourth merge feature A(2) based on E(I) and E(D).

[0125] For example, A(1), A(2), A(1), A(2) can be obtained by calculating the outer product, as follows:

[0126]

[0127]

[0128]

[0129]

[0130] It should be understood that A(1), A(2), A(1), and A(2) above are all in matrix form.

[0131] (2) Generate the comprehensive semantic feature E(IJ) corresponding to the target feedback information based on A(1), A(2), A(1), and A(3).

[0132] In some embodiments, weight matrices W1, W2, W3, and W4 corresponding to A(1), A(2), A(1), and A(3) can be obtained, and A(1), A(2), A(1), and A(2) can be weighted based on the aforementioned weight matrices to obtain the comprehensive semantic feature E(IJ). The aforementioned weight matrices W1, W2, W3, and W4 can be obtained during the training process of the decision model. This application designs a decision model with weight matrices and combines it with a tree model, replacing the relatively simple gating unit in typical hybrid expert systems. This allows the tree model to fully utilize its feature recognition and inductive capabilities, enabling the decision model to more directionally summarize conclusions and significantly improve classification accuracy.

[0133] In some embodiments, the process of generating the comprehensive semantic feature E(IJ) may include: multiplying A(1) by W1 to obtain A. ’ (1) Multiplying A(2) by W2 gives A ’ (2) Multiply A(1) by W3 to get A ’ (1) Multiplying A(2) by W4 gives A ’ (2). Furthermore, A... ’ (1) The elements in each column are summed to obtain a one-dimensional vector E(I1), and A is then used to calculate the vector E(I1). ’ (2) The elements are summed column by column to obtain a one-dimensional vector E(I2), and A is then used to calculate the vector E(I2). ’ The elements in (1) are summed column by column to obtain a one-dimensional vector E(1), and A ’ The elements in (2) are summed column by column to obtain a one-dimensional vector E(2). The dimensions of E(I1), E(I2), E(1), and E(2) are concatenated to obtain E(IJ), as follows:

[0134]

[0135] (3) Based on the comprehensive semantic features E(IJ), determine the target product to which the target feedback information is directed.

[0136] In some embodiments, the decision model may include a third product classification model, which classifies E(IJ) to obtain the probability of each product in a preset product set. Assuming the number of products in the preset product set is S, after inputting E(IJ) into the third product classification model, the output of the model is a probability vector of dimension S, corresponding to the probability of each product in the preset product set. Then, the product with the highest probability in the probability vector is determined as the target product for the target feedback information. This application does not limit the structure of the third product classification model; for example, a DNN network structure or any other network structure with classification functionality can be used.

[0137] Figure 7The training process of the overall algorithm model is as follows: Multiple semantic classification models (Model A, Model B, Models F1 to FN, Models G1 to GM, Model C, and Model D) can be pre-trained separately. These models are trained separately using specialized datasets with different labels, achieving ideal accuracy on specific classification tasks, before being concatenated into the overall algorithm model. During subsequent overall training of the algorithm model, tree models (i.e., the first product classification model and the second product classification model) can be trained first. Once the tree models achieve a relatively ideal accuracy, a decision model is added for further training to further improve classification accuracy. It should be noted that during the overall training of the algorithm model, parameter fine-tuning of the pre-trained semantic classification models is not necessary; otherwise, it may reduce classification performance.

[0138] S140: Perform the target operation based on the target product.

[0139] In this application, after determining the target product to which the target feedback information is directed, the processor 620 can perform target operations based on the target product according to the needs of the application scenario. In some embodiments, the target operation may include: pushing the target feedback information to the product manager corresponding to the target product. That is, the processor 620 can determine the product manager corresponding to the target product based on the target product determined in S130, and push the target feedback information to the product manager so that the product manager can be informed of the target feedback information in a timely manner, thereby improving the target product or responding to the target feedback information in a timely manner. It should be noted that this application does not limit the specific content of the target operation, and it can be set according to the needs of the actual application scenario.

[0140] The following example illustrates the process by which the processor 620 executes a target operation, including pushing target feedback information to the product manager corresponding to the target product. Figure 8 A flowchart of a feedback information push method P200 provided according to an embodiment of this specification is shown. Figure 8 The method shown can be considered as one possible implementation of S140. As previously described, the computing device 600 can execute the feedback information pushing method P200 described herein. Specifically, the processor 620 can read the instruction set stored in its local storage medium and then execute the feedback information pushing method P200 described herein according to the instructions in the instruction set. Figure 8 As shown, method P200 may include:

[0141] S210: Obtain the historical feedback set corresponding to the target product. The historical feedback set includes at least one feedback cluster. The historical feedback information in the same feedback cluster has the same semantics, and the historical feedback information in different feedback clusters has different semantics.

[0142] In this application, all historical feedback information for the target product is maintained in a historical feedback set. To facilitate the maintenance and management of historical feedback information, the historical feedback information in the historical feedback set is stored in the form of clusters (or groups). Historical feedback information in the same feedback cluster has the same semantics, while historical feedback information in different feedback clusters has different semantics. By maintaining historical feedback information with the same semantics in the same cluster, the repeated processing of multiple feedback information with the same actual semantics can be avoided. In some embodiments, the clustering process for feedback information can be implemented using a pre-trained word-to-word vector (word2vec) model, a Transformer model, a Bidirectional Encoder Representation from Transformers (BERT) model, or other deep models. These models can group synonymous feedback information with significant wording differences into one category, thereby making the clustering results more accurate. In addition, in clustering algorithms, algorithms that do not require specifying the number of clusters can be used, such as Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), because the number of feedback clusters in real-world applications is difficult to estimate.

[0143] S220: Determine whether there exists a target feedback cluster in the historical feedback set that has the same semantics as the target feedback information.

[0144] In some embodiments, the similarity between the semantics of the target feedback information and the semantics of each feedback cluster can be calculated separately. If the calculated maximum similarity is greater than or equal to a preset threshold, it indicates that the semantics of the target feedback information are the same as the semantics of the feedback cluster corresponding to the maximum similarity. In this case, the feedback cluster corresponding to the maximum similarity is determined as the target feedback cluster. If the calculated maximum similarity is less than the preset threshold, it indicates that the semantics of the target feedback information are not the same as the semantics of any feedback cluster in the historical feedback set, that is, there is no target feedback cluster in the historical feedback set.

[0145] If it does not exist, then execute S230.

[0146] If it exists, then execute S240.

[0147] S230: Generate a new feedback cluster, add the target feedback information to the new feedback cluster, and push the target feedback information to the product manager of the target product in a targeted manner.

[0148] If the target feedback cluster does not exist in the historical feedback set, it indicates that the target feedback information is being submitted for the first time. Therefore, the processor 620 can generate a new feedback cluster from the historical feedback set and add the current feedback cluster to the new cluster. Then, the processor 620 pushes the target feedback information to the product manager of the target product. In some embodiments, the processor 620 can determine the importance level of the new feedback cluster based on the semantics of the target feedback information. Based on at least one of the importance level of the new feedback cluster and the number of cluster information in the new feedback cluster, a target push method is determined, and then the target feedback information is pushed to the product manager using the target push method. The target push method may include one or more of email, instant messaging, and SMS. It should be noted that pushing the target feedback information to the product manager can refer to pushing the original content of the target feedback information or the semantics corresponding to the target feedback information.

[0149] S240: Add the target feedback information to the target feedback cluster. If it is determined that the updated target feedback cluster meets the preset conditions, push the target feedback information to the product manager.

[0150] If a target feedback cluster exists in the historical feedback set, it indicates that a user has already submitted similar feedback. In this case, the processor 620 can add the target feedback information to the target feedback cluster. Furthermore, the processor 620 can update the number of feedback messages in the target feedback cluster, which represents the number of times similar feedback information has been submitted, and to some extent reflects the urgency of the target feedback cluster needing processing. It is understood that when the target feedback cluster is initially generated, the processor 620 has already pushed the corresponding feedback information to the product manager. Therefore, to avoid frequent pushes to the product manager, the processor 620 can determine whether the updated target feedback cluster meets preset reminder conditions. In some embodiments, the processor 620 can determine whether the target feedback cluster meets the preset reminder conditions based on at least one of the following: the number of cluster information in the updated target feedback cluster, the importance level of the target feedback cluster, and the duration of existence of the target feedback cluster. If the preset reminder conditions are met, the target feedback information is pushed to the product manager. If the preset reminder conditions are not met, there is no need to repeatedly push the target feedback information to the manager. When pushing target feedback information to the product manager, the target push method can be determined based on at least one of the following: the importance level of the target feedback cluster, the number of cluster information in the target feedback cluster, and then the target feedback information is pushed to the product manager using the target push method. The target push method can include one or more of email, instant messaging, and SMS. It should be noted that pushing target feedback information to the product manager can refer to pushing the original content of the target feedback information, pushing historical feedback information with the same semantics as the target feedback information, or pushing the semantics of the target feedback information.

[0151] In some embodiments, after pushing the target feedback information to the product manager, the process may further include: obtaining the response information corresponding to the target feedback information and sending the response information to the user who submitted the target feedback information. Specifically, after the product manager responds to the feedback cluster in which the target feedback information belongs, the response information can be sent to each user who submitted feedback information in that feedback cluster, so that the users who submitted feedback information can be informed of the response status in a timely manner, thus achieving a feedback loop.

[0152] After accurately identifying the target product for which the target feedback information is intended, this solution automatically pushes the target feedback information to the product manager, ensuring that user feedback reaches the product manager promptly and accurately. Furthermore, after the product manager responds, the solution automatically sends the response to the user who submitted the feedback, forming an automated closed-loop problem-solving process. This not only improves the user feedback experience but also enhances product optimization efficiency and reduces the cost of manually classifying and distributing feedback information.

[0153] In summary, the feedback information processing method P100 and system 001 provided in this specification include: acquiring target feedback information; performing semantic classification on the target feedback information using multiple semantic recognition models to obtain multiple semantic classification results, wherein each semantic classification model is trained to classify specific semantics, and different semantic classification models target different specific semantics; then determining the target product targeted by the target feedback information based on the multiple semantic classification results; and then performing a target operation based on the target product. Since each semantic classification model is trained to classify specific semantics, each semantic classification model corresponds to a relatively simple classification task (e.g., binary or tri-class classification) and has high accuracy in the classification task. Furthermore, the different semantic classification models target different specific semantics, resulting in multiple semantic classification results containing more semantic information, i.e., a more comprehensive semantic understanding of the target feedback information. Therefore, determining the target product targeted by the target feedback information based on multiple semantic classification results can improve the accuracy of the determined target product.

[0154] This specification, in another aspect, provides a non-transitory storage medium storing at least one set of executable instructions for processing feedback information. When the executable instructions are executed by a processor, they instruct the processor to implement the steps of the feedback information processing method P100 described in this specification. In some possible embodiments, various aspects of this specification can also be implemented as a program product comprising program code. When the program product is run on a computing device 600, the program code causes the computing device 600 to perform the steps of the feedback information processing P100 described in this specification. The program product for implementing the above method may employ a portable compact disc read-only memory (CD-ROM) containing program code and may run on the computing device 600. However, the program product of this specification is not limited thereto. In this specification, a readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system. The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof. Program code for performing the operations described herein can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on computing device 600, partially on computing device 600, as a standalone software package, partially on computing device 600 and partially on a remote computing device, or entirely on a remote computing device.

[0155] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0156] In summary, after reading this detailed disclosure, those skilled in the art will understand that the foregoing detailed disclosure is presented by way of example only and is not restrictive. Although not explicitly stated herein, those skilled in the art will understand that this specification requires various reasonable changes, improvements, and modifications to the embodiments. These changes, improvements, and modifications are intended to be made by this specification and are within the spirit and scope of the exemplary embodiments described herein.

[0157] Furthermore, certain terms in this specification have been used to describe embodiments of this specification. For example, "an embodiment," "an embodiment," and / or "some embodiments" mean that a particular feature, structure, or characteristic described in connection with that embodiment may be included in at least one embodiment of this specification. Therefore, it is to be emphasized and understood that two or more references to "an embodiment" or "an embodiment" or "alternative embodiment" in various parts of this specification do not necessarily refer to the same embodiment. Moreover, specific features, structures, or characteristics may be suitably combined in one or more embodiments of this specification.

[0158] It should be understood that in the foregoing description of the embodiments in this specification, various features are combined in a single embodiment, drawing, or description for the purpose of simplifying the description and aiding in the understanding of a feature. However, this does not mean that the combination of these features is necessary, and those skilled in the art may readily identify some of the devices as separate embodiments when reading this specification. That is, the embodiments in this specification can also be understood as an integration of multiple secondary embodiments. It is also valid when each secondary embodiment contains fewer than all the features of a single foregoing disclosed embodiment.

[0159] Each patent, patent application, publication of the patent application, and other materials such as articles, books, specifications, publications, documents, articles, etc., cited herein may be incorporated by reference. All contents used for all purposes, except for any history of prosecution documents relating to it, that may be inconsistent with or conflict with this document, or any such history of prosecution documents that may have a limiting effect on the widest extent of the claims, are now or hereafter associated with this document. For example, in the event of any inconsistency or conflict between the description, definition, and / or use of terms associated with any of the included materials and the terms, description, definition, and / or used in connection with this document, the terms used herein shall prevail.

[0160] Finally, it should be understood that the embodiments disclosed herein are illustrative of the principles of the embodiments described in this specification. Other modified embodiments are also within the scope of this specification. Therefore, the embodiments disclosed in this specification are merely examples and not limitations. Those skilled in the art can implement the applications described in this specification using alternative configurations based on the embodiments in this specification. Therefore, the embodiments in this specification are not limited to the embodiments precisely described in the applications.

Claims

1. A method for processing feedback information, comprising: Obtain target feedback information; Multiple semantic classification results are obtained by semantically classifying the target feedback information using multiple semantic classification models. Each semantic classification model is trained to classify specific semantics, which are used to distinguish different product categories, and different semantic classification models target different specific semantics. The multiple semantic classification models include: a first coarse semantic classification model, N first specialized semantic classification models, and a user classification model. The semantic classification result corresponding to the first coarse semantic classification model includes the probability of multiple product categories under the first product classification method. The semantic classification result corresponding to each of the first specialized semantic classification models includes the probability of at least some products in the same product category under the first product classification method. The semantic classification result corresponding to the user classification model includes the probability of multiple preset user types. Based on the semantic classification results corresponding to the first coarse semantic classification model, the N first specialized semantic classification models, and the user classification model, the first semantic feature corresponding to the target feedback information is determined. The first semantic feature is classified using a first product classification model to determine the target product to which the target feedback information is directed; and Perform the target operation based on the target product.

2. The method according to claim 1, wherein, At least some of the semantic classification models are organized in a tree structure, wherein the tree structure includes K layers, and the semantic granularity of the specific semantics targeted by the semantic classification model is progressively refined according to the order from layer 1 to layer K, where K is an integer greater than 1; and The target feedback information is semantically classified using multiple semantic classification models, including: semantically classifying the target feedback information layer by layer according to the tree structure using the semantic classification model at each layer.

3. The method according to claim 1, wherein, At least some of the semantic classification models target different semantic dimensions for the specific semantic meaning.

4. The method according to claim 1, wherein, The plurality of semantic classification models further includes: a second coarse semantic classification model and M second specialized semantic classification models, wherein The semantic classification results corresponding to the second coarse semantic classification model include: the probabilities of multiple product categories under the second product classification method. The semantic classification result corresponding to each second specialized semantic classification includes: the probability of at least some products in the same product category under the second product classification method.

5. The method according to claim 4, wherein, The first semantic feature is classified using a first product classification model to determine the target product to which the target feedback information is directed, including: The first semantic feature is classified using the first product classification model to obtain the first product classification result; Based on the second coarse semantic classification model, the M second specialized semantic classification models, and the semantic classification results corresponding to each of the user classification models, the second semantic feature corresponding to the target feedback information is determined, and the second semantic feature is classified using the second product classification model to obtain the second product classification result; and Based on the first product classification result and the second product classification result, the target product to which the target feedback information is directed is determined.

6. The method according to claim 5, wherein, The plurality of semantic classification models further include: a feedback topic classification model, wherein the semantic classification result corresponding to the feedback topic classification model includes: the probability of multiple preset feedback topics; and Determining the target product for the target feedback information based on the first product classification result and the second product classification result includes: determining the target product for the target feedback information based on the first product classification result, the second product classification result, the semantic classification result corresponding to the user classification model, and the semantic classification result corresponding to the feedback topic classification model.

7. The method according to claim 6, wherein, Based on the first product classification result, the second product classification result, the semantic classification result corresponding to the user classification model, and the semantic classification result corresponding to the feedback topic classification model, the target product to which the target feedback information is directed is determined, including: Based on the first product classification result and the semantic classification result corresponding to the user classification model, a first merged feature is generated, and based on the first product classification result and the semantic classification result corresponding to the feedback topic classification model, a second merged feature is generated. Based on the second product classification result and the semantic classification result corresponding to the user classification model, a third merged feature is generated, and based on the second product classification result and the semantic classification result corresponding to the feedback topic classification model, a fourth merged feature is generated. Based on the first merging feature, the second merging feature, the third merging feature, and the fourth merging feature, a comprehensive semantic feature corresponding to the target feedback information is generated; and Based on the comprehensive semantic features, the target product to which the target feedback information is directed is determined.

8. The method according to claim 7, wherein, Based on the first merging feature, the second merging feature, the third merging feature, and the fourth merging feature, a comprehensive semantic feature corresponding to the target feedback information is generated, including: Obtain the weight matrices corresponding to the first merging feature, the second merging feature, the third merging feature, and the fourth merging feature; and The first merged feature, the second merged feature, the third merged feature, and the fourth merged feature are weighted and calculated based on the weight matrix to obtain the comprehensive semantic feature.

9. The method according to claim 7, wherein, Based on the comprehensive semantic features, determining the target product to which the target feedback information is directed includes: The comprehensive semantic features are classified using a third product classification model to obtain the probability of each product in the preset product set; and The product with the highest probability is identified as the target product for the target feedback information.

10. The method according to claim 1, wherein, The target operation includes: pushing the target feedback information to the product manager corresponding to the target product.

11. The method of claim 10, wherein, The target feedback information is pushed to the product manager corresponding to the target product, including: Obtain the historical feedback set corresponding to the target product. The historical feedback set includes at least one feedback cluster, where historical feedback information within the same cluster has the same semantics, and historical feedback information within different clusters has different semantics. Determine whether a target feedback cluster exists in the historical feedback set that has the same semantics as the target feedback information, and select one of the first and second schemes to execute based on the determination result, wherein... The first solution includes: when the target feedback cluster does not exist in the historical feedback set, generating a new feedback cluster, adding the target feedback information to the new feedback cluster, and pushing the target feedback information to the product manager; The second approach includes: when the target feedback cluster does not exist in the historical feedback set, adding the target feedback information to the target feedback cluster; and if it is determined that the updated target feedback cluster meets the preset reminder conditions, pushing the target feedback information to the product manager.

12. The method of claim 11, wherein, In either the first or second solution, the target feedback information is pushed to the product manager, including: The target push method is determined based on at least one of the following: the importance level of the feedback cluster to which the target feedback information belongs, and the number of cluster information within the feedback cluster to which the target feedback information belongs; and The target feedback information is pushed to the product manager using the target push method.

13. The method according to claim 11, wherein, In the second scheme, determining that the updated target feedback cluster meets preset reminder conditions includes: Based on at least one of the following: the number of cluster information in the updated target feedback cluster, the importance level of the target feedback cluster, and the existence duration of the target feedback cluster, it is determined that the target feedback cluster satisfies the preset reminder condition.

14. The method of claim 10, wherein, After pushing the target feedback information to the product manager corresponding to the target product, the process also includes: Obtain the response information corresponding to the target feedback information; and The response information is sent to the user who submitted the target feedback information.

15. The method according to claim 1, wherein, The target feedback information includes at least one of the following: text information, image information, audio information, and video information.

16. A feedback information processing system, comprising: At least one storage medium storing at least one instruction set for processing feedback information; as well as At least one processor is communicatively connected to the at least one storage medium. When the feedback information processing system is running, the at least one processor reads the at least one instruction set and executes the feedback information processing method according to any one of claims 1 to 15 according to the instructions of the at least one instruction set.

Images