Method and apparatus for generating a customer service consultation abstract

Abstract

This invention discloses a method and apparatus for generating customer service consultation summaries, relating to the field of computer technology. One specific embodiment of the method includes: decomposing the customer service consultation content into at least one consultation task; for each consultation task, determining the form data corresponding to the consultation task, and extracting features from the consultation task to obtain a consultation task vector; embedding the consultation task vector and the form data into a pre-trained graph model to obtain an embedded graph model, and extracting features from the embedded graph model to obtain a graph model embedding vector; obtaining the consultation question and key summary information based on the graph model embedding vector; and generating a customer service consultation summary based on the consultation question and the key summary information. This embodiment can automatically generate summaries based on multiple features involved in the customer service consultation process, resulting in objective, standardized, and highly efficient summary generation.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and apparatus for generating customer service inquiry summaries. Background Technology

[0002] In the e-commerce sector, tens of millions of online customer service inquiries are initiated daily. These include user questions about sizes, product expiration dates, packaging, and after-sales service, such as returns, exchanges, invoices, and shipping times. This complex communication scenario results in individual conversations ranging from hundreds to thousands of words. Without an effective summary of user inquiries, recording and resolving user issues becomes significantly more cumbersome and costly. Currently, methods for generating summaries of online customer service inquiries primarily include manual methods and intelligent text-based methods such as template-based summaries.

[0003] In the process of realizing this invention, the inventors discovered at least the following problems in the prior art:

[0004] Generating consultation summaries manually is subjective, and different people may summarize them in different ways. Generating consultation summaries using templates or other intelligent text-based methods mainly relies on the user's consultation text, resulting in low feature utilization. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method and apparatus for generating customer service consultation summaries, which can automatically generate summaries based on multiple features involved in the customer service consultation process. The summary generation results are objective, standardized, and highly efficient.

[0006] To achieve the above objectives, according to one aspect of the present invention, a method for generating customer service inquiry summaries is provided.

[0007] A method for generating customer service inquiry summaries includes:

[0008] Each customer service inquiry is broken down into dialogue segments to obtain at least one inquiry task.

[0009] For each consultation task, the form data corresponding to the consultation task is determined, and feature extraction is performed on the consultation task to obtain a consultation task vector;

[0010] The consultation task vector and the form data are embedded into a pre-trained graph model to obtain an embedded graph model, and features are extracted from the embedded graph model to obtain a graph model embedding vector.

[0011] The consultation question and key summary information are obtained based on the graph model embedding vectors.

[0012] A customer service inquiry summary is generated based on the inquiry question and the key summary information.

[0013] Optionally, breaking down customer service inquiries into dialogue segments includes:

[0014] For each sentence in the customer service consultation content, the dialogue relevance is calculated based on the association relationship between the sentence and adjacent sentences, where adjacent sentences refer to sentences that are adjacent to the sentence in the context of the customer service consultation content;

[0015] The dialogue is decomposed based on the dialogue relevance and a preset threshold.

[0016] Optionally, the consultation task vector and the form data are embedded into a pre-trained graph model to obtain an embedded graph model, including:

[0017] Determine the positions of the consultation task vector and the form data in the graph model, and embed the consultation task vector and the form data into the graph model according to the positions;

[0018] The number of nodes in the embedded graph model will be expanded according to the set number of windows;

[0019] For the data in each window, the number of co-occurrences between different nodes is calculated as the weight of the edges between the nodes, thus obtaining the embedding graph model.

[0020] Optionally, the consultation question and key summary information are obtained based on the graph model embedding vector, including:

[0021] The consultation question is obtained by performing graph network association based on the graph model embedding vectors.

[0022] Key summary information is generated based on the graph model embedding vectors and pre-trained dialogue vectors.

[0023] Optionally, after obtaining the graph model embedding vector, the following steps are also included:

[0024] A centrally symmetric matrix is ​​established based on the relationships between the nodes of the embedded graph model;

[0025] Furthermore, the graph network association based on the graph model embedding vectors to obtain the consultation question includes:

[0026] Perform format transformation on the graph model embedding vector;

[0027] The format-transformed graphical model embedding vector is concatenated with the central symmetric matrix to obtain the concatenated vector;

[0028] The score of the hidden layer in the embedded graph model is obtained by multiplying the concatenated vector with a preset weight;

[0029] The consultation question is determined based on the score of the hidden layer.

[0030] Optionally, after generating a customer service consultation summary based on the consultation question and the key summary information, the method further includes:

[0031] The system compares and learns with the customer service inquiry summary and preset summary content to obtain summary content with a similarity exceeding the similarity threshold.

[0032] The customer service response corresponding to the summary content will be used as the response result of the consultation task corresponding to the customer service consultation summary and fed back to the user.

[0033] Optionally, after obtaining the consultation question and key summary information based on the graph model embedding vector, the method further includes:

[0034] Generate a follow-up judgment result, which is used to determine whether the consultation task needs to be followed up;

[0035] Furthermore, after using the customer service response result corresponding to the summary content as the response result of the consultation task corresponding to the customer service consultation summary and feeding it back to the user, it also includes:

[0036] The consultation tasks corresponding to the customer service consultation summary set do not need to be tracked.

[0037] According to another aspect of the present invention, an apparatus for generating customer service inquiry summaries is provided.

[0038] An apparatus for generating customer service inquiry summaries, comprising:

[0039] The dialogue decomposition module is used to decompose customer service inquiries into at least one inquiry task.

[0040] The task feature extraction module is used to determine the form data corresponding to each consultation task and extract features from the consultation task to obtain a consultation task vector.

[0041] The task embedding module is used to embed the consultation task vector and the form data into a pre-trained graph model to obtain an embedded graph model, and to extract features from the embedded graph model to obtain a graph model embedding vector.

[0042] An embedded vector calculation module is used to obtain consultation questions and key summary information based on the embedded vectors of the graph model.

[0043] The summary generation module is used to generate a customer service consultation summary based on the consultation question and the key summary information.

[0044] According to another aspect of the present invention, an electronic device for generating customer service inquiry summaries is provided.

[0045] An electronic device for generating customer service inquiry summaries includes: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the method for generating customer service inquiry summaries provided in the embodiments of the present invention.

[0046] According to another aspect of the present invention, a computer-readable medium is provided.

[0047] A computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method for generating customer service inquiry summaries provided in embodiments of the present invention.

[0048] One embodiment of the above invention has the following advantages or beneficial effects: At least one consultation task is obtained by decomposing the customer service consultation content into dialogue; for each consultation task, the corresponding form data is determined, and feature extraction is performed on the consultation task to obtain a consultation task vector; the consultation task vector and form data are embedded into a pre-trained graph model to obtain an embedded graph model, and feature extraction is performed on the embedded graph model to obtain a graph model embedding vector; the consultation question and key summary information are obtained based on the graph model embedding vector; a customer service consultation summary is generated based on the consultation question and key summary information. This method effectively combines, associates, and summarizes the user's consultation task vector with the user's characteristics, text features, and basic product attributes related to the consultation task vector, thus solving the problem of low feature utilization when generating customer service consultation summaries. Subsequently, when a new question arises, its customer service consultation content can be cleverly associated with the graph model, achieving template-based processing. This solves the problems of subjectivity and inconsistent results in manually generated customer service consultation summaries, realizing automated summary generation based on multiple features involved in the customer service consultation process. The summary generation results are objective, standardized, and highly efficient. At the same time, by using a comparative learning approach, users' actual problems can be effectively summarized through comparison. By dynamically comparing and segmenting historical consultation nodes, both the unification of related problems and the discovery of new problems can be fully automated and follow a standard format.

[0049] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0050] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein:

[0051] Figure 1 This is a schematic diagram illustrating the main steps of a method for generating customer service inquiry summaries according to an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of the implementation process of an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the main modules of an apparatus for generating customer service inquiry summaries according to an embodiment of the present invention;

[0054] Figure 4 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied;

[0055] Figure 5 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Detailed Implementation

[0056] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0057] The inventor analyzed existing technologies and found that there are currently two methods that can be used to automatically generate summaries of online customer service inquiries:

[0058] (1) Manual method. This method mainly relies on manual compilation. After a user inquires about a problem, regardless of whether the problem has been resolved, the content of the chat is summarized manually, a consultation summary is filled out, and follow-up is carried out.

[0059] (2) Text-based intelligent approach. This approach mainly relies on text-related algorithms. By using text analysis, text generation and other related algorithms to summarize the content, or by directly using a fixed format template to extract the required content and automatically generate it based on the set template.

[0060] Regarding the two existing solutions mentioned above, the inventors have found the following two main drawbacks:

[0061] (1) Weak inductive reasoning

[0062] Technical solutions proactively summarized by customer service staff are often difficult to standardize due to the inclusion of personal opinions. This can lead to misunderstandings when other personnel, such as customer service staff and delivery drivers, or delivery drivers and purchasing staff, need to work together, requiring them to review the chat logs again. Even among customer service staff, some use concise language, while others handle different parts, making coordination difficult. While template-based intelligent text processing can alleviate these problems, e-commerce inquiries encompass tens of thousands of types, such as invoices, price protection, returns, and shipping, making template creation complex and difficult to categorize.

[0063] (2) Low feature utilization

[0064] When customers inquire about something, they are invariably focused on the product; the core of the question lies in the inquiry about a specific product. Current methods primarily rely on the user's inquiry text. Although the product SKU (SKU refers to a single product; each product has one SKU for easy identification by e-commerce brands) is sent during the inquiry, the summary of the inquiry is still not based on the product itself. This is disastrous for later maintenance, product tracking, and handling of frequently asked questions about the product.

[0065] To address the aforementioned technical problems in existing technologies, this invention provides a method and apparatus for generating customer service consultation summaries. Addressing the issue of weak inductive reasoning, it primarily employs a comparative learning approach. This method effectively summarizes users' actual questions through comparison, dynamically comparing and analyzing historical consultation nodes. Whether unifying related questions or discovering new questions, it is fully automated and follows a standard format. Addressing the issue of low feature utilization, it effectively combines user consultations, user characteristics, textual features, and basic product attributes. It cleverly utilizes graph representation to correlate and summarize all features. Then, when new questions arise, their consultations are cleverly associated with the graph, achieving template-based processing.

[0066] Figure 1 This is a schematic diagram illustrating the main steps of a method for generating customer service inquiry summaries according to an embodiment of the present invention. Figure 1 As shown, the method for generating customer service consultation summaries in this embodiment of the invention mainly includes the following steps S101 to S105.

[0067] Step S101: Deconstruct the customer service inquiry content to obtain at least one inquiry task;

[0068] Step S102: For each consultation task, determine the form data corresponding to the consultation task, and perform feature extraction on the consultation task to obtain the consultation task vector;

[0069] Step S103: Embed the consultation task vector and the form data into a pre-trained graph model to obtain an embedded graph model, and extract features from the embedded graph model to obtain a graph model embedding vector;

[0070] Step S104: Obtain the consultation question and key summary information based on the graph model embedding vector;

[0071] Step S105: Generate a customer service consultation summary based on the consultation question and the key summary information.

[0072] According to an embodiment of the present invention, the customer service consultation content refers to the specific conversation content when a user consults through online customer service, mainly including the chat content and hyperlinks when the user connects. The conversation between the user and customer service can be distinguished based on the customer service consultation content in JSON format stored in the backend. Each message sent by the user and customer service records a unique identifier and indicates who sent it; therefore, the conversation between the user and customer service can be directly distinguished based on the customer service consultation content in JSON format.

[0073] Dialogue decomposition primarily involves breaking down the user's inquiry into different inquiry tasks based on the user's expression. Specifically, dialogue decomposition can be based on the correlation between conversations. In one embodiment of the present invention, dialogue decomposition of customer service inquiry content may specifically include:

[0074] For each sentence in the customer service consultation content, the dialogue relevance is calculated based on the association relationship between the sentence and adjacent sentences, where adjacent sentences refer to sentences that are adjacent to the sentence in the context of the customer service consultation content;

[0075] The dialogue is decomposed based on the dialogue relevance and a preset threshold.

[0076] Specifically, dialogue relevance can be calculated using the following formula:

[0077]

[0078] Where A and B refer to the user and customer service respectively, d n (A, B) represents the relevance of the customer service consultation content in the dialogue, i and j are the number of dialogue lines respectively, n is the total number of dialogue lines included in the customer service consultation content, U i D represents the content of the i-th row. j This represents the content of line j, where Distance represents the current computing session U. i and D j The correlation is represented by cos, where cos represents the distance after representing the session vector, where 01 represents a negative correlation between up and down, 10 represents a negative correlation between down and up, and 11 represents a positive correlation between up and down. Taking U as an example...01 For example, it represents the dialogue U in the i-th row. i The dialogue U in the (i+1)th row adjacent to it i+1 It is a negative correlation; with U 10 For example, it represents the dialogue U in the i-th row. i The dialogue U in the (i-1)th row adjacent to it i-1 It is a negative correlation; with U 11 For example, it represents the dialogue U in the i-th row. i The dialogue U in the (i+1)th row adjacent to it i+1 A positive correlation indicates a strong connection between the two sentences, while a negative correlation indicates a weak or no connection. When calculating the correlation between two sentences in a dialogue, common text and semantic analysis methods can be used. For example, feature vectors can be extracted from both sentences, and then cosine similarity can be calculated based on these feature vectors to characterize the correlation between the two sentences.

[0079] After calculating the dialogue relevance of customer service inquiries, the dialogue can be broken down according to pre-set thresholds to obtain at least one inquiry task, with each task corresponding to multiple dialogue lines. For each inquiry task, the user's historical forms are queried based on the content of the user's inquiries, mainly including orders, after-sales orders, and online inquiry records, and these are summarized into the user's basic information, which is then stored in the user's basic information database.

[0080] Next, for each consultation task, its relevant information is embedded into a pre-trained graph model. The graph model primarily relies on a user's basic information database, an order database, and a product database. When constructing the graph model, for example, the product's SKU (Stock Keeping Unit, the smallest unit of inventory measurement; in e-commerce, it generally refers to a single product) is used as the primary key. Relevant information is aggregated and linked to create a graph that associates users and products. Each product has a unique SKU and SPU (Standard Product Unit, the smallest unit for aggregating product information; products with the same attribute values ​​and characteristics can form a single SPU). The SKU needs to be recorded under the SPU. This is because if a user only selects different product models, their expected purchase intent remains the same.

[0081] When embedding consultation tasks, feature extraction is first required to obtain a consultation task vector. After decomposing customer service consultation content into consultation tasks, for each consultation task, that consultation task is used as a positive sample, and other consultation tasks are used as negative samples. Feature extraction is performed on the positive and negative samples through sentence pre-training to obtain the corresponding consultation task vector. For example, Albert pre-training can be used to convert the sentences into vectors.

[0082] According to one embodiment of the present invention, embedding the consultation task vector and the form data into a pre-trained graph model to obtain an embedded graph model may specifically include:

[0083] Determine the positions of the consultation task vector and the form data in the graph model, and embed the consultation task vector and the form data into the graph model according to the positions;

[0084] The number of nodes in the embedded graph model will be expanded according to the set number of windows;

[0085] For the data in each window, the number of co-occurrences between different nodes is calculated as the weight of the edges between the nodes, thus obtaining the embedding graph model.

[0086] Specifically, the form data corresponding to the consultation task (e.g., order information, after-sales order information, etc.) is embedded into the consultation task vector using a graph method. The embedding method primarily relies on the previously generated graph model. For the user's form data and consultation task vector, their positions in the graph model are directly calculated. Then, the nodes are expanded according to a set number of windows (i.e., the number of adjacent nodes in the associated graph model). For the data in each window, the co-occurrence frequency (i.e., the number of times the same node appears) between different nodes is calculated as a weight, resulting in the embedded graph model. Subsequently, based on the relationships between nodes in the graph model, a central symmetric matrix and the graph embedding vector are established. Each element in the graph embedding vector represents the number of times a node appears in the central symmetric matrix.

[0087] According to one embodiment of the present invention, obtaining consultation questions and key summary information based on the graph model embedding vector may specifically include: performing graph network association based on the graph model embedding vector to obtain the consultation question; and generating key summary information based on the graph model embedding vector and pre-trained dialogue vectors.

[0088] In one embodiment of the present invention, after obtaining the graph model embedding vector, the method further includes: establishing a central symmetric matrix based on the relationships between nodes in the embedded graph model. Furthermore, performing graph network association based on the graph model embedding vector to obtain the consultation question includes: transforming the format of the graph model embedding vector; concatenating the transformed graph model embedding vector with the central symmetric matrix to obtain a concatenated vector; multiplying the concatenated vector by preset weights to obtain the score of the hidden layer in the embedded graph model; and determining the consultation question based on the score of the hidden layer.

[0089] In embodiments of this invention, graph network association is primarily performed by calculating the score representation of the hidden layers of the network based on the graph model embedding vectors. First, the graph model embedding vectors are format-transformed to allow concatenation with a centrally symmetric matrix. Then, they are concatenated with the centrally symmetric matrix, and the result multiplied by the weights is used as the graph network embedding score. The purpose of this process is mainly to infer the specific questions a user needs regarding a particular product by combining product and order information, such as price, discounts, and shelf life, and to include the results as part of a summary.

[0090] According to another embodiment of the present invention, when generating key summary information based on the graph model embedding vector and pre-trained dialogue vectors, the pre-trained dialogue vectors are, for example, dialogue vectors for different consultation tasks obtained by grouping and training different consultation tasks. By inputting the graph model embedding vectors and the pre-trained dialogue vectors into a text generation model, key summary information can be generated based on the content of the consultation task. Furthermore, after obtaining the consultation question and key summary information based on the graph model embedding vectors, the method further includes: generating a follow-up judgment result, which is used to determine whether the consultation task needs to be tracked. Additionally, a summary for each consultation task and a customer service consultation summary can be generated based on the key summary information and the consultation question, and the summaries for each consultation task and the customer service consultation summary are standardized to generate a standardized summary.

[0091] According to another embodiment of the present invention, after generating a customer service consultation summary based on the consultation question and the key summary information, the method further includes: performing comparative learning on the customer service consultation summary and preset summary content to obtain summary content with a similarity exceeding a similarity threshold; using the customer service response result corresponding to the summary content as the response result of the consultation task corresponding to the customer service consultation summary, and feeding it back to the user. Based on the aforementioned steps S101 to S105, a customer service consultation summary has been obtained, including the consultation goal and possible solutions of the customer service consultation. Here, the comparison learning between the customer service consultation summary and preset summary content can be used to determine whether it conforms to the consultation intent of a certain consultation dialogue content. Specifically, a similarity threshold can be set to determine whether it conforms to the consultation intent. If the set similarity threshold is exceeded, it can be determined that the model has provided an appropriate solution, and the customer service response result of the corresponding summary content can be used as the response result of the consultation task corresponding to the customer service consultation summary, and fed back to the user. Furthermore, after using the customer service response result corresponding to the summary content as the response result of the consultation task corresponding to the customer service consultation summary and feeding it back to the user, the consultation task corresponding to the customer service consultation summary can also be set not to be tracked.

[0092] The implementation process of the present invention will be described below with reference to specific embodiments. Figure 2 This is a schematic diagram of the implementation process of an embodiment of the present invention, as shown below. Figure 2As shown, the system for generating customer service consultation summaries according to the present invention can be mainly divided into three layers: a user consultation layer, a model calculation layer, and a system storage layer. The system storage layer is mainly used to process raw and intermediate data; the model calculation layer mainly performs separate calculations and processing on the data to output the required items; and the user consultation layer can receive user input and requests as needed. Specifically, the user consultation layer mainly processes the consultation content of online customer service users, converting it into model-recognizable data for further calculation. This mainly includes breaking down the customer service consultation content into dialogue segments and determining the corresponding form data for each consultation task for form data aggregation. The system storage layer mainly stores relevant library information and form information, as well as intermediately generated summaries and order tracking information. It is worth noting that if customer service personnel have modified the relevant data, it can be synchronized to the model calculation layer to correct model parameters and thresholds. The model calculation layer mainly calculates based on the input data and outputs different requirement results. Specifically, this includes: generating a graph model based on data from the product database, order database, and user database (i.e., the user's basic information database) in the system storage layer; embedding the customer service consultation content into the graph model to obtain the embedded graph model, thereby obtaining the graph model embedding vector and the central symmetric matrix; performing graph network association based on the graph model embedding vector and the central symmetric matrix to obtain the user's consultation question, and simultaneously inputting the graph model embedding vector into the task discrimination network to obtain key summary information and determine whether the task needs to be tracked; based on the consultation question and key summary information, the customer service consultation summary corresponding to the customer service consultation content can be obtained; then, joint calculation of relationship scores is performed, and the customer service consultation summary is compared and learned with the preset summary content to obtain the consultation task response result and the order tracking discrimination result. Based on the response result, the response can be queried, and based on the order tracking discrimination result, it can be determined whether the work order needs to be tracked, thereby identifying the work orders that need to be tracked.

[0093] Figure 3 This is a schematic diagram of the main modules of an apparatus for generating customer service inquiry summaries according to an embodiment of the present invention. Figure 3 As shown, the device 300 for generating customer service consultation summaries in this embodiment of the invention mainly includes a dialogue decomposition module 301, a task feature extraction module 302, a task embedding module 303, an embedding vector calculation module 304, and a summary generation module 305.

[0094] The dialogue decomposition module 301 is used to decompose customer service inquiries into at least one inquiry task.

[0095] The task feature extraction module 302 is used to determine the form data corresponding to each consultation task and extract features from the consultation task to obtain a consultation task vector.

[0096] The task embedding module 303 is used to embed the consultation task vector and the form data into a pre-trained graph model to obtain an embedded graph model, and to extract features from the embedded graph model to obtain a graph model embedding vector.

[0097] The embedding vector calculation module 304 is used to obtain the consultation question and key summary information based on the embedding vector of the graph model.

[0098] The summary generation module 305 is used to generate a customer service consultation summary based on the consultation question and the key summary information.

[0099] According to one embodiment of the present invention, the dialogue disassembly module 301 can also be used for:

[0100] For each sentence in the customer service consultation content, the dialogue relevance is calculated based on the association relationship between the sentence and adjacent sentences, where adjacent sentences refer to sentences that are adjacent to the sentence in the context of the customer service consultation content;

[0101] The dialogue is decomposed based on the dialogue relevance and a preset threshold.

[0102] According to another embodiment of the present invention, the task embedding module 303 can also be used for:

[0103] Determine the positions of the consultation task vector and the form data in the graph model, and embed the consultation task vector and the form data into the graph model according to the positions;

[0104] The number of nodes in the embedded graph model will be expanded according to the set number of windows;

[0105] For the data in each window, the number of co-occurrences between different nodes is calculated as the weight of the edges between the nodes, thus obtaining the embedding graph model.

[0106] According to another embodiment of the present invention, the embedded vector calculation module 304 can also be used for:

[0107] The consultation question is obtained by performing graph network association based on the graph model embedding vectors.

[0108] Key summary information is generated based on the graph model embedding vectors and pre-trained dialogue vectors.

[0109] According to another embodiment of the present invention, the task embedding module 303 can also be used for:

[0110] After obtaining the graph model embedding vector, a centrally symmetric matrix is ​​established based on the relationships between the nodes of the embedded graph model;

[0111] Furthermore, the embedded vector computation module 304 can also be used for:

[0112] Perform format transformation on the graph model embedding vector;

[0113] The format-transformed graphical model embedding vector is concatenated with the central symmetric matrix to obtain the concatenated vector;

[0114] The score of the hidden layer in the embedded graph model is obtained by multiplying the concatenated vector with a preset weight;

[0115] The consultation question is determined based on the score of the hidden layer.

[0116] According to another embodiment of the present invention, the apparatus 300 for generating customer service inquiry summaries may further include an inquiry response module (not shown in the figure), used for:

[0117] After generating a customer service consultation summary based on the consultation question and the key summary information, the customer service consultation summary is compared with the preset summary content to obtain summary content with a similarity exceeding the similarity threshold.

[0118] The customer service response corresponding to the summary content will be used as the response result of the consultation task corresponding to the customer service consultation summary and fed back to the user.

[0119] According to another embodiment of the present invention, the embedded vector calculation module 304 can also be used for:

[0120] Generate a follow-up judgment result, which is used to determine whether the consultation task needs to be followed up;

[0121] Furthermore, the consultation response module (not shown in the diagram) can also be used for:

[0122] After the customer service response result corresponding to the summary content is used as the response result of the consultation task corresponding to the customer service consultation summary and fed back to the user, the consultation task corresponding to the customer service consultation summary is set to not need to be tracked.

[0123] According to the technical solution of the present invention, at least one consultation task is obtained by decomposing the customer service consultation content into dialogue; for each consultation task, the form data corresponding to the consultation task is determined, and the consultation task is feature extracted to obtain a consultation task vector; the consultation task vector and form data are embedded into a pre-trained graph model to obtain an embedded graph model, and the embedded graph model is feature extracted to obtain a graph model embedding vector; the consultation question and key summary information are obtained according to the graph model embedding vector; a customer service consultation summary is generated according to the consultation question and key summary information. The graph representation method can effectively combine, associate, and summarize the user's consultation task vector with the user's characteristics, text features, and basic product attributes related to the consultation task vector, thereby solving the problem of low feature utilization when generating customer service consultation summaries. Then, when a new question appears, its customer service consultation content can be cleverly associated with the graph model, achieving template-based processing. This solves the problems of subjectivity and inconsistent results in manually generated customer service consultation summaries, realizing automated summary generation based on multiple features involved in the customer service consultation process. The summary generation results are objective, standardized, and highly efficient. At the same time, by using a comparative learning approach, users' actual problems can be effectively summarized through comparison. By dynamically comparing and segmenting historical consultation nodes, both the unification of related problems and the discovery of new problems can be fully automated and follow a standard format.

[0124] Figure 4 An exemplary system architecture 400 is shown, which can be applied to a method or apparatus for generating customer service inquiry summaries according to embodiments of the present invention.

[0125] like Figure 4 As shown, system architecture 400 may include terminal devices 401, 402, and 403, a network 404, and a server 405. Network 404 serves as the medium for providing communication links between terminal devices 401, 402, and 403 and server 405. Network 404 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0126] Users can use terminal devices 401, 402, and 403 to interact with server 405 via network 404 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 401, 402, and 403, such as shopping applications, web browser applications, e-commerce applications, customer service tools, etc. (for example only).

[0127] Terminal devices 401, 402, and 403 can be various electronic devices with displays that support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0128] Server 405 can be a server providing various services, such as a backend management server supporting customer service inquiries sent by users using terminal devices 401, 402, and 403 (this is just an example). The backend management server can process the received customer service inquiry summary generation request and other data by deconstructing the customer service inquiry content into at least one inquiry task; for each inquiry task, determine the corresponding form data and extract features from the inquiry task to obtain an inquiry task vector; embed the inquiry task vector and the form data into a pre-trained graph model to obtain an embedded graph model, and extract features from the embedded graph model to obtain a graph model embedding vector; obtain the inquiry question and key summary information based on the graph model embedding vector; generate a customer service inquiry summary based on the inquiry question and the key summary information, and feed back the processing result (e.g., the generated customer service inquiry summary—this is just an example) to the terminal device.

[0129] It should be noted that the method for generating customer service inquiry summaries provided in this embodiment of the invention is generally executed by server 405, and correspondingly, the device for generating customer service inquiry summaries is generally located in server 405.

[0130] It should be understood that Figure 4 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0131] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer system 500 suitable for implementing terminal devices or servers of the present invention. Figure 5 The terminal device or server shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0132] like Figure 5 As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 502 or programs loaded from storage section 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the system 500. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0133] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 510 as needed so that computer programs read from it can be installed into storage section 508 as needed.

[0134] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs the functions defined above in the system of this invention.

[0135] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can 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 a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, 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 device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0136] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0137] The units or modules described in the embodiments of the present invention can be implemented in software or hardware. The described units or modules can also be housed in a processor; for example, a processor can be described as including a dialogue decomposition module, a task feature extraction module, a task embedding module, an embedding vector calculation module, and a summary generation module. The names of these units or modules do not necessarily limit the specific unit or module itself; for example, a dialogue decomposition module can also be described as "a module for decomposing customer service consultation content into dialogue to obtain at least one consultation task."

[0138] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs, which, when executed by the device, cause the device to include: deconstructing customer service consultation content into at least one consultation task; for each consultation task, determining the form data corresponding to the consultation task, and extracting features from the consultation task to obtain a consultation task vector; embedding the consultation task vector and the form data into a pre-trained graph model to obtain an embedded graph model, and extracting features from the embedded graph model to obtain a graph model embedding vector; obtaining consultation questions and key summary information based on the graph model embedding vector; and generating a customer service consultation summary based on the consultation questions and the key summary information.

[0139] According to the technical solution of the present invention, at least one consultation task is obtained by decomposing the customer service consultation content into dialogue; for each consultation task, the form data corresponding to the consultation task is determined, and the consultation task is feature extracted to obtain a consultation task vector; the consultation task vector and form data are embedded into a pre-trained graph model to obtain an embedded graph model, and the embedded graph model is feature extracted to obtain a graph model embedding vector; the consultation question and key summary information are obtained according to the graph model embedding vector; a customer service consultation summary is generated according to the consultation question and key summary information. The graph representation method can effectively combine, associate, and summarize the user's consultation task vector with the user's characteristics, text features, and basic product attributes related to the consultation task vector, thereby solving the problem of low feature utilization when generating customer service consultation summaries. Then, when a new question appears, its customer service consultation content can be cleverly associated with the graph model, achieving template-based processing. This solves the problems of subjectivity and inconsistent results in manually generated customer service consultation summaries, realizing automated summary generation based on multiple features involved in the customer service consultation process. The summary generation results are objective, standardized, and highly efficient. At the same time, by using a comparative learning approach, users' actual problems can be effectively summarized through comparison. By dynamically comparing and segmenting historical consultation nodes, both the unification of related problems and the discovery of new problems can be fully automated and follow a standard format.

[0140] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method of generating a customer care consultation summary, the method comprising: include: Each customer service inquiry is broken down into dialogue segments to obtain at least one inquiry task. For each consultation task, the form data corresponding to the consultation task is determined, and feature extraction is performed on the consultation task to obtain a consultation task vector; The consultation task vector and the form data are embedded into a pre-trained graph model to obtain an embedded graph model, and features are extracted from the embedded graph model to obtain a graph model embedding vector. A centrally symmetric matrix is ​​established based on the relationships between the nodes of the embedded graph model; Obtaining consultation questions and key summary information based on the graph model embedding vector includes: transforming the format of the graph model embedding vector; concatenating the transformed graph model embedding vector with the central symmetric matrix to obtain a concatenated vector; multiplying the concatenated vector with preset weights to obtain the score of the hidden layer in the embedded graph model; determining the consultation question based on the score of the hidden layer; and generating key summary information based on the graph model embedding vector and pre-trained dialogue vectors. A customer service inquiry summary is generated based on the inquiry question and the key summary information.

2. The method of claim 1, wherein, Breaking down customer service inquiries into dialogues includes: For each sentence in the customer service consultation content, the dialogue relevance is calculated based on the association relationship between the sentence and adjacent sentences, where adjacent sentences refer to sentences that are adjacent to the sentence in the context of the customer service consultation content; The dialogue is decomposed based on the dialogue relevance and a preset threshold.

3. The method of claim 1, wherein, Embedding the consultation task vector and the form data into a pre-trained graph model yields an embedded graph model, including: Determine the positions of the consultation task vector and the form data in the graph model, and embed the consultation task vector and the form data into the graph model according to the positions; The number of nodes in the embedded graph model will be expanded according to the set number of windows; For the data in each window, the number of co-occurrences between different nodes is calculated as the weight of the edges between the nodes, thus obtaining the embedding graph model.

4. The method of claim 1, wherein, After generating a customer service consultation summary based on the consultation question and the key summary information, the process also includes: The system compares and learns with the customer service inquiry summary and preset summary content to obtain summary content with a similarity exceeding the similarity threshold. The customer service response corresponding to the summary content will be used as the response result of the consultation task corresponding to the customer service consultation summary and fed back to the user.

5. The method of claim 4, wherein, After obtaining the consultation question and key summary information based on the graph model embedding vector, the following is also included: Generate a follow-up judgment result, which is used to determine whether the consultation task needs to be followed up; Furthermore, after using the customer service response result corresponding to the summary content as the response result of the consultation task corresponding to the customer service consultation summary and feeding it back to the user, it also includes: The consultation tasks corresponding to the customer service consultation summary set do not need to be tracked.

6. An apparatus for generating customer service inquiry summaries, characterized in that, include: The dialogue decomposition module is used to decompose customer service inquiries into at least one inquiry task. The task feature extraction module is used to determine the form data corresponding to each consultation task and extract features from the consultation task to obtain a consultation task vector. The task embedding module is used to embed the consultation task vector and the form data into a pre-trained graph model to obtain an embedded graph model, and to extract features from the embedded graph model to obtain a graph model embedding vector; and to establish a centrally symmetric matrix based on the relationship between the nodes of the embedded graph model. An embedded vector calculation module is used to obtain consultation questions and key summary information based on the embedded vectors of the graph model. The embedding vector calculation module is further configured to: perform format transformation on the graph model embedding vector; concatenate the format-transformed graph model embedding vector with the central symmetric matrix to obtain a concatenated vector; multiply the concatenated vector with preset weights to obtain the score of the hidden layer in the embedded graph model; determine the consultation question based on the score of the hidden layer; and generate key summary information based on the graph model embedding vector and the pre-trained dialogue vector. The summary generation module is used to generate a customer service consultation summary based on the consultation question and the key summary information.

7. An electronic device for generating a customer care consultation summary, the electronic device comprising: include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-5.

8. A computer readable medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

Images