Customer service session dispatching method and apparatus, device, medium, product

Abstract

The application discloses a customer service conversation scheduling method and device, equipment, medium and product, and the method comprises the following steps: obtaining dialogue data of any conversation in an intelligent customer service system, wherein the dialogue data comprises question text of a question user and reply text of the system; extracting deep semantic information of a word vector corresponding to each word segmentation from the dialogue data to obtain a corresponding semantic vector; calculating a correlation matrix between a label vector of each label in a label library and each semantic vector, and determining a label corresponding to a label vector satisfying a preset correlation matching condition with a full-amount semantic vector as a hit label according to the correlation matrix; connecting a conversation with a hit label to a corresponding agent interface of the hit label, so that a question user of the conversation continues the conversation with an agent user of the agent interface. Therefore, the intelligent customer service system is improved in the efficiency of shunting and scheduling, the waiting time of the question user is saved, and the cost is reduced and the benefit is increased for a merchant user.

Description

Technical Field

[0001] The present application relates to the field of intelligent customer service technology, and in particular to a customer service session scheduling method and its corresponding apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] On e-commerce platforms, consumer users often conduct pre-sales consultations, in-sales inquiries, and after-sales feedback with merchant users through third-party chat tools or built-in chat tools on the e-commerce platform. The intelligent customer service system of the e-commerce platform is responsible for responding to questions raised by consumer users. The intelligent customer service system generally provides two types of responses: intelligent robot response and seat user response, which are specifically dispatched by the intelligent customer service system based on actual conditions.

[0003] Due to the large number of live chat users, different customers have different types of questions, with varying degrees of importance and urgency for merchants. Typically, agents respond sequentially based on the order of the chat list, or at the agent's discretion. Without prior knowledge of the content and type of customer questions, it's difficult to effectively prioritize and sort them. When merchants have multiple agents, each responsible for answering different questions, the system can't automatically triage conversations, requiring administrators to manually triage them, which is time-consuming and labor-intensive.

[0004] In existing solutions, a sorting function is generally added to the system to filter conversations by time to reduce customer waiting time, or manual filtering is performed based on keywords. This processing method is inefficient and cannot achieve intelligent diversion and scheduling of conversations. It cannot truly and effectively reduce the waiting time of consumer users, so it needs to be improved. Summary of the Invention

[0005] The primary purpose of the present application is to solve at least one of the above problems and provide a customer service session scheduling method and its corresponding device, computer equipment, computer-readable storage medium, and computer program product.

[0006] In order to meet the various objectives of this application, this application adopts the following technical solutions:

[0007] A customer service session scheduling method provided to meet one of the purposes of this application includes the following steps:

[0008] Obtaining conversation data from any conversation in the intelligent customer service system, including the text of the question generated by the user asking the question and the text of the reply generated by the system;

[0009] Extracting deep semantic information of word vectors corresponding to each word segment from the conversation data to obtain semantic vectors corresponding to each word vector;

[0010] Calculate the correlation matrix between the label vector corresponding to each label in the preset label library and each of the semantic vectors, and determine the label corresponding to the label vector that meets the preset correlation matching conditions with the full semantic vector as the hit label based on the correlation matrix;

[0011] The conversation corresponding to the conversation data with the hit tag determined is connected to the agent interface corresponding to the hit tag, so that the agent user corresponding to the agent interface and the questioning user of the conversation can continue the conversation.

[0012] In a further embodiment, extracting deep semantic information of the word vector corresponding to each word segment from the conversation data to obtain the semantic vector corresponding to each word vector includes the following steps:

[0013] Concatenating the question text and the reply text in the conversation data in the order of their creation to form a conversation text;

[0014] Performing a preset standardization preprocessing on the dialogue text to make it into a standard text;

[0015] Performing word segmentation processing on the standard text to obtain a word segmentation sequence of the standard text;

[0016] Querying a preset word vector table to determine the word vector corresponding to each word in the word segmentation sequence, and obtaining a word vector sequence corresponding to the standard text;

[0017] A text extraction model that has been pre-trained to a convergent state is used to extract deep semantic information from each word vector in the word vector sequence with reference to context information and mapped to a high-dimensional space to obtain a semantic vector corresponding to each word vector.

[0018] In a further embodiment, a correlation matrix is ​​calculated between the label vectors corresponding to each label in a preset label library and each of the semantic vectors, and a label corresponding to the label vector that meets a preset correlation matching condition with the full semantic vector is determined as a hit label based on the correlation matrix, including the following steps:

[0019] A preset correlation algorithm is used to calculate a correlation matrix between the label vector corresponding to each label in the preset label library and each of the semantic vectors, where the correlation matrix includes a correlation coefficient mapping each semantic vector to each label vector;

[0020] Based on the correlation coefficients, a preset voting algorithm is applied to calculate the comprehensive correlation coefficient corresponding to each of the label vectors mapped from all semantic vectors to represent the probability that the conversation data belongs to each label vector;

[0021] Screening is performed according to preset relevant matching conditions to determine that the one or more label vectors with the highest relevant probability are the label vectors most relevant to the full semantic vector, and the corresponding labels are hit labels.

[0022] In an extended embodiment, before the step of calculating the correlation matrix between the tag vector corresponding to each tag in the preset tag library and each of the semantic vectors, the following steps are included:

[0023] Obtain tag configuration information corresponding to the intelligent customer service system, wherein the tag configuration information includes mapping relationship data between a tag and one or more keywords;

[0024] A label encoding model that has been pre-trained to a convergent state is used to encode each keyword in the label configuration information to obtain a label vector corresponding to the corresponding deep semantic information, and the label vector is stored in the label library.

[0025] In a further embodiment, accessing a session corresponding to the conversation data with a hit tag determined to an agent interface corresponding to the hit tag includes the following steps:

[0026] Query the annotation tags corresponding to each preset agent interface, and determine the agent interface whose annotation tags completely contain all hit tags as the agent interface corresponding to the hit tag;

[0027] Establishing a data communication link between the determined agent interface and the user who asked the question in the conversation corresponding to the conversation data with the hit tag, so that the agent user corresponding to the agent interface and the user who asked the question in the conversation can continue the conversation;

[0028] A notification message indicating the switching of seat users is pushed to the chat interface of the questioning user side corresponding to the conversation.

[0029] In an extended embodiment, after the step of determining, according to the correlation matrix, a label corresponding to a label vector that satisfies a preset correlation matching condition with the full semantic vector as a hit label, the following steps are included:

[0030] Respond to any agent user's session management request and push a session list to the agent, which contains the mapping relationship data between all ongoing sessions of the intelligent customer service system and their hit tags;

[0031] receiving a hit tag specified by the agent user, and updating the conversation list so that the conversation list only includes conversations corresponding to the specified hit tag;

[0032] In response to the agent user selecting a participation request of any one session in the session list, a data communication link between the agent user and the question user of the session is established to continue the session by the agent user and the question user.

[0033] A customer service session scheduling device is provided for adapting to one of the purposes of the present application, comprising: a data acquisition module, a semantic extraction module, a label determination module, and a communication establishment module, wherein the data acquisition module is configured to acquire dialogue data of any session in an intelligent customer service system, the dialogue data comprising question text generated by a question user of the session and reply text generated by the system; the semantic extraction module is configured to extract deep semantic information of word vectors corresponding to each word piece in the dialogue data from the dialogue data to obtain semantic vectors corresponding to each word vector; the label determination module is configured to calculate a correlation matrix between label vectors corresponding to each label in a preset label library and each semantic vector, and determine a label corresponding to a label vector that satisfies a preset correlation matching condition with the full semantic vector as a hit label according to the correlation matrix; and the communication establishment module is configured to connect a session corresponding to dialogue data of which a hit label is determined to a corresponding agent interface of the hit label, so as to continue the session by a corresponding agent user of the agent interface and a question user of the session.

[0034] In a deepened embodiment, the semantic extraction module comprises: a text splicing sub-module configured to sequentially splice question text and reply text in the dialogue data according to their creation time to form dialogue text; a text specification sub-module configured to perform a preset standardization preprocessing on the dialogue text to make it a standard text; a text word segmentation sub-module configured to perform word segmentation processing on the standard text to obtain a word segmentation sequence of the standard text; a word segmentation coding sub-module configured to query a preset word vector table to determine word vectors corresponding to each word piece in the word segmentation sequence and obtain a word vector sequence corresponding to the standard text; and a semantic extraction sub-module configured to extract deep semantic information of each word vector in the word vector sequence by referring to context information using a text extraction model pre-trained to a convergent state and map the deep semantic information to a high-dimensional space to obtain semantic vectors corresponding to each word vector.

[0035] In a further embodiment, the label determination module includes: a matrix construction submodule, which is used to use a preset correlation algorithm to calculate the correlation matrix between the label vector corresponding to each label in the preset label library and each of the semantic vectors, and the correlation matrix contains the correlation coefficient of each semantic vector mapped to each label vector; a coefficient synthesis submodule, which is used to apply a preset voting algorithm based on the correlation coefficient to calculate the comprehensive correlation coefficient corresponding to all semantic vectors mapped to each of the label vectors to represent the relevant probability of the conversation data belonging to each label vector; a label hit submodule, which is used to screen according to preset relevant matching conditions and determine that the one or more label vectors with the highest correlation probability are the label vectors most relevant to the full semantic vector, and the corresponding label is the hit label.

[0036] In an extended embodiment, the customer service session scheduling device of the present application further includes: a configuration acquisition module for acquiring label configuration information corresponding to the intelligent customer service system, wherein the label configuration information includes mapping relationship data between the label and one or more keywords; a label extraction module for encoding each keyword in the label configuration information using a label encoding model pre-trained to a convergent state, obtaining a label vector corresponding to the corresponding deep semantic information, and storing the label vector in the label library.

[0037] In a further embodiment, the communication establishment module includes: an agent determination submodule, which is used to query the annotation tags corresponding to each preset agent interface, and determine the agent interface whose annotation tags completely contain all hit tags as the agent interface corresponding to the hit tag; a link establishment submodule, which is used to establish a data communication link between the determined agent interface and the questioning user of the session corresponding to the conversation data with the hit tag, so that the agent user corresponding to the agent interface and the questioning user of the session continue the conversation; a notification push submodule, which is used to push a notification message representing the switching of the agent user to the chat interface on the questioning user side corresponding to the session.

[0038] In an extended embodiment, the customer service session scheduling device of the present application further includes: an initial push submodule, used to respond to a session management request of any agent user and push a session list to the agent, wherein the session list contains mapping relationship data between all ongoing sessions of the intelligent customer service system and their hit tags; an update push submodule, used to receive a hit tag specified by the agent user and update the session list so that the session list only contains sessions corresponding to the specified hit tag; a call switching submodule, used to respond to the agent user's request to participate in any one of the sessions in the session list, and establish a data communication link between the agent user and the questioning user of the session, so that the agent user and the questioning user can continue the session.

[0039] A computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the customer service session scheduling method described in the present application.

[0040] A computer-readable storage medium is provided to meet another purpose of the present application, which stores a computer program implemented according to the customer service session scheduling method in the form of computer-readable instructions. When the computer program is called and executed by a computer, it executes the steps included in the method.

[0041] A computer program product provided to meet another purpose of the present application includes a computer program / instruction, which, when executed by a processor, implements the steps of the method described in any embodiment of the present application.

[0042] Compared with the prior art, the advantages of this application are as follows:

[0043] First, this application targets conversations in an intelligent customer service system. By calculating a correlation matrix between the semantic vectors obtained by word segmentation of the conversation data corresponding to the conversation and the label vectors of preset labels, the conversation data is matched with labels that meet preset matching conditions based on the correlation matrix. The conversation data is then diverted according to the labels to which it belongs and allocated to the agent interface that matches the label, so that a data communication connection is established between the questioning user of the conversation and the agent interface, so that the agent user of the agent interface can be in a conversation with corresponding content according to his preset label. Automatic diversion and scheduling of conversations is achieved by intelligently and automatically matching agent interfaces for conversations without relying on manual allocation, thereby improving the service effectiveness of the intelligent customer service system, reducing the waiting time for questioning users to be assigned agent users, and improving user experience.

[0044] Secondly, this application actually uses tags to play the role of semantic probes, allowing the intelligent customer service system to define tags by setting semantic probes. Tags can take effect in real time through configuration, and can indicate the importance, different functions, different products, etc. of each conversation, and can identify conversations in real time, so that the seat users and conversations can be associated with tags and matched, avoiding the strong rule matching phenomenon that relies entirely on keywords in traditional technologies, and can better recall conversations with similar semantics but not containing keywords, making the operation of the intelligent customer service system more robust.

[0045] In addition, the implementation of the technical solution of this application can enable large-scale customer service scenarios such as e-commerce platforms to avoid a large amount of manpower work, omitting the need for administrators of each merchant user to manually browse conversations, thereby saving corresponding implementation costs and achieving economies of scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0047] Figure 1 This is a flowchart of a typical embodiment of the customer service session scheduling method of the present application;

[0048] Figure 2 This is a flow chart of extracting semantic vectors from conversation data in an embodiment of the present application;

[0049] Figure 3 Schematic diagram of the process of matching semantic vectors and label vectors in an embodiment of the present application;

[0050] Figure 4 A flowchart of the label configuration process in an embodiment of the present application;

[0051] Figure 5 Schematic diagram of the process of scheduling an agent interface for a conversation in an embodiment of the present application;

[0052] Figure 6 Schematic diagram of the process of selecting conversations by an agent user in an embodiment of the present application;

[0053] Figure 7 This is a functional block diagram of the customer service session scheduling device of this application;

[0054] Figure 8 This is a schematic diagram of the structure of a computer device used in this application. DETAILED DESCRIPTION

[0055] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and are not to be construed as limiting the present application.

[0056] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any units and all combinations of one or more associated listed items.

[0057] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0058] It will be understood by those skilled in the art that the terms "client," "terminal," and "terminal device" as used herein include both devices that are wireless signal receivers, i.e., devices that only have wireless signal receivers without transmission capabilities, and devices that have receiving and transmitting hardware capable of two-way communication over a two-way communication link. Such devices may include: cellular or other communication devices such as personal computers and tablet computers, which have single-line displays, multi-line displays, or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service), which may combine voice, data processing, fax, and / or data communication capabilities; PDA (Personal Digital Assistant), which may include a radio frequency receiver, a pager, Internet / Intranet access, a web browser, a notepad, a calendar, and / or a GPS (Global Positioning System) receiver; and conventional laptop and / or palmtop computers or other devices, which have and / or include a radio frequency receiver. As used herein, the terms "client," "terminal," or "terminal device" may be portable, transportable, or installed in a vehicle (air, sea, and / or land), or may be adapted and / or configured to operate locally and / or in a distributed manner at any other location on Earth and / or in space. As used herein, the terms "client," "terminal," or "terminal device" may also refer to a communication terminal, an Internet terminal, or a music / video playback terminal, such as a PDA, an MID (Mobile Internet Device), and / or a mobile phone with music / video playback capabilities, or may include a smart TV, a set-top box, or other device.

[0059] The hardware referred to by names such as "server", "client", and "service node" in this application is essentially an electronic device with capabilities equivalent to those of a personal computer. It is a hardware device that has the necessary components revealed by the von Neumann principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. Computer programs are stored in its memory, and the central processing unit loads the program stored in the external memory into the internal memory for execution, executes the instructions in the program, and interacts with the input and output devices to complete specific functions.

[0060] It should be noted that the concept of "server" referred to in this application can also be extended to server clusters. Based on the network deployment principles understood by those skilled in the art, the servers described should be logically divided. In physical space, these servers can be independent of each other but callable through interfaces, or integrated into a single physical computer or a computer cluster. Those skilled in the art should understand this flexibility and should not use it to constrain the implementation of the network deployment method of this application.

[0061] Unless expressly specified, one or more technical features of the present application can be deployed on a server for implementation and accessed by a client through a remote call to obtain an online service interface provided by the server, or can be directly deployed and run on a client for implementation.

[0062] Unless expressly specified otherwise, the neural network models referenced or may be referenced in this application may be deployed on a remote server and remotely called on the client, or may be deployed and directly called on a client with sufficient device capabilities. In some embodiments, when it runs on the client, its corresponding intelligence may be obtained through transfer learning to reduce the requirements for the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.

[0063] Unless explicitly specified, the various data involved in this application can be stored remotely on a server or on a local terminal device, as long as they are suitable for being called by the technical solution of this application.

[0064] Those skilled in the art should be aware that although the various methods of this application are described based on the same concept and thus exhibit commonality, unless otherwise specified, these methods can be independently executed. Similarly, the various embodiments disclosed in this application are all based on the same inventive concept. Therefore, concepts with the same expression, as well as concepts that are appropriately transformed for convenience despite different expression, should be understood as equivalent.

[0065] Unless expressly stated to be mutually exclusive, the various embodiments disclosed in this application may be cross-combined with the relevant technical features of the various embodiments to flexibly construct new embodiments, as long as such combination does not deviate from the creative spirit of this application and can meet the needs of the prior art or resolve certain deficiencies in the prior art. Those skilled in the art should be aware of such flexibility.

[0066] A customer service conversation scheduling method of the present application can be programmed as a computer program product and deployed in a client or server for execution. For example, in the e-commerce platform application scenario of the present application, it is generally deployed in a server for implementation. The method can be executed by accessing an interface opened after the computer program product is run and conducting human-computer interaction with the process of the computer program product through a graphical user interface.

[0067] An exemplary application scenario of the present application is an application in an e-commerce platform based on independent stations. Each independent station is a merchant instance of the e-commerce platform, has an independent access domain name, and its actual owner is responsible for publishing and updating products.

[0068] Each independent website merchant instance can be configured with the intelligent customer service system provided by the e-commerce platform to introduce an intelligent customer service robot. The intelligent customer service system is used to provide consulting services to relevant consumer questioning users, create a corresponding session, and the questioning user enters the corresponding chat interface of the session and enters the question to be consulted as the question text. After the intelligent customer service system of the e-commerce platform receives the question text, in one case, the intelligent customer service robot uses the question text to perform semantic matching with the question set in the knowledge base pre-configured for the independent website, matches the preset question that is most semantically similar to the question text, and determines the standard question in the question set to which the preset question belongs based on the preset question. Then, the pre-stored answer set mapped to the standard question is called, and one of the preset answers is determined as the target answer. It is output to the chat interface as a reply text to answer the consumer user's question and meet his or her consultation needs. In another case, the session is switched to the agent interface, allowing the questioning user and the agent user to have a direct manual conversation, and the agent user gives a corresponding reply text based on the question text submitted by the questioning user. In these two cases, the intelligent customer service system can switch and schedule according to the specific situation. The above two situations can be switched between each other by the intelligent customer service system, and either situation can be configured as the default situation of the intelligent customer service system.

[0069] In the latter case, when there are multiple agent interfaces (i.e., multiple corresponding agent users), assigning corresponding agent interfaces to conversations typically requires a method for splitting the conversations. This allows conversations with different characteristics to be dispatched to different agent interfaces, with different agents responsible for answering user questions. Therefore, for the latter manual response scenario listed above, conversation scheduling is one of the required features in this application scenario.

[0070] Refer to the application scenarios disclosed above. Figure 1 In a typical embodiment, the customer service session scheduling method of the present application includes the following steps:

[0071] Step S1300, obtaining dialogue data of any session in the intelligent customer service system, the dialogue data including question text generated by a questioning user of the session and reply text generated by the system:

[0072] For each session, the chat record generated by the questioning user and the robot or the agent user in the chat interface interaction includes the question text proposed by the questioning user and the reply text automatically replied by the agent user or the robot, which is usually stored in the database to constitute the dialogue data corresponding to the session.

[0073] Generally, the intelligent customer service system concurrently responds to the session requests of different questioning users and creates a corresponding session for each questioning user, so the intelligent customer service system generally maintains multiple sessions at the same time, and each session generates the dialogue data.

[0074] When the consumer user logged in at the client device first enters the chat interface of the intelligent customer service system, the intelligent robot is generally called by the intelligent customer service system for initial response to send welcome information and the like to the chat interface, and the consumer user, i.e., the questioning user, then inputs the related question text to the chat interface to obtain the corresponding reply text.

[0075] The dialogue data can be the chat record corresponding to the full historical data of the corresponding session, or the chat record corresponding to the current question and answer period (usually the same day), which can be determined by those skilled in the art.

[0076] Step S1400, extracting deep semantic information of the word vector corresponding to each word piece in the dialogue data to obtain a semantic vector corresponding to each word vector:

[0077] First, the dialogue data is taken as a complete text, and after standardized preprocessing according to a preset, word segmentation is performed to obtain a word segmentation sequence, then a preset word vector table is used to encode the word segmentation to obtain a word vector of each word piece, a corresponding word vector sequence is obtained, and a preset neural network model is used to extract deep semantic information of the word vector sequence to finally obtain a semantic vector corresponding to each word vector. The semantic vector is a representation of the deep semantic features of the corresponding word vector. Considering that the dialogue data itself is organized in the form of one question and one answer, the neural network model is suitable to use a corresponding model with context information processing capability, in particular, a neural network model with a multi-head attention mechanism. For specific variations of this step, more detailed embodiments will be provided later in this application, which will be omitted here.

[0078] Step S1500: Calculate the correlation matrix between the label vectors corresponding to each label in the preset label library and each of the semantic vectors. According to the correlation matrix, determine the label corresponding to the label vector that meets the preset correlation matching condition with the full semantic vector as the hit label:

[0079] The intelligent customer service system is equipped with a tag library, which stores mapping relationship data between multiple tags and one or more keywords used to describe the tags. The tags and keywords are generally created, modified, deleted, and maintained by the management user of the intelligent customer service system. In a specific application example, the tags can be divided according to the importance of the matters involved, such as "light", "serious", "postponable", "urgent", etc. The keywords corresponding to each tag can be words that describe the specific business type, business scenario, and transaction content. For example, for the tag "serious", its keywords can be described as "broken", "angry", "quality problem", "defect", "return", etc. Similarly, for the "urgent tag", its keywords can correspond to "return", "not buying", "expired", etc. It can be seen that the keywords in the tag library are used to describe and define the role of a tag, and the tag itself can serve as a type identifier. Those skilled in the art set the type classification standards by themselves, set multiple tags, and use the corresponding keywords of the tag to describe and define the tag accordingly. Based on this principle, in another embodiment, the tags can also be divided according to the business links involved in the e-commerce order, for example, according to the different stages of "pre-sales", "sales", and "after-sales", thereby facilitating the management of seat users according to the order business links. Such as these, those skilled in the art can flexibly implement them.

[0080] For the tags in the tag library, a neural network model that has been pre-trained to convergence can be used to extract deep semantic information from the corresponding keywords and construct a tag vector. This tag vector can be extracted in real time or pre-extracted and stored in the tag library for easy access. In short, the tag vector can be used to semantically represent the keywords of the corresponding tags, making it easier for the intelligent customer service system to use the tag vector for semantic matching.

[0081] For the full semantic vector determined by the conversation data of a conversation, a preset correlation algorithm can be used to calculate the correlation between it and the label vector corresponding to the full label, thereby obtaining a correlation matrix. The correlation algorithm can be implemented by using any algorithm for calculating data distance, such as cosine similarity algorithm, Euclidean distance, Pearson correlation coefficient, Jaccard algorithm, etc. After calculation by the correlation algorithm, a corresponding correlation coefficient can be determined between the full semantic vector and the full label vector, and between each semantic vector and each label vector. Based on this, the normalized comprehensive correlation coefficient of the full semantic vector mapped to each label vector is calculated. Then, based on the comprehensive correlation coefficient of each label vector, the preset correlation matching condition can be used to determine one or more label vectors that are most relevant to the full semantic vector. The labels corresponding to these label vectors are the hit labels corresponding to the conversation data of the conversation.

[0082] The relevant matching condition can be preset to adopt the maximum value among all comprehensive correlation coefficients, and thus the label corresponding to the maximum value is the hit label; it can also be preset to a preset threshold for screening all comprehensive correlation coefficients, and thus the label corresponding to the label vector whose comprehensive correlation coefficient is greater than the preset threshold is the hit label.

[0083] Step S1600: Connect the conversation corresponding to the conversation data with the hit tag to the agent interface corresponding to the hit tag, so that the agent user corresponding to the agent interface and the questioning user of the conversation can continue the conversation:

[0084] In the intelligent customer service system, each agent interface is pre-configured with tags from the tag library. For example, an agent user's agent interface is configured to be associated with the tags "urgent" and "important". If the hit tag of a subsequent session matches the tag pre-marked by this agent interface, the session can be assigned to this agent interface for processing.

[0085] Once a conversation's dialog data has its corresponding hit tag identified, the agent interface matching the hit tag can be found from all agent interfaces based on a preset matching strategy, specifically determined by the matching strategy. For example, in one embodiment, the matching strategy is configured so that when the agent interface's annotated tags completely encompass all of the conversation's hit tags, the two are considered a match, and the conversation can be diverted to the agent interface for processing. In another embodiment, the matching strategy is configured so that when one of the agent interface's annotated tags matches one of the conversation's hit tags, the conversation can also be diverted to the agent interface for processing.

[0086] It can be seen that this application opens up many flexible configurations, which facilitate flexible settings in the background of the intelligent customer service system, including flexible setting of the matching strategies, flexible setting of the relevant matching conditions, etc., opening up more abundant functions for relevant merchant users who configure the intelligent customer service system.

[0087] When a conversation is assigned to a matching agent interface, the intelligent customer service system establishes a data communication link between the questioning user of the conversation and the agent user corresponding to the agent interface. The chat interface of the questioning user will subsequently display the reply text of the agent user instead of the robot's automatic reply text. Therefore, the conversation is actually switched from the robot's automatic answering mode to the manual customer service mode.

[0088] When there are multiple agent interfaces and a large number of questioning users interacting concurrently, the intelligent customer service system can, through the above steps of this application, respectively schedule a matching agent interface for the conversation of each questioning user in real time, without manual intervention, and can achieve accurate answering task allocation, which is very efficient.

[0089] According to the above typical embodiments and their alternative embodiments, it can be seen that the technical solution of the present application has many positive effects, including but not limited to the following aspects:

[0090] First, this application targets conversations in an intelligent customer service system. By calculating a correlation matrix between the semantic vectors obtained by word segmentation of the conversation data corresponding to the conversation and the label vectors of preset labels, the conversation data is matched with labels that meet preset matching conditions based on the correlation matrix. The conversation data is then diverted according to the labels to which it belongs and allocated to the agent interface that matches the label, so that a data communication connection is established between the questioning user of the conversation and the agent interface, so that the agent user of the agent interface can be in a conversation with corresponding content according to his preset label. Automatic diversion and scheduling of conversations is achieved by intelligently and automatically matching agent interfaces for conversations without relying on manual allocation, thereby improving the service effectiveness of the intelligent customer service system, reducing the waiting time for questioning users to be assigned agent users, and improving user experience.

[0091] Secondly, this application actually uses tags to play the role of semantic probes, allowing the intelligent customer service system to define tags by setting semantic probes. Tags can take effect in real time through configuration, and can indicate the importance, different functions, different products, etc. of each conversation, and can identify conversations in real time, so that the seat users and conversations can be associated with tags and matched, avoiding the strong rule matching phenomenon that relies entirely on keywords in traditional technologies, and can better recall conversations with similar semantics but not containing keywords, making the operation of the intelligent customer service system more robust.

[0092] In addition, the implementation of the technical solution of this application can enable large-scale customer service scenarios such as e-commerce platforms to avoid a large amount of manpower work, omitting the need for administrators of each merchant user to manually browse conversations, thereby saving corresponding implementation costs and achieving economies of scale.

[0093] See also Figure 2 In a further embodiment, step S1400, extracting deep semantic information of the word vector corresponding to each word segment from the conversation data to obtain the semantic vector corresponding to each word vector, includes the following steps:

[0094] Step S1410: Concatenate the question text and the reply text in the conversation data in the order of their creation to form a conversation text.

[0095] In the intelligent customer service system, the question text and reply text of the dialogue data are organized according to their generation time, that is, the creation time, and this time actually also plays a guiding role in the context relationship. Therefore, the question text and reply text in the dialogue data can be orderly spliced ​​into the same continuous text according to the creation time to form the dialogue text.

[0096] Step S1420: Performing a preset standardization preprocessing on the dialogue text to make it a standard text:

[0097] Then, according to a preset standardized preprocessing procedure, the dialogue text is formatted and preprocessed, including, for example, removing stop words such as modal particles and auxiliary words, removing spaces, removing punctuation symbols, expressions, noise information, etc. For example, if the input text is "Why hasn't my express arrived yet? ah ah ah???", after preprocessing, the text "Why hasn't my express arrived yet" is obtained. The text after preprocessing is the standard text.

[0098] Step S1430: perform word segmentation processing on the standard text to obtain a word segmentation sequence of the standard text:

[0099] Furthermore, a preset word segmentation model is used to perform word segmentation processing on the standard text. The word segmentation model can adopt a statistical-based machine learning algorithm or a dictionary-based word segmentation algorithm, such as HMM, CRF, SVM, and deep learning algorithms constructed based on basic models such as LSTM+CRF, Bert+CRF, etc. Those skilled in the art can implement it flexibly and obtain a word segmentation sequence consisting of all the words in the standard text by performing word segmentation processing on the standard text.

[0100] Step S1440: query a preset word vector table to determine the word vector corresponding to each word in the word segmentation sequence, and obtain the word vector sequence corresponding to the standard text:

[0101] Then, the word segmentation sequence is vector-encoded. Specifically, each word segmentation can be converted into a word vector representation by querying a word vector table, thereby obtaining a word vector sequence corresponding to the word segmentation sequence. The word vector table is pre-trained and can be a word vector table obtained by a known pre-trained model such as Bert, Word2Vec, GloVe, Fasttext, ELMO, etc., and those skilled in the art can flexibly select it.

[0102] Step S1450: Using a text extraction model that has been pre-trained to a convergent state, extract deep semantic information from each word vector in the word vector sequence with reference to context information and map it to a high-dimensional space to obtain a semantic vector corresponding to each word vector:

[0103] Finally, a text extraction model suitable for extracting deep semantic information from the word vector sequence is used, pre-selected and trained to convergence. This model can be constructed using basic models such as Bert, RNN, and CNN, and in particular, using basic models with a multi-head attention mechanism, such as Bert, Electra, Transformer, LSTM, and BiLSTM. The addition of a multi-head attention mechanism allows for full reference to contextual information between word vectors during semantic extraction, resulting in more effective representation learning. This type of model is trained using a sufficient training dataset to acquire the ability to extract deep semantic information from the word vector sequence and map it to a high-dimensional space. The high-dimensional vector sequence obtained for the word vector sequence is a semantic vector sequence, which contains semantic vectors corresponding to each word vector. Each semantic vector represents a deep semantic representation of a corresponding word vector. At this point, a deep semantic representation of the conversation data is achieved, which can then be used to participate in the subsequent matching process with the label vector.

[0104] This embodiment achieves deep semantic representation of the conversation data by performing more specific semantic processing on the conversation data. Since the context information of the conversation data is referenced during the semantic extraction process, the obtained semantic vector has a more effective feature representation capability, which helps guide its matching with the label vector, thereby improving the matching accuracy.

[0105] See also Figure 3 In a further embodiment, step S1500, calculating a correlation matrix between the label vectors corresponding to each label in a preset label library and each of the semantic vectors, and determining, based on the correlation matrix, a label corresponding to a label vector that satisfies a preset correlation matching condition with the full semantic vector as a hit label, includes the following steps:

[0106] Step S1510, a preset correlation algorithm is used to calculate a correlation matrix between each label vector corresponding to each label in a preset label library and each semantic vector, the correlation matrix including a correlation coefficient of each semantic vector mapping to each label vector:

[0107] As described above, the correlation algorithm can use any algorithm for calculating data distance, such as cosine similarity algorithm, Euclidean distance, Pearson correlation coefficient, Jaccard algorithm, etc. For example, in the embodiment, the cosine similarity algorithm is used as the preset correlation algorithm to calculate the similarity value between each label vector and each semantic vector as the correlation coefficient, thereby obtaining a correlation matrix, in which one dimension corresponds to each label vector and the other dimension corresponds to each semantic vector, and each element corresponds to the correlation coefficient between the corresponding label vector and the corresponding semantic vector.

[0108] Step S1520, according to the correlation coefficient, a preset voting algorithm is applied to calculate a comprehensive correlation coefficient of all semantic vectors mapping to each label vector to represent the correlation probability of the dialogue data belonging to each label vector:

[0109] As can be easily understood, in the correlation matrix, for each label vector, the correlation coefficients corresponding to each semantic vector are summarized to obtain a comprehensive correlation coefficient, which can indicate the correlation probability of the full semantic vector of the dialogue data mapping to the label vector. The summary method can be to take the average or the sum of the correlation coefficients corresponding to the label vector of the full semantic vector, as long as it is normalized to a standard numerical space, such as [0, 1], to facilitate comparison with other comprehensive correlation coefficients. In specific implementation, a pooling layer can be used for processing. Thus, a comprehensive correlation coefficient sequence corresponding to the full label vector can be obtained, and each comprehensive correlation coefficient in the sequence represents the correlation probability of the dialogue data belonging to the label corresponding to the current comprehensive correlation coefficient.

[0110] Step S1530, according to a preset correlation matching condition, one or more label vectors with the highest correlation probability are determined as the most relevant label vectors of the full semantic vector, and the labels corresponding to the label vectors are hit labels:

[0111] As mentioned above, a relevant matching condition can be preset to screen the correlation matrix. For example, the relevant matching condition gives a preset threshold, and each comprehensive correlation coefficient in the comprehensive correlation coefficient sequence is compared with the preset threshold to reduce the comprehensive correlation coefficient below the preset threshold, and only retain the comprehensive correlation coefficient above the preset threshold. The labels corresponding to these retained comprehensive correlation coefficients may include one or more, which can be determined as the hit labels of this application, that is, the hit labels corresponding to the conversation data to which the full semantic vector belongs. After determining these hit labels, they can be used to match the agent interface for the corresponding conversation of the conversation data.

[0112] In this embodiment, a correlation matrix is ​​used to match labels to semantic vectors corresponding to conversations, facilitating further matching of conversations to agent interfaces based on these labels. This computational complexity is minimal and fast, making it particularly suitable for high-concurrency scenarios, providing efficient computation for massive conversations and thus improving the response rate of the intelligent customer service system in assigning agent interfaces to conversations. Unlike existing techniques that rely on strong rule-based matching based on keywords, this semantic matching process, based on deep semantic information, enables a more robust intelligent customer service system.

[0113] See also Figure 4 In an extended embodiment, in step S1500, before the step of calculating the correlation matrix between the tag vector corresponding to each tag in the preset tag library and each of the semantic vectors, the following steps are included:

[0114] Step S1100: Obtain tag configuration information corresponding to the intelligent customer service system, wherein the tag configuration information includes mapping relationship data between a tag and one or more keywords:

[0115] To facilitate routine maintenance by backend administrators of the intelligent customer service system, the system provides a maintenance function for administrators in the backend, providing a configuration page for administrators to access tag configuration information. This tag configuration information allows administrators to describe and define tags in the tag library. Administrators use this tag configuration information to provide tags and one or more keywords used to describe and define the tags, thereby enabling timely updates to the tag library.

[0116] As mentioned above, tags are divided according to certain type classification standards. Correspondingly, they can be described by one or more of the aforementioned keywords. Mapping relationship data is formed between tags and keywords and is submitted to the intelligent customer service system for associated storage.

[0117] Step S1200: Encode each keyword in the tag configuration information using a tag encoding model that has been pre-trained to a convergent state to obtain a tag vector corresponding to the corresponding deep semantic information, and store the tag vector in the tag library:

[0118] Furthermore, a label encoding model that has been pre-trained to a convergent state is used to encode each keyword in the label configuration information to obtain a corresponding label vector. The specific process may be to first convert each keyword of the label into a corresponding word vector by querying a preset word vector table to obtain a word vector sequence, and then the label encoding model extracts deep semantic information based on each word vector in the word vector sequence and maps it to a high-dimensional space to obtain a corresponding label vector. This label vector can be stored in the label library together with the corresponding label.

[0119] Similar to the text extraction model, the label encoding model can be constructed using basic models such as BERT, RNN, and CNN, or further constructed using basic models with a multi-head attention mechanism, such as BERT, Electra, Transformer, LSTM, and BiLSTM. It is then trained with a sufficient training dataset until convergence, making it suitable for learning the representation of keywords corresponding to the labels. Because the implementation of the text extraction model and the label encoding model is essentially the same, in one alternative embodiment, both can be implemented using the same neural network model, thereby further reducing training costs.

[0120] It should be noted that this embodiment does not necessarily depend on the execution of step S1300 and step S1400 first, as both steps can be performed concurrently.

[0121] In this embodiment, a tag configuration function is connected to the intelligent customer service system to facilitate the maintenance of the tag library and facilitate merchant users to update the tags in the tag library in a timely manner. In addition, since the configuration of tags in the tag library is technically decoupled from the call of its tag vector, the added tags can also take effect immediately, thereby improving the intelligence level of the intelligent customer service system.

[0122] See also Figure 5 In a further embodiment, step S1600 of accessing the session corresponding to the conversation data with the hit tag determined to the agent interface corresponding to the hit tag includes the following steps:

[0123] Step S1610: Query the annotation tags corresponding to each preset agent interface, and determine the agent interface whose annotation tags completely include all hit tags as the agent interface corresponding to the hit tag:

[0124] As mentioned above, the merchant user can pre-associate the tags in the tag library with the corresponding agent interface of each agent user in the background of the intelligent customer service system, so as to make the tags become the labeling tags corresponding to the agent interface, so as to indicate the session corresponding to the tags suitable for the agent user to handle.

[0125] Therefore, in order to realize the shunting of the session corresponding to the dialogue data, it is necessary to first query the labeling tags carried by each agent interface. In this embodiment, according to the preset matching strategy, when the labeling tags carried by an agent interface completely contain all hit tags of a session, the agent interface is determined as the agent interface corresponding to the hit tags.

[0126] Step S1620, establishing a data communication link between the determined agent interface and the question user of the session corresponding to the dialogue data of the determined hit tag, so as to continue the session by the agent user of the agent interface and the question user:

[0127] The session is in the state of data communication between the robot of the intelligent customer service system and the question user in the original state. After the agent interface is determined, the system can further assign the session to the determined agent interface, establish a data communication link between the agent interface and the question user, so that the agent user of the agent interface can directly communicate with the question user in real time, thereby realizing the switching from the robot to the manual customer service, and continuing the session by the agent user and the question user.

[0128] Step S1630, pushing a notification message representing the switching agent user to the chat interface of the question user side corresponding to the session:

[0129] In order to facilitate the question user to know the response mechanism of the intelligent customer service system, a notification message representing the switching agent user can be constructed, and the notification message is pushed to the chat interface of the user side of the question user to be displayed.

[0130] This embodiment further perfects the business closed loop of the intelligent customer service system for session dispatching agent interface, allows the intelligent customer service system to configure the labeling tags for the agent interface in the background, and after the hit tags of the session are determined, the matching of the hit tags and the labeling tags is used to match the corresponding agent interface for the session. Since this process is processed in real time, the process of manually browsing the session and assigning the session to the agent user for processing by the administrator is realized, and the cost is reduced and the efficiency is improved.

[0131] Please refer to Figure 6 In the extended embodiment, after the step of determining the label corresponding to the label vector satisfying the preset correlation matching condition as the hit label according to the correlation matrix in the step S1500, the following steps are included:

[0132] Step S1700: respond to any agent user's session management request and push a session list to the agent, the session list containing mapping relationship data between all ongoing sessions of the intelligent customer service system and their hit tags:

[0133] This embodiment further provides a mechanism for agent users to select the sessions that they are suitable for handling. Thus, the agent user can initiate a session management request to the intelligent customer service system. In response to the request, the intelligent customer service system pushes a list of currently existing sessions in the system to the agent user. The session list includes all ongoing sessions of the intelligent customer service system and the mapping relationship data between the hit tags, which are displayed in the graphical user interface of the client device where the agent user is located for the agent user to view.

[0134] Step S1800: Receive a hit tag specified by the agent user, and update the conversation list so that the conversation list only includes conversations corresponding to the specified hit tag:

[0135] When an agent wants to handle a conversation corresponding to a specific hit tag, they can select a hit tag from the conversation list in their graphical user interface and submit it to the intelligent customer service system, instructing it to filter the conversations for them. The intelligent customer service system then filters the conversation list based on the specified hit tag, creating a new conversation list containing only those conversations corresponding to the specified hit tag. Similarly, the updated conversation list is also pushed to the user's graphical user interface for display.

[0136] Step S1900: In response to the agent user's request to participate in any one of the conversations in the conversation list, a data communication link is established between the agent user and the user who asked the question in the conversation, so that the agent user and the user who asked the question can continue the conversation.

[0137] Based on the conversation list, the agent user can select any conversation in the conversation list, thereby triggering a participation request to be sent to the intelligent customer service system. The intelligent customer service system then establishes a data communication link between the agent user and the user who asked the question in the conversation, thereby continuing the conversation. The agent user then answers the question from the user who asked the question, realizing manual customer service access.

[0138] It should be noted that the various steps of this embodiment can be processed concurrently with the step S1600 , and the execution of this embodiment does not need to depend on the execution of the step S1600 .

[0139] The implementation of this embodiment enables agent users to filter conversations that are suitable for their own processing, allowing them to obtain a list of conversations suitable for their own processing based on their subjectively selected hit tags. This allows agents to prioritize and take the initiative when replying to messages from consumer users, thereby improving the response efficiency of the intelligent customer service system.

[0140] See also Figure 7 A customer service conversation scheduling device provided to meet one of the purposes of this application is a functional embodiment of the customer service conversation scheduling method of this application. The device includes: a data acquisition module 1300, a semantic extraction module 1400, a label determination module 1500, and a communication establishment module 1600, wherein the data acquisition module 1300 is used to obtain the conversation data of any conversation in the intelligent customer service system, and the conversation data includes the question text generated by the questioning user of the conversation and the reply text generated by the system; the semantic extraction module 1400 is used to extract the corresponding words of each segment from the conversation data. The deep semantic information of the word vector is used to obtain the semantic vector corresponding to each word vector; the label determination module 1500 is used to calculate the correlation matrix between the label vector corresponding to each label in the preset label library and each of the semantic vectors, and determine the label corresponding to the label vector that meets the preset relevant matching conditions with the full semantic vector as the hit label according to the correlation matrix; the communication establishment module 1600 is used to connect the conversation corresponding to the conversation data with the hit label determined to the agent interface corresponding to the hit label, so that the agent user corresponding to the agent interface and the questioning user of the conversation can continue the conversation.

[0141] In a further embodiment, the semantic extraction module 1400 includes: a text splicing submodule, which is used to splice the question text and the reply text in the conversation data in an orderly manner according to the chronological order of their creation to form a conversation text; a text standardization submodule, which is used to perform preset standardization preprocessing on the conversation text to make it a standard text; a text word segmentation submodule, which is used to perform word segmentation processing on the standard text to obtain a word segmentation sequence of the standard text; a word segmentation encoding submodule, which is used to query a preset word vector table, determine the word vector corresponding to each word in the word segmentation sequence, and obtain the word vector sequence corresponding to the standard text; a semantic extraction submodule, which is used to use a text extraction model pre-trained to a convergent state to refer to context information to extract the deep semantic information of each word vector in the word vector sequence and map it to a high-dimensional space to obtain a semantic vector corresponding to each word vector.

[0142] In a further embodiment, the label determination module 1500 includes: a matrix construction submodule, which is used to use a preset correlation algorithm to calculate the correlation matrix between the label vector corresponding to each label in the preset label library and each of the semantic vectors, and the correlation matrix includes the correlation coefficient of each semantic vector mapped to each label vector; a coefficient synthesis submodule, which is used to apply a preset voting algorithm based on the correlation coefficient to calculate the comprehensive correlation coefficient corresponding to all semantic vectors mapped to each of the label vectors to represent the relevant probability of the conversation data belonging to each label vector; a label hit submodule, which is used to screen according to preset relevant matching conditions and determine that the one or more label vectors with the highest correlation probability are the label vectors most relevant to the full semantic vector, and the corresponding label is the hit label.

[0143] In an extended embodiment, the customer service session scheduling device of the present application further includes: a configuration acquisition module for acquiring label configuration information corresponding to the intelligent customer service system, wherein the label configuration information includes mapping relationship data between the label and one or more keywords; a label extraction module for encoding each keyword in the label configuration information using a label encoding model pre-trained to a convergent state, obtaining a label vector corresponding to the corresponding deep semantic information, and storing the label vector in the label library.

[0144] In a further embodiment, the communication establishment module 1600 includes: an agent determination submodule, which is used to query the annotation tags corresponding to each preset agent interface, and determine the agent interface whose annotation tags completely contain all hit tags as the agent interface corresponding to the hit tag; a link establishment submodule, which is used to establish a data communication link between the determined agent interface and the questioning user of the session corresponding to the conversation data with the hit tag, so that the agent user corresponding to the agent interface and the questioning user of the session continue the conversation; a notification push submodule, which is used to push a notification message representing the switching of the agent user to the chat interface on the questioning user side corresponding to the session.

[0145] In an extended embodiment, the customer service session scheduling device of the present application further includes: an initial push submodule, used to respond to a session management request of any agent user and push a session list to the agent, wherein the session list contains mapping relationship data between all ongoing sessions of the intelligent customer service system and their hit tags; an update push submodule, used to receive a hit tag specified by the agent user and update the session list so that the session list only contains sessions corresponding to the specified hit tag; a call switching submodule, used to respond to the agent user's request to participate in any one of the sessions in the session list, and establish a data communication link between the agent user and the questioning user of the session, so that the agent user and the questioning user can continue the session.

[0146] In order to solve the above technical problems, the embodiment of the present application also provides a computer device. Figure 8 As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions, and the database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor may implement a customer service session scheduling method. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor may execute the customer service session scheduling method of the present application. The network interface of the computer device is used to connect and communicate with the terminal. Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0147] In this embodiment, the processor is used to execute Figure 7 The memory stores the program code and various data required to execute the modules and submodules in the customer service session scheduling device. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program code and data required to execute all modules and submodules in the customer service session scheduling device of this application. The server can call the server's program code and data to execute the functions of all submodules.

[0148] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the customer service session scheduling method of any embodiment of the present application.

[0149] The present application also provides a computer program product, comprising a computer program / instruction, which implements the steps of the method described in any embodiment of the present application when executed by one or more processors.

[0150] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments of the present application can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of the method. The aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0151] To sum up, this application can comprehensively improve the intelligence level of the intelligent customer service system, realize accurate matching of conversations and agent interfaces based on the semantic correlation between the conversation data and preset tags, improve the diversion and scheduling efficiency of the intelligent customer service system, save waiting time for users asking questions, and reduce costs and increase efficiency for merchant users who configure the intelligent customer service system.

[0152] Those skilled in the art will appreciate that the steps, measures, and schemes in the various operations, methods, and processes discussed in this application may be interchanged, modified, combined, or deleted. Furthermore, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be interchanged, modified, rearranged, decomposed, combined, or deleted. Furthermore, steps, measures, and schemes in the prior art that are similar to those disclosed in this application may also be interchanged, modified, rearranged, decomposed, combined, or deleted.

[0153] The above description is only part of the implementation methods of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A customer service session scheduling method, characterized in that: Applied to e-commerce platforms, it includes the following steps: Obtaining conversation data for any conversation in the intelligent customer service system, including chat records corresponding to the full historical data of the conversation or chat records corresponding to the current question-and-answer period of the conversation; the conversation data includes the text of the question generated by the user asking the question in the conversation and the text of the reply generated by the system; Extracting deep semantic information of word vectors corresponding to each word segment from the conversation data to obtain semantic vectors corresponding to each word vector, including: using a text extraction model pre-trained to a convergent state to refer to context information, extracting deep semantic information from each word vector in the conversation data and mapping it to a high-dimensional space to obtain semantic vectors corresponding to each word vector; Calculate the correlation matrix between the label vector corresponding to each label in the preset label library and each of the semantic vectors, and determine the label corresponding to the label vector that meets the preset relevant matching conditions with the full semantic vector as the hit label based on the correlation matrix, including: using a preset correlation algorithm to calculate the correlation matrix between the label vector corresponding to each label in the preset label library and each of the semantic vectors, the correlation matrix including the correlation coefficient of each semantic vector mapped to each label vector; based on the correlation coefficient, apply a preset voting algorithm to calculate the comprehensive correlation coefficient corresponding to all semantic vectors mapped to each of the label vectors to represent the relevant probability of the conversation data belonging to each label vector; screen according to the preset relevant matching conditions, determine the one or more label vectors with the highest correlation probability as the label vector most relevant to the full semantic vector, and the label corresponding to the label is the hit label; The session corresponding to the conversation data with a determined hit tag is connected to the agent interface corresponding to the hit tag, so that the agent user corresponding to the agent interface and the questioning user of the session continue the session, including: querying the annotation tags corresponding to each preset agent interface, and determining the agent interface whose annotation tags completely contain all hit tags as the agent interface corresponding to the hit tag; establishing a data communication link between the determined agent interface and the questioning user of the session corresponding to the conversation data with a determined hit tag, so that the agent user corresponding to the agent interface and the questioning user of the session continue the session; and pushing a notification message representing the switching of agent users to the chat interface on the questioning user side corresponding to the session.

2. The customer service session scheduling method according to claim 1, characterized in that: Extracting deep semantic information of word vectors corresponding to each word segment from the conversation data to obtain semantic vectors corresponding to each word vector includes the following steps: Concatenating the question text and the reply text in the conversation data in the order of their creation to form a conversation text; Performing a preset standardization preprocessing on the dialogue text to make it into a standard text; Performing word segmentation processing on the standard text to obtain a word segmentation sequence of the standard text; Querying a preset word vector table to determine the word vector corresponding to each word in the word segmentation sequence, and obtaining a word vector sequence corresponding to the standard text; A text extraction model that has been pre-trained to a convergent state is used to extract deep semantic information from each word vector in the word vector sequence with reference to context information and mapped to a high-dimensional space to obtain a semantic vector corresponding to each word vector.

3. The customer service session scheduling method according to any one of claims 1 to 2, characterized in that: Before the step of calculating the correlation matrix between the tag vector corresponding to each tag in the preset tag library and each of the semantic vectors, the following steps are included: Obtain tag configuration information corresponding to the intelligent customer service system, wherein the tag configuration information includes mapping relationship data between a tag and one or more keywords; A label encoding model that has been pre-trained to a convergent state is used to encode each keyword in the label configuration information to obtain a label vector corresponding to the corresponding deep semantic information, and the label vector is stored in the label library.

4. The customer service session scheduling method according to any one of claims 1 to 2, characterized in that: After the step of determining, according to the correlation matrix, the label corresponding to the label vector that satisfies the preset correlation matching condition with the full semantic vector as the hit label, the following steps are included: Respond to any agent user's session management request and push a session list to the agent, which contains the mapping relationship data between all ongoing sessions of the intelligent customer service system and their hit tags; receiving a hit tag specified by the agent user, and updating the conversation list so that the conversation list only includes conversations corresponding to the specified hit tag; In response to the agent user's request to participate in any one of the conversations in the conversation list, a data communication link is established between the agent user and the questioning user of the conversation, so that the agent user and the questioning user continue the conversation.

5. A customer service conversation scheduling device, characterized in that: Applied in e-commerce platforms, including: A data acquisition module is used to acquire conversation data from any conversation in the intelligent customer service system. The conversation data of any conversation includes chat records corresponding to the full historical data of the conversation or chat records corresponding to the current question-and-answer period of the conversation; the conversation data includes the question text generated by the user asking the question in the conversation and the reply text generated by the system; A semantic extraction module is configured to extract deep semantic information of the word vector corresponding to each word segment in the conversation data, and obtain a semantic vector corresponding to each word vector, including: using a text extraction model pre-trained to a convergent state to refer to context information, extracting deep semantic information from each word vector in the conversation data, and mapping it into a high-dimensional space to obtain a semantic vector corresponding to each word vector; A tag determination module is used to calculate a correlation matrix between the tag vector corresponding to each tag in a preset tag library and each of the semantic vectors, and determine, based on the correlation matrix, the tag corresponding to the tag vector that meets the preset relevant matching conditions with the full semantic vector as a hit tag, including: using a preset correlation algorithm to calculate a correlation matrix between the tag vector corresponding to each tag in the preset tag library and each of the semantic vectors, the correlation matrix including the correlation coefficient of each semantic vector mapped to each tag vector; based on the correlation coefficient, applying a preset voting algorithm to calculate the comprehensive correlation coefficient corresponding to all semantic vectors mapped to each of the tag vectors to represent the relevant probability of the conversation data belonging to each tag vector; screening according to the preset relevant matching conditions, determining that the one or more tag vectors with the highest correlation probability are the tag vectors most relevant to the full semantic vector, and the corresponding tags are the hit tags; A communication establishment module is used to connect the session corresponding to the conversation data with a determined hit tag to the seat interface corresponding to the hit tag, so that the seat user corresponding to the seat interface and the questioning user of the session can continue the session, including: querying the annotation tags corresponding to each preset seat interface, and determining the seat interface whose annotation tags completely contain all hit tags as the seat interface corresponding to the hit tag; establishing a data communication link between the determined seat interface and the questioning user of the session corresponding to the conversation data with a determined hit tag, so that the seat user corresponding to the seat interface and the questioning user of the session can continue the session; pushing a notification message representing the switching of seat users to the chat interface on the questioning user side corresponding to the session.

6. The customer service conversation scheduling device according to claim 5, characterized in that: Also includes: A configuration acquisition module is used to obtain tag configuration information corresponding to the intelligent customer service system, wherein the tag configuration information includes mapping relationship data between a tag and one or more keywords; The label extraction module is used to encode each keyword in the label configuration information using a label encoding model pre-trained to a convergent state, obtain a label vector corresponding to the corresponding deep semantic information, and store the label vector in the label library.

7. The customer service conversation scheduling device according to claim 5, characterized in that: Also includes: The initial push submodule is used to respond to the session management request of any agent user and push a session list to the agent, which contains the mapping relationship data between all ongoing sessions of the intelligent customer service system and their hit tags; An update push submodule is configured to receive a hit tag specified by the agent user and update the conversation list so that the conversation list only includes conversations corresponding to the specified hit tag; The call switching submodule is used to respond to the agent user's request to participate in any one of the conversations in the conversation list, establish a data communication link between the agent user and the questioning user of the conversation, so that the agent user and the questioning user can continue the conversation.

8. A computer device comprising a central processing unit and a memory, characterized in that: The central processing unit is configured to call and run a computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 4.

9. A computer-readable storage medium, characterized in that It stores a computer program implemented according to the method described in any one of claims 1 to 4 in the form of computer-readable instructions, and when the computer program is called and executed by a computer, the steps included in the corresponding method are executed.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 4 are implemented.

Images