User feedback information collection methods, devices, and user feedback systems

By acquiring and processing user feedback information in the in-vehicle intelligent voice system, and utilizing feedback information extraction models and knowledge graph technology, the problem of ineffective application of user feedback information has been solved, thereby achieving system performance optimization and user experience improvement.

CN115689603BActive Publication Date: 2026-11-17IFLYTEK CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202211185180.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-09-27
Publication Date
2026-11-17
Estimated Expiration
2042-09-27

AI Technical Summary

Technical Problem

Existing in-vehicle intelligent voice systems cannot effectively utilize user feedback information from different functions, resulting in wasted feedback information and an inability to apply it to system performance optimization in a timely manner.

Method used

By acquiring the user's contextual dialogue text, using the feedback information extraction model to determine the user feedback information and feature representation, and calculating its relevance to the product function knowledge graph, the user feedback information is filled into the knowledge graph to achieve function upgrade.

Benefits of technology

This improved the effective use of user feedback information, enabling timely application to the performance optimization of the in-vehicle intelligent voice system, avoiding information waste, and enhancing the user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115689603B_ABST
    Figure CN115689603B_ABST
Patent Text Reader

Abstract

This invention relates to the field of data processing technology, providing a method, apparatus, and system for extracting user feedback information. This method writes user feedback information after using various product functions into a product function knowledge graph. This allows developers to promptly understand the user experience of each product function, facilitating subsequent updates and upgrades and improving user experience. Furthermore, this method has good scalability; subsequent maintenance and upgrades of the product function knowledge graph are sufficient to quickly collect user feedback information corresponding to each product function. Specifically, for in-vehicle intelligent voice systems, it can quickly identify user feedback information for each function, enabling effective application of this feedback and timely optimization of the system's performance, thus avoiding the waste of user feedback information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

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

[0002] Currently, intelligent voice systems, with their unique advantages, have been widely used in various industries, especially in-vehicle intelligent voice systems, which can help users reduce their reliance on manual operation of in-vehicle devices and increase driving safety.

[0003] In existing technologies, users typically interact with an in-vehicle intelligent voice system via voice. The system parses the user's voice and provides feedback. After receiving the feedback, the user can provide their evaluation of the feedback—user feedback information—to the system, enabling more accurate feedback to be provided in the future.

[0004] However, due to the numerous functions of current in-vehicle intelligent voice systems, and the vast amount of user feedback for different functions, these systems cannot effectively utilize this feedback. This leads to a waste of user feedback information and prevents its timely application in subsequent performance optimization. Therefore, there is an urgent need to provide a method for collecting user feedback information. Summary of the Invention

[0005] This invention provides a method, apparatus, and system for collecting user feedback information to address the deficiencies in the prior art.

[0006] This invention provides a method for collecting user feedback information, comprising:

[0007] Obtain the user's contextual dialogue text, which is obtained based on the user's contextual dialogue speech conversion;

[0008] Based on the feedback information extraction model, the user feedback information and user feedback feature representation in the contextual dialogue text are determined.

[0009] Calculate the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and fill the user feedback information into the product function knowledge graph based on the correlation.

[0010] The feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags.

[0011] According to a user feedback information collection method provided by the present invention, the step of determining the user feedback information and user feedback feature representation in the contextual dialogue text based on a feedback information extraction model includes, prior to:

[0012] Determine the language information corresponding to the contextual dialogue voice;

[0013] Accordingly, the step of determining the user feedback information and user feedback feature representation in the contextual dialogue text based on the feedback information extraction model specifically includes:

[0014] The contextual dialogue text and the language information are input into the feedback information extraction model to obtain the user feedback information and the user feedback feature representation output by the feedback information extraction model.

[0015] According to a method for collecting user feedback information provided by the present invention, the user feedback information includes evaluation tendency information and evaluation vocabulary information;

[0016] The feedback information extraction model is trained based on the following method:

[0017] Based on the contextual dialogue text, the pre-determined multilingual pre-trained model is jointly trained for evaluation bias classification tasks and evaluation vocabulary extraction tasks to obtain the feedback information extraction model.

[0018] According to a user feedback information collection method provided by the present invention, the node feature representation is determined based on the following method:

[0019] The text information of each product function node and the text information of each edge in the product function knowledge graph are encoded to obtain the encoding result;

[0020] Based on the encoding results and the relationship between each product function node and each edge in the product function knowledge graph, the node feature representation is determined.

[0021] According to a user feedback information collection method provided by the present invention, the step of filling the user feedback information into the product function knowledge graph based on the correlation specifically includes:

[0022] Select the target product function node with the highest relevance from the product function knowledge graph, and fill the target product function node with the user feedback information.

[0023] According to a user feedback information collection method provided by the present invention, the step of filling the user feedback information into the product function knowledge graph based on the correlation includes:

[0024] Identify the target product function nodes with user feedback information in the product function knowledge graph;

[0025] Based on user feedback information at the target product's functional nodes, the functional nodes of the target product are upgraded.

[0026] The present invention also provides a user feedback information collection device, comprising:

[0027] The acquisition module is used to acquire the user's contextual dialogue text, which is obtained based on the user's contextual dialogue speech conversion;

[0028] The extraction module is used to extract user feedback information and user feedback feature representation in the context dialogue text based on the feedback information extraction model.

[0029] The collection module is used to calculate the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and to fill the user feedback information into the product function knowledge graph based on the correlation.

[0030] The feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags.

[0031] The present invention also provides a user feedback system, comprising: a voice module and the aforementioned user feedback information collection device; the voice module and the user feedback information collection device are connected.

[0032] The voice module is used to acquire the user's contextual dialogue voice, convert the contextual dialogue voice into contextual dialogue text, determine the response information corresponding to the contextual dialogue voice, and provide the response information to the user.

[0033] The voice module is also used to send the contextual dialogue voice and / or the contextual dialogue text to the user feedback information collection device.

[0034] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the user feedback information collection method as described above.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the user feedback information collection method as described above.

[0036] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the user feedback information collection method as described above.

[0037] The user feedback information collection method, apparatus, and user feedback system provided by this invention first acquire the user's contextual dialogue text. Then, based on a feedback information extraction model, it determines the user feedback information and user feedback feature representation in the contextual dialogue text. Subsequently, it calculates the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and based on this correlation, fills the user feedback information into the product function knowledge graph. This method writes user feedback information after using each product function into the product function knowledge graph, which allows developers to promptly grasp the user experience of each product function, facilitating subsequent updates and upgrades to product functions and improving user experience. Moreover, this method has good scalability; subsequent maintenance and upgrades only require the product function knowledge graph to quickly collect user feedback information corresponding to each product function. In particular, for in-vehicle intelligent voice systems, it can quickly determine the user feedback information for each function, enabling effective application of user feedback information for different functions and timely application of user feedback information to subsequent performance optimization of the in-vehicle intelligent voice system, avoiding the waste of user feedback information. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on the drawings described below without creative effort.

[0039] Figure 1 This is a flowchart illustrating the user feedback information collection method provided by the present invention;

[0040] Figure 2 This is a schematic diagram of the product function knowledge graph corresponding to the car voice function library in the user feedback information collection method provided by the present invention;

[0041] Figure 3 This is a schematic diagram illustrating the implementation process of the user feedback information collection method provided by the present invention;

[0042] Figure 4 This is a schematic diagram of the updated product function knowledge graph in the user feedback information collection method provided by the present invention;

[0043] Figure 5 This is a schematic diagram of the user feedback information collection device provided by the present invention;

[0044] Figure 6 This is a schematic diagram of the user feedback system provided by the present invention;

[0045] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0047] Current in-vehicle intelligent voice systems cannot effectively utilize user feedback information for different functions, leading to a waste of this feedback and hindering its timely application in subsequent performance optimization. Furthermore, every user is highly concerned about the user experience and interaction efficiency of intelligent cockpit products, and service providers increasingly need to intelligently analyze user interaction information to provide a superior user experience. Therefore, user feedback and rapid analysis during use are extremely important. To address this, this invention provides a method for collecting user feedback information. This method can be applied not only to in-vehicle intelligent voice systems but also to voice systems in other scenarios, such as user feedback systems.

[0048] Figure 1 This is a flowchart illustrating a user feedback information collection method provided in an embodiment of the present invention, such as... Figure 1 As shown, the method includes:

[0049] S1, Obtain the user's contextual dialogue text, which is obtained based on the user's contextual dialogue speech conversion;

[0050] S2, Based on the feedback information extraction model, determine the user feedback information and user feedback feature representation in the contextual dialogue text;

[0051] S3, calculate the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and fill the user feedback information into the product function knowledge graph based on the correlation.

[0052] The feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags.

[0053] Specifically, the user feedback information collection method provided in this embodiment of the invention is executed by a user feedback information collection device. This device can be configured in a user feedback system or in a server. The server can be a local server or a cloud server. The local server can be a computer, tablet computer or other smart device. This embodiment of the invention does not make any specific limitations on this.

[0054] First, execute step S1 to obtain the user's contextual dialogue text. This user can be a user of the product function, such as a driver. The product function can be any product function in the product function library, which can be an in-vehicle function library or a function library from other systems; no specific limitation is made here.

[0055] The contextual dialogue text refers to the user-side text-based contextual dialogue when a user interacts with the intelligent voice system corresponding to the product function library. This contextual dialogue text can determine the user's usage of the product functions and their attitude towards those functions. This contextual dialogue text can be obtained by converting the user's contextual dialogue speech; this contextual dialogue speech is the user-side voice-based contextual dialogue.

[0056] It is understood that the contextual dialogue text or contextual dialogue voice refers to the contextual dialogue within a single conversation. The contextual dialogue text may include one or more statements.

[0057] Then, step S2 is executed. After obtaining the contextual dialogue text, the user feedback information and its feature representation are determined by combining the feedback information extraction model. User feedback information can be used to characterize the user's evaluation of the target product function involved in the contextual dialogue text, i.e., the evaluation information of the product function used by the user. For example, it may include the user's attitude towards the target product function and evaluation vocabulary. User feedback feature representation can be used to characterize the deep features of the user feedback information.

[0058] The feedback information extraction model can be obtained by training an initial model using contextual dialogue text samples carrying user feedback information labels. This initial model can be built based on a neural network.

[0059] When training the initial model, contextual dialogue text samples can be input into the initial model to obtain the user feedback information detection results output by the initial model. Then, the loss function value is calculated through the user feedback information labels and the user feedback information detection results. When the loss function value converges, the trained feedback information extraction model is obtained.

[0060] When applying the feedback information extraction model, the contextual dialogue text can be input into the model to obtain user feedback information output by the model. Simultaneously, the feedback information extraction model also outputs a user feedback feature representation. This user feedback feature representation is an intermediate product of the feedback information extraction model in obtaining user feedback information. This user feedback feature representation can be called a contextual dialogue aggregation feature, and can be in vector form or feature map form; no specific limitation is made here.

[0061] Next, step S3 is executed to calculate the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph. This product function knowledge graph is the knowledge graph formed by the various product functions in the product function library. This product function knowledge graph can include multiple product function nodes and edges connecting two product function nodes. Product function nodes can include multiple levels, such as... Figure 2 As shown. Taking the automotive voice function library in the in-vehicle function library as an example, product functions can be divided into primary product functions such as in-vehicle application functions and automotive component control functions. Each primary product function can include secondary product functions, such as navigation and music functions under in-vehicle application functions, and air conditioning and window functions under automotive component control functions. Each secondary product function can also include tertiary product functions, such as location search, location navigation, and map zoom functions under navigation functions, music playback, favorite music, and loop mode functions under music functions, and temperature adjustment, fan speed adjustment, and airflow direction adjustment functions under air conditioning functions.

[0062] Level 1, Level 2, and Level 3 product functions all have corresponding product function nodes in the product function knowledge graph, namely, Level 1 product function node, Level 2 product function node, and Level 3 product function node, respectively.

[0063] Each product function node in the product function knowledge graph has a corresponding node feature representation. This node feature representation can be extracted by inputting the product function knowledge graph into a graph attention network. The node feature representation has the same form as the user feedback feature representation, which can be either a vector form or a feature map form. No specific limitation is made here.

[0064] Subsequent product feature expansions can flexibly configure the information of product feature nodes and edges in the product feature knowledge graph, enabling rapid expansion. After expansion, the latest node feature representations can be obtained, completing the update and construction of the product feature knowledge graph.

[0065] Here, the correlation can be calculated by determining the cosine distance between the user feedback feature representation and the node feature representation of each product function node. The calculated cosine distance is used to characterize the corresponding correlation. A larger cosine distance indicates a stronger correlation, and a smaller cosine distance indicates a weaker correlation.

[0066] By combining user feedback information and its relevance, this information can be incorporated into the product feature knowledge graph. For example, based on relevance, the evaluation subject corresponding to the user feedback can be determined, i.e., the target product feature in the product feature knowledge graph. Then, the user feedback information can be added to the target product feature section of the knowledge graph and stored. This allows for updating the product feature knowledge graph. The updated knowledge graph not only contains each product feature node and the edges connecting adjacent product feature nodes, but also the corresponding user feedback information for each product feature node.

[0067] If a user's historical feedback for a target product function already exists in that function before the feedback is entered, that feedback can be used to replace the existing historical feedback to ensure that the user feedback entered into the product function knowledge graph is up-to-date.

[0068] The user feedback information collection method provided in this embodiment of the invention first obtains the user's contextual dialogue text, then determines the user feedback information and user feedback feature representation in the contextual dialogue text based on a feedback information extraction model; subsequently, it calculates the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and fills the user feedback information into the product function knowledge graph based on this correlation. This method writes user feedback information after using each product function into the product function knowledge graph, which allows developers to promptly grasp the user experience of each product function, facilitating subsequent updates and upgrades to the product function and improving user experience. Moreover, this method has good scalability; subsequent maintenance and upgrades of the product function knowledge graph are sufficient to quickly collect user feedback information corresponding to each product function. In particular, for in-vehicle intelligent voice systems, it can quickly determine the user feedback information for each function, enabling effective application of user feedback information for different functions and timely application of user feedback information to subsequent performance optimization of the in-vehicle intelligent voice system, avoiding the waste of user feedback information.

[0069] Based on the above embodiments, the user feedback information collection method provided in this embodiment of the invention, prior to determining the user feedback information and user feedback feature representation in the contextual dialogue text based on the feedback information extraction model, includes:

[0070] Determine the language information corresponding to the contextual dialogue voice;

[0071] Accordingly, the step of determining the user feedback information and user feedback feature representation in the contextual dialogue text based on the feedback information extraction model specifically includes:

[0072] The contextual dialogue text and the language information are input into the feedback information extraction model to obtain the user feedback information and the user feedback feature representation output by the feedback information extraction model.

[0073] Specifically, in this embodiment of the invention, after obtaining the user's contextual dialogue text and before determining the user feedback information and user feedback feature representation in the contextual dialogue text through a feedback information extraction model, the language information corresponding to the contextual dialogue speech can be determined first. This language information can be obtained by inputting the contextual dialogue speech into a language recognition model, which then identifies and outputs the language information of the contextual dialogue speech. This language recognition model can be obtained by training an initial model with speech samples carrying language labels. This initial model can be a neural network model, and its specific structure can be set as needed; no specific limitations are made here. Language information can include Chinese [ch], English [en], etc., and Chinese can further include Mandarin, Tibetan, Uyghur, etc.

[0074] Furthermore, when determining user feedback information and user feedback feature representations in the contextual dialogue text based on the feedback information extraction model, both the contextual dialogue text and language information can be input into the feedback information extraction model, which then outputs the user feedback information and user feedback feature representations. When inputting both the contextual dialogue text and language information into the feedback information extraction model, the language information can be concatenated with the contextual comparison text to obtain a concatenated result, which can then be used as input to the feedback information extraction model.

[0075] Because the feedback information extraction model incorporates language information into its input, the contextual dialogue text samples used during the initial model training can also carry language information labels. Therefore, during initial model training, contextual dialogue text samples can be input into the initial model to obtain the user feedback information detection results and language recognition results output by the initial model. Then, a first loss function value is calculated using the user feedback information labels and the user feedback information detection results, and a second loss function value is calculated using the language labels and the language recognition results. Finally, the first and second loss function values ​​are added to calculate the target loss function value. When the target loss function value converges, the trained feedback information extraction model is obtained. The resulting feedback information extraction model is a multilingual feedback information extraction model.

[0076] In this embodiment of the invention, by incorporating language information, the feedback information extraction model is not limited by region or user, and can accurately collect user feedback information. Furthermore, by filling the product function knowledge graph with a matching scheme, the user experience of product functions can be quickly and in real-time perceived.

[0077] Based on the above embodiments, the user feedback information collection method provided in this embodiment of the invention includes evaluation tendency information and evaluation vocabulary information;

[0078] The feedback information extraction model is trained based on the following method:

[0079] Based on the contextual dialogue text, the pre-determined multilingual pre-trained model is jointly trained for evaluation bias classification tasks and evaluation vocabulary extraction tasks to obtain the feedback information extraction model.

[0080] Specifically, in this embodiment of the invention, user feedback information may include evaluation bias information and evaluation vocabulary information. Evaluation bias information refers to the user's attitude towards the target product function involved in the contextual dialogue text, and may include positive, negative, and neutral. Positive indicates a good evaluation, negative indicates a bad evaluation, and neutral indicates a neutral evaluation. Evaluation vocabulary information refers to the words used by the user to express their attitude towards the target product function involved in the contextual dialogue text, and may include, for example, good, bad, accurate, inaccurate, right, wrong, etc. Here, neither evaluation bias information nor evaluation vocabulary information is specifically limited.

[0081] Furthermore, the feedback information extraction model needs to have evaluation bias classification and evaluation vocabulary extraction functions. Therefore, when training the initial model, a pre-trained multi-functional pre-trained model can be used as the initial model. On this basis, the multilingual pre-trained model can be jointly trained for evaluation bias classification and evaluation vocabulary extraction tasks to obtain a feedback information extraction model suitable for multiple languages. This allows the feedback information extraction model to perform evaluation bias classification and evaluation vocabulary extraction tasks simultaneously, and output evaluation bias information and evaluation vocabulary information at the same time.

[0082] Understandably, multilingual pre-trained models can be trained using regular corpora in different languages. This improves the adaptability of the feedback information extraction model to different languages ​​and also increases its training efficiency. The evaluation bias information and evaluation vocabulary information output by the feedback information extraction model are usually represented in probability form, that is, the model outputs positive and negative evaluation probabilities and the probability that each word in the context dialogue text belongs to the evaluation vocabulary.

[0083] In this embodiment of the invention, the multilingual pre-trained model can be a multi-BERT model. The mathematical representation of the feedback information extraction model can be:

[0084] P i c ,P i e C i =multi-BERT([L i ,S i ])

[0085] Among them, L i S represents the language information corresponding to the user's contextual dialogue voice during the i-th contextual dialogue, such as Chinese [ch], English [en], etc., usually in the form of word vectors; i This represents the contextual dialogue text of the user during the i-th contextual dialogue, such as "Help me navigate to XXX -> go to XXX -> I can't seem to find this location," and has S. i =[t1,t2,......,t n ], t n P represents the nth statement in the context dialogue text, and can be in vector form; i c Let P represent the probability of a user's positive or negative evaluation during the i-th contextual dialogue. i e C represents the probability that each word in the context dialogue text belongs to the evaluation vocabulary. i The representation of user feedback features during the i-th contextual dialogue process is called the contextual dialogue aggregated feature, which can be in vector form.

[0086] In this embodiment of the invention, the introduction of a multilingual pre-trained model not only improves the adaptability of the feedback information extraction model to different languages ​​but also enhances its training efficiency. Furthermore, the feedback information extraction model obtained through multi-task training can simultaneously perform multiple tasks and accurately output evaluation bias information and evaluation vocabulary information, facilitating a rapid and comprehensive understanding of user feedback information.

[0087] Based on the above embodiments, the user feedback information collection method provided in this embodiment of the invention determines the node feature representation based on the following method:

[0088] The text information of each product function node and the text information of each edge in the product function knowledge graph are encoded to obtain the encoding result;

[0089] Based on the encoding results and the relationship between each product function node and each edge in the product function knowledge graph, the node feature representation is determined.

[0090] Specifically, in this embodiment of the invention, when determining the node feature representation, the text information of each product function node and the text information of each edge in the product function knowledge graph can be encoded to obtain the encoding result. The text information of each product function node is the node keyword text of each product function node, and the text information of each edge is the edge keyword text of each edge. This encoding process can be implemented through encoding networks such as the BERT pre-trained model.

[0091] For example, such as Figure 2 As shown, the keyword text for navigation, location navigation, and other nodes of each product function node in the product function knowledge graph is denoted as t1, t2, ..., and the keyword text for the first-level function, second-level function, third-level function, etc., of each edge in the product function knowledge graph is denoted as te. 12 ...

[0092] Using an encoding network, node keyword text and edge keyword text can be encoded separately. The encoding results can include node feature representations and edge feature representations. The node feature representations can be denoted as f1, f2, ..., and the edge feature representations can be denoted as e. 12 ...

[0093] Subsequently, combining the encoding results and the relationships between each product function node and its edges in the product function knowledge graph, Graph Attention Networks (GAT) can be used to determine the node feature representation of each product function node. That is, the encoding results and the product function knowledge graph can both be input into GAT to obtain the node feature representation of each product function node in the product function knowledge graph output by GAT. The node feature representation of each product function node can be denoted as N1, N2, ...

[0094] The calculation process for determining the node feature representation of each product functional node described above can be achieved through the following formula:

[0095] f n =BERT(t) n )

[0096] e ij =BERT(te ij )

[0097] N j =GAT([N i ,e ij ,f j ])N0=f0

[0098] In this embodiment of the invention, by encoding the text information of each product function node and the text information of each edge in the product function knowledge graph, and combining the encoding results with the relationship between each product function node and each edge in the product function knowledge graph, the determined node feature representation can represent both the node information and the relationship between each node, thus having the advantage of comprehensive expression.

[0099] Based on the above embodiments, the user feedback information collection method provided in this embodiment of the invention, wherein the user feedback information is filled into the product function knowledge graph based on the correlation, specifically includes:

[0100] Select the target product function node with the highest relevance from the product function knowledge graph, and fill the target product function node with the user feedback information.

[0101] Specifically, in this embodiment of the invention, when filling user feedback information into the product function knowledge graph based on relevance, the target product function node with the highest relevance can be selected from the product function knowledge graph first. The mathematical representation of this process is as follows:

[0102]

[0103] Where m is the total number of product function nodes in the product function knowledge graph, and N is... j Let y be the node feature representation of the j-th product function node in the product function knowledge graph, and y be the target product function node.

[0104] Afterwards, user feedback information can be added to the target product feature nodes to enrich the relevant information of the target product feature nodes.

[0105] Based on the above embodiments, the user feedback information collection method provided in this embodiment of the invention, which involves filling the user feedback information into the product function knowledge graph based on the correlation, further includes:

[0106] Identify the target product function nodes with user feedback information in the product function knowledge graph;

[0107] Based on user feedback information at the target product's functional nodes, the functional nodes of the target product are upgraded.

[0108] Specifically, in this embodiment of the invention, after filling the user feedback information into the product function knowledge graph based on relevance, the target product function node with user feedback information in the product function knowledge graph can be determined first. The target product function node can be one or more, depending on the actual situation.

[0109] Then, based on user feedback at the target product's functional nodes, the functional nodes are upgraded. In other words, user feedback provides guidance for developers on upgrading the target product's functional nodes, ensuring that the upgraded nodes meet user needs, reduce negative user feedback, and ultimately improve the user experience.

[0110] Figure 3 This is a flowchart illustrating the implementation process of the user feedback information collection method provided in this embodiment of the invention, such as... Figure 3 As shown, the process includes:

[0111] First, the language information corresponding to the contextual dialogue speech and the contextual dialogue text are input into the feedback information extraction model to obtain the evaluation bias information, evaluation vocabulary information, and user feedback feature representation output by the feedback information extraction model. The language information can be "ch", the contextual dialogue text can be "Help me navigate to XXX->to XXX->Why can't I find this place?", the obtained evaluation bias information can be negative, and the evaluation vocabulary information can include "XXX", "place", and "can't find it".

[0112] Then, the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph 31 is calculated. Figure 3 The node characteristics of each product function node are represented as N1, N2, N3, N4, etc. Subsequently, based on this relevance, user feedback information can be filled into the target product function node with the highest relevance in the product function knowledge graph to update the product function knowledge graph. The updated product function knowledge graph is as follows: Figure 4 As shown.

[0113] In summary, the user feedback information collection method provided in this embodiment of the invention extracts information categories from the user's contextual dialogue text and considers the language information of the contextual dialogue voice, facilitating rapid identification, classification, and extraction of results across different languages ​​and texts. Furthermore, when a user provides feedback again, the contextual dialogue text can be quickly matched with the corpus type and the resulting data analyzed to determine the user's problem. This method has good scalability; subsequent maintenance and upgrades to the product function knowledge graph are sufficient to quickly collect user feedback attitudes and evaluation words for each function. Simultaneously, the use of a multilingual feedback information extraction model is not limited by region or user, accurately collecting user evaluation direction and content. By filling the product function knowledge graph with user feedback information through feature representation relevance, the user experience of product functions can be quickly and in real-time perceived, thereby improving the efficiency of user voice interaction problem feedback and the accuracy of problem matching, indirectly improving the user experience when user feedback problems are resolved.

[0114] like Figure 5 As shown, based on the above embodiments, this embodiment of the invention provides a user feedback information collection device, including:

[0115] The acquisition module 51 is used to acquire the user's contextual dialogue text, which is obtained based on the user's contextual dialogue speech conversion;

[0116] Extraction module 52 is used to determine the user feedback information and user feedback feature representation in the context dialogue text based on the feedback information extraction model;

[0117] The collection module 53 is used to calculate the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and to fill the user feedback information into the product function knowledge graph based on the correlation.

[0118] The feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags.

[0119] Based on the above embodiments, the user feedback information collection device provided in this embodiment of the invention further includes a language information determination module, used for:

[0120] Determine the language information corresponding to the contextual dialogue voice;

[0121] Accordingly, the extraction module is specifically used for:

[0122] The contextual dialogue text and the language information are input into the feedback information extraction model to obtain the user feedback information and the user feedback feature representation output by the feedback information extraction model.

[0123] Based on the above embodiments, the user feedback information collection device provided in this embodiment of the invention includes evaluation tendency information and evaluation vocabulary information;

[0124] The user feedback information collection device further includes a training module, used for:

[0125] Based on the contextual dialogue text, the pre-determined multilingual pre-trained model is jointly trained for evaluation bias classification tasks and evaluation vocabulary extraction tasks to obtain the feedback information extraction model.

[0126] Based on the above embodiments, the user feedback information collection device provided in this embodiment of the invention further includes a feature representation determination module, used for:

[0127] The text information of each product function node and the text information of each edge in the product function knowledge graph are encoded to obtain the encoding result;

[0128] Based on the encoding results and the relationship between each product function node and each edge in the product function knowledge graph, the node feature representation is determined.

[0129] Based on the above embodiments, the user feedback information collection device provided in this embodiment of the invention, wherein the collection module is specifically used for:

[0130] Select the target product function node with the highest relevance from the product function knowledge graph, and fill the target product function node with the user feedback information.

[0131] Based on the above embodiments, the user feedback information collection device provided in this embodiment of the invention further includes an application module, used for:

[0132] Identify the target product function nodes with user feedback information in the product function knowledge graph;

[0133] Based on user feedback information at the target product's functional nodes, the functional nodes of the target product are upgraded.

[0134] Specifically, the functions of each module in the user feedback information collection device provided in this embodiment of the invention correspond one-to-one with the operation flow of each step in the above method-like embodiments, and the achieved effects are also the same. For details, please refer to the above embodiments, and this will not be repeated in this embodiment of the invention.

[0135] like Figure 6 As shown, based on the above embodiments, this embodiment of the invention provides a user feedback system, including: a voice module 61 and a user feedback information collection device 62 provided in the above embodiments; the voice module 61 and the user feedback information collection device 62 are connected.

[0136] The voice module 61 is used to acquire the user's contextual dialogue voice, convert the contextual dialogue voice into contextual dialogue text, determine the response information corresponding to the contextual dialogue voice, and provide the response information to the user.

[0137] The voice module 61 is also used to send contextual dialogue voice and / or contextual dialogue text to the user feedback information collection device.

[0138] Specifically, in this embodiment of the invention, the user feedback system can be an in-vehicle intelligent voice system, used to respond to the user's voice requests, that is, to find the response information corresponding to the user's contextual dialogue voice and provide it to the user. This function is implemented through the voice module in the user feedback system. Simultaneously, the user feedback system also includes a user feedback information collection device. The voice module can send the contextual dialogue voice and / or contextual dialogue text to the user feedback information collection device. After receiving the contextual dialogue voice and / or contextual dialogue text, the user feedback information collection device can first convert the contextual dialogue voice into contextual dialogue text. Alternatively, it can directly use the contextual dialogue text to execute the user feedback information collection method, thereby realizing the collection of user feedback information and updating the product function knowledge graph. In this way, the user feedback system can have the function of collecting user feedback information.

[0139] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include: a processor 710, a communications interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communications interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute the user feedback information collection method provided in the above embodiments. This method includes: acquiring the user's contextual dialogue text, which is obtained based on the user's contextual dialogue speech conversion; determining the user feedback information and user feedback feature representation in the contextual dialogue text based on a feedback information extraction model; calculating the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and filling the user feedback information into the product function knowledge graph based on the correlation; wherein the feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags.

[0140] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0141] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the user feedback information collection method provided in the above embodiments. The method includes: acquiring the user's contextual dialogue text, which is obtained based on the user's contextual dialogue speech conversion; determining the user feedback information and user feedback feature representation in the contextual dialogue text based on a feedback information extraction model; calculating the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and filling the user feedback information into the product function knowledge graph based on the correlation; wherein the feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags.

[0142] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the user feedback information collection method provided in the above embodiments. The method includes: acquiring the user's contextual dialogue text, the contextual dialogue text being obtained based on the user's contextual dialogue speech conversion; determining user feedback information and user feedback feature representation in the contextual dialogue text based on a feedback information extraction model; calculating the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and filling the user feedback information into the product function knowledge graph based on the correlation; wherein the feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags.

[0143] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0144] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for collecting user feedback information, characterized in that, include: Obtain the user's contextual dialogue text, which is obtained based on the user's contextual dialogue speech conversion; Based on the feedback information extraction model, the user feedback information and user feedback feature representation in the contextual dialogue text are determined. The user feedback feature representation is the contextual dialogue aggregation feature of the feedback information extraction model in the process of obtaining the user feedback information, which is used to characterize the deep features of the user feedback information. The correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph is calculated, and the user feedback information is filled into the product function knowledge graph based on the correlation. The product function knowledge graph includes product function nodes at multiple levels, and adjacent product function nodes are connected by edges representing hierarchical relationships. The node feature representation is extracted by inputting the product function knowledge graph into a graph attention network. The feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags. The node feature representation is determined based on the following method: The text information of each product function node and the text information of each edge in the product function knowledge graph are encoded through an encoding network to obtain the encoding result. The encoding results and the relationships between each product function node and each edge in the product function knowledge graph are input into the graph attention network to determine the node feature representation.

2. The user feedback information collection method according to claim 1, characterized in that, The step of determining user feedback information and user feedback feature representation in the contextual dialogue text based on the feedback information extraction model includes: Determine the language information corresponding to the contextual dialogue voice; Accordingly, the step of determining the user feedback information and user feedback feature representation in the contextual dialogue text based on the feedback information extraction model specifically includes: The contextual dialogue text and the language information are input into the feedback information extraction model to obtain the user feedback information and the user feedback feature representation output by the feedback information extraction model.

3. The user feedback information collection method according to claim 2, characterized in that, The user feedback information includes evaluation bias information and evaluation vocabulary information; The feedback information extraction model is trained based on the following method: Based on the contextual dialogue text, the pre-determined multilingual pre-trained model is jointly trained for evaluation bias classification tasks and evaluation vocabulary extraction tasks to obtain the feedback information extraction model.

4. The user feedback information collection method according to claim 1, characterized in that, The step of filling the user feedback information into the product function knowledge graph based on the correlation specifically includes: Select the target product function node with the highest relevance from the product function knowledge graph, and fill the target product function node with the user feedback information.

5. The user feedback information collection method according to any one of claims 1-4, characterized in that, Based on the correlation, the user feedback information is filled into the product function knowledge graph, followed by: Identify the target product function nodes with user feedback information in the product function knowledge graph; Based on user feedback information at the target product's functional nodes, the functional nodes of the target product are upgraded.

6. A user feedback information collection device, characterized in that, include: The acquisition module is used to acquire the user's contextual dialogue text, which is obtained based on the user's contextual dialogue speech conversion; The extraction module is used to extract user feedback information and user feedback feature representation in the context dialogue text based on the feedback information extraction model. The user feedback feature representation is the contextual dialogue aggregation feature of the feedback information extraction model in the process of obtaining the user feedback information, which is used to characterize the deep features of the user feedback information. The collection module is used to calculate the correlation between the user feedback feature representation and the node feature representation of each product function node in the product function knowledge graph, and to fill the user feedback information into the product function knowledge graph based on the correlation; the product function knowledge graph includes product function nodes at multiple levels, and adjacent product function nodes are connected by edges representing hierarchical relationships; the node feature representation is extracted by inputting the product function knowledge graph into a graph attention network; The feedback information extraction model is trained based on contextual dialogue text samples carrying user feedback information tags. It also includes a feature representation determination module, used for: The text information of each product function node and the text information of each edge in the product function knowledge graph are encoded through an encoding network to obtain the encoding result. The encoding results and the relationships between each product function node and each edge in the product function knowledge graph are input into the graph attention network to determine the node feature representation.

7. A user feedback system, characterized in that, include: The voice module and the user feedback information collection device as described in claim 6; The voice module and the user feedback information collection device are connected; The voice module is used to acquire the user's contextual dialogue voice, convert the contextual dialogue voice into contextual dialogue text, determine the response information corresponding to the contextual dialogue voice, and provide the response information to the user. The voice module is also used to send the contextual dialogue voice and / or the contextual dialogue text to the user feedback information collection device.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the user feedback information collection method as described in any one of claims 1-5.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the user feedback information collection method as described in any one of claims 1-5.

Citation Information

Patent Citations

  • Network complaint intelligent delimitation method and system

    CN108776900A

  • Poem recommendation method based on combination of LDA topic model and poem knowledge graph

    CN113127627A

  • Vehicle user feedback information collection method and device

    CN114999024A

  • Speech recognition method, related device, electronic equipment and storage medium

    CN115019773A