Session generation method, device and equipment and computer readable storage medium

Abstract

The invention discloses a session generation method, device and equipment and a computer readable storage medium, and is applied to the technical field of computers, and the method comprises the steps: marking a to-be-marked session by using a large model, and obtaining a marking result; counting the labeling result, and determining a session strategy according to a statistical result; and generating an outbound session according to the session policy. According to the method, the content in the session is automatically marked with the corresponding label through the large model technology, and the labeling result is obtained; and a session strategy is given by counting all labeling results, and an outbound session is generated according to the session strategy. According to the method, automatic labeling is realized through a large model, the labor cost of manual labeling and statistical strategy analysis is released, sessions in different scenes can be labeled by utilizing the method, the optimal session strategy is automatically generated according to the labeled statistical result, and the session robot can be quickly established and optimized, so that intelligent outbound is realized.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a session generation method, device, equipment and computer-readable storage medium. Background Art

[0002] When building corresponding AI (artificial intelligence) outbound calls for different outbound call scenarios, it is necessary to analyze the historical outbound call conversations in the scenario, so as to know what kind of words and outbound call strategies are more acceptable to the outbound call recipients in the corresponding scenario. Therefore, it is necessary to annotate the historical outbound call conversations, determine the optimal outbound call strategy based on the annotated results, and optimize the invitation bot (robot) based on the optimal outbound call strategy to generate more effective and reliable conversations. Currently, manual annotation is generally used, but due to the huge amount of historical outbound call data, the manual annotation method is inefficient and relies on manual experience, and manual annotation cannot quickly generate conversation strategies.

[0003] Therefore, how to improve annotation efficiency and accuracy and generate conversation strategies quickly and effectively is a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a conversation generation method, device, equipment and computer-readable storage medium, which solve the problems of low annotation efficiency and poor accuracy in the prior art, and the inability to effectively and quickly generate conversation strategies.

[0005] To solve the above technical problems, the present invention provides a session generation method, comprising:

[0006] Use the large model to annotate the conversation to be annotated and obtain the annotation result;

[0007] Performing statistics on the annotation results, and determining a conversation strategy according to the statistical results;

[0008] An outbound call session is generated according to the session strategy.

[0009] Optionally, the large model is used to annotate the session to be annotated, and the annotation results are obtained, including:

[0010] Inputting the conversation to be annotated, the matching rules and the prompt words into the large model, so that the large model structurally decomposes the conversation to be annotated according to the prompt words, matches each part of the conversation content after the decomposition with the corresponding knowledge base, and determines the annotation result of each part of the conversation content according to the matching rules;

[0011] The conversation to be annotated is a historical invitation conversation in text form in the same conversation scenario.

[0012] Optionally, before matching the disassembled parts of the conversation content with the corresponding knowledge base, the method further includes:

[0013] The historical conversation text is analyzed, and the knowledge base is established according to the analysis result, wherein the key of the knowledge base is a tag, and the value of the knowledge base is the content in the historical conversation text.

[0014] Optionally, the large model is used to structurally decompose the conversation to be annotated according to the prompt word, including:

[0015] Using the large model, preliminarily disassemble the to-be-annotated session according to the first-level disassembly node to obtain a preliminary disassembly result;

[0016] Using the large model, performing secondary disassembly on the preliminary disassembly result according to the secondary disassembly node;

[0017] The second-level disassembly node is a child node of a first-level disassembly node.

[0018] Optionally, the large model is used to structurally decompose the conversation to be annotated according to the prompt word, including:

[0019] Using the large model, the conversation to be annotated is divided into a greeting part, an opening part, an invitation strategy part, a guidance part after rejection, and a guidance operation difficulty part. The invitation strategy part includes an invitation purpose part, an invitation process part, and an invitation question type part.

[0020] Optionally, statistics are collected on the annotation results, and a session strategy is determined according to the statistical results, including:

[0021] Obtaining the marking result of a target session; the target session is a session in which the invitation is successfully completed among the sessions to be marked;

[0022] According to the splitting nodes, distributed statistics are performed on the annotation results of the target session to obtain statistical results under each splitting node;

[0023] The labeled result with the largest proportion in the statistical results is used as the conversation strategy.

[0024] Optionally, after taking the annotation result with the largest proportion in the statistical results as the session strategy, the method further includes:

[0025] When the largest-proportioned annotation result in the statistical results changes, it is determined whether the proportion of the largest-proportioned annotation result currently is greater than a preset threshold;

[0026] If it is greater, the session strategy is updated according to the annotation result with the largest current proportion.

[0027] The present invention also provides a session generation device, comprising:

[0028] Annotation module, used to annotate the session to be annotated using the large model to obtain the annotation result;

[0029] A statistical module, used to collect statistics on the annotation results and determine the conversation strategy according to the statistical results;

[0030] A generating module is used to generate an outbound call session according to the session strategy.

[0031] The present invention also provides a session generation device, comprising:

[0032] Memory for storing computer programs;

[0033] A processor is used to implement the session generation method as described above when executing the computer program.

[0034] The present invention also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, the session generation method as described above is implemented.

[0035] It can be seen that the present invention uses the big model technology to automatically mark the content in the conversation with the corresponding annotations to obtain the annotation results; and gives the conversation strategy by counting all the annotation results, and generates the outbound call conversation according to the conversation strategy. This method realizes automatic annotation through the big model, releases the manpower cost of manual annotation and analysis of statistical strategies, and can use this method to annotate conversations in different scenarios, automatically generate the optimal conversation strategy according to the annotated statistical results, and can quickly build and optimize the conversation robot, thereby realizing intelligent outbound calls.

[0036] In addition, the present invention also provides a session generation device, equipment and computer-readable storage medium, which also have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for generating a session provided by an embodiment of the present invention;

[0039] Figure 2 An example diagram of a session provided by an embodiment of the present invention;

[0040] Figure 3 A distribution diagram of an example of a greeting method provided in an embodiment of the present invention;

[0041] Figure 4 An example diagram of an automated labeling process provided by an embodiment of the present invention;

[0042] Figure 5 An example diagram of interaction provided by an embodiment of the present invention;

[0043] Figure 6 A schematic diagram of the structure of a session generation device provided by an embodiment of the present invention;

[0044] Figure 7 A schematic diagram of the structure of a session generation device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] In order to improve the outbound calling effect of BOT (intelligent outbound call system / outbound call robot / intelligent conversation system / conversational robot), it is necessary to mark the historical conversations, determine the optimal conversation strategy based on the marking results, and use the optimal conversation strategy to optimize BOT so that BOT can generate effective outbound call conversations.

[0047] Currently, conversation annotation is still in the manual annotation stage or non-automatic annotation. As the number of historical conversations reaches millions, the existing manual annotation methods have the following defects:

[0048] (1) Low efficiency: Manual labeling is slow and time-consuming, and the time it takes to produce labeling is not conducive to rapid iteration of BOT. Multiple iterations and repeated adjustments are required, which seriously affects the project progress.

[0049] (2) Poor accuracy: The manual annotation volume is limited and the coverage of the annotated sessions is small, so it is not comprehensive, which affects the decision-making of the overall outbound call strategy.

[0050] (3) Poor reliability: The objectivity of the data depends on each person’s understanding of the standards and familiarity with the business. Therefore, the cognition and standards of manual labeling may not be unified, and cognitive bias may be large, and the credibility of the data will be retained.

[0051] In order to solve the above problems, the present invention uses a large model to automatically label the conversations to be labeled. That is, for a large number of conversations to be labeled, the large model's excellent natural language understanding and generation capabilities are fully utilized to automatically label the invitation strategies used in the conversations to be labeled, freeing up the manpower for manual annotation of conversations and reducing the cost of manual annotation.

[0052] Please refer to Figure 1 , Figure 1 A flow chart of a session generation method provided by an embodiment of the present invention. The method may include:

[0053] S101: Annotate the conversation to be annotated using the large model to obtain an annotation result.

[0054] The executor of this embodiment is a terminal. This embodiment does not limit the type of terminal, as long as it can complete the operation of session generation. The large model in this embodiment is a large language model built based on deep learning technology, such as the GPT series model. This method can be applied to various vehicle scenarios that require the use of automatic annotation to generate the optimal invitation strategy, such as online car-hailing, trucks, takeout, express delivery and other invitation scenarios; or it can be applied to various telemarketing scenarios that require the use of automatic annotation to generate the optimal invitation strategy, such as product sales, market research and other scenarios that require targeted design of speech and design of outbound call strategies. It should be noted that the sessions to be annotated in this embodiment are historical sessions, that is, each session to be annotated is obtained by recycling the historical invitation sessions of BOT.

[0055] Furthermore, the above-mentioned use of the big model to annotate the conversation to be annotated to obtain the annotation results can specifically include the following steps: inputting the conversation to be annotated, matching rules and prompt words into the big model, so that the big model performs structured decomposition of the conversation to be annotated according to the prompt words, matching each part of the conversation content after the decomposition with the corresponding knowledge base, and determining the annotation results of each part of the conversation content according to the matching rules; wherein the conversation to be annotated is a historical invitation conversation in text form in the same conversation scenario.

[0056] The purpose of this step is to achieve multi-level annotation of the conversation to be annotated, that is, no longer a single label is summarized for a certain conversation, but the entire conversation to be annotated is structurally disassembled, and each part after the disassembly is annotated separately. The historical invitation conversations in each conversation scenario are automatically annotated, so that the outbound call strategy of the BOT in each conversation scenario can be optimized. Specifically, this embodiment inputs the conversation to be annotated, the matching rules and the prompt words into the big model, and uses the excellent natural language understanding and generation capabilities of the big model to automatically achieve structural disassembly and label annotation. The prompt words in this embodiment are predefined prompt words by the user. The role of the prompt words is to guide the big model to generate specific outputs or complete the task; the matching rules in this embodiment can be understood as the top TOP-K (Top-K recall is a probability-based vocabulary selection strategy used to optimize output results in natural language processing tasks) knowledge points recalled, that is, the number of knowledge points that need to be recalled. For example, when using a large model to label the invitation strategy in a conversation to be labeled, the specific prompt word may be "The following is a conversation between the inviter and the driver. Determine whether the opening remarks in this conversation are set for accepting an order. If so, return YES, otherwise return NO."

[0057] Furthermore, the above-mentioned conversation generation method may specifically include the following steps: before matching the disassembled conversation contents with the corresponding knowledge base, analyzing the historical conversation text, and establishing a knowledge base based on the analysis results, wherein the key of the knowledge base is the label, and the value of the knowledge base is the content in the historical conversation text.

[0058] This step is to build a database, analyze the conversation text, and build a database based on the analysis results. For example, in the greeting database, the greeting types (keys of the database) can include real surnames, no titles, real name titles, and popular titles. Among them, the greeting content "Hello, Master Wang" belongs to the real surname title type; the greeting content "Hello" belongs to the no title type; the greeting content "Hello, Master Wang San" belongs to the real name title type; the greeting content "Hello, Brother Wang" belongs to the popular title type.

[0059] Furthermore, the above method of making the large model perform structural decomposition of the labeled conversation according to the prompt words may specifically include the following steps:

[0060] Step 21: Using the large model, perform a preliminary decomposition of the to-be-annotated session according to the first-level decomposition nodes to obtain a preliminary decomposition result;

[0061] Step 22: Using the large model, perform secondary disassembly on the preliminary disassembly result according to the secondary disassembly node; wherein the secondary disassembly node is a child node of the first disassembly node.

[0062] To better understand this step, please refer to Figure 2 , Figure 2 The process of disassembling and analyzing the conversation is shown in the figure below. The following example shows that the initial disassembly is to perform a first-level node disassembly on the entire conversation to be annotated, for example, the conversation is split into 1.0 opening remarks, 2.1 retention, 4.1 purchase membership-rights activation guidance, 4.2 purchase membership-operation guidance, 4.3 purchase membership-waiting for operation, etc. The contents of each part after the initial split node is further analyzed in detail. For example, at the opening remarks node, it is split into the opening words, whether the real name is involved, and whether there is an order setting, etc.; at the retention no intention node, it is split into the interest points, methods, and speech structure of retention, etc. This splitting is for the large model to be able to annotate the annotated conversation in detail and obtain more detailed annotation results. If data interaction occurs between the split nodes, the input and output data must be explained in combination with the interaction steps.

[0063] Furthermore, this embodiment proposes a structured decomposition example, such as using a large model to split the conversation to be labeled into a greeting part, an opening part, an invitation strategy part, a guidance part after rejection, and a guidance operation difficulty part. The invitation strategy part includes an invitation purpose part, an invitation process part, and an invitation question type part. The content of each part after the split is analyzed to determine the label. For details, please refer to Table 1.

[0064] Table 1 Marking table

[0065]

[0066] S102: Count the annotation results and determine the conversation strategy according to the statistical results.

[0067] Furthermore, the above-mentioned statistical analysis of the annotation results and determination of the conversation strategy according to the statistical results may specifically include the following steps:

[0068] Step 31: Obtain the marking result of the target session; the target session is the session with successful invitation in the sessions to be marked;

[0069] Step 32: Perform distributed statistics on the annotation results of the target session according to the split nodes to obtain the statistical results under each split node;

[0070] Step 33: The annotation result with the largest proportion in the statistical results is used as the conversation strategy.

[0071] For ease of understanding, we will take the greeting method as an example to illustrate the statistics of various greeting types in successful invitations. The distribution chart of the statistical results can be found in Figure 3 , Figure 3This is an example diagram of the distribution of a greeting method provided by an embodiment of the present invention. The label of the greeting type with the widest distribution range, that is, the largest proportion, is used as the conversation strategy, so that the conversation strategy is optimal. Alternatively, the big model can be used to analyze and summarize the calls of the inviter with the highest invitation success rate to derive the invitation strategy.

[0072] Furthermore, the session generation method may also include the following steps: when the annotation result with the largest proportion in the statistical results changes, it is determined whether the proportion of the current annotation result with the largest proportion is greater than a preset threshold; if greater, the session strategy is updated according to the current annotation result with the largest proportion.

[0073] In order to update the BOT's session strategy in time, when the annotation result with the largest proportion in the statistical results changes, it should be updated in time. However, in order to avoid BOT abnormality or failure caused by too frequent updates, a preset threshold is set to determine whether the update condition is met. When the update condition is met, the annotation result with the largest proportion is used as the latest session strategy. The preset threshold in this embodiment can be set by the user.

[0074] S103: Generate an outbound call session according to the session policy.

[0075] This embodiment uses the conversation strategy to adjust the BOT's outbound call conversation, so that the BOT generates an appropriate outbound call conversation based on the reply strategy, thereby increasing the probability of a successful invitation. If the optimal strategy for the greeting method determined in the above 102 is to involve the real name, and the optimal strategy for the opening words is to pass the account review, then the BOT's greeting method should be optimized to "Hello, Liu Asan, Master Liu, hello", and the opening words should be optimized to "Your account has been reviewed and approved."

[0076] The conversation generation method provided by the embodiment of the present invention is applied to automatically mark the content in the conversation with the corresponding annotations through the big model technology to obtain the annotation results; and the conversation strategy is given by counting all the annotation results, and the outbound call conversation is generated according to the conversation strategy. This method realizes automatic annotation through the big model, releases the manpower cost of manual annotation and analysis and statistical strategy, and can use this method to annotate conversations in different scenarios, automatically generate the optimal conversation strategy according to the annotated statistical results, and can quickly build and optimize the conversation robot, thereby realizing intelligent outbound calls. In this way, when facing different outbound call scenarios, the BOT with the optimal outbound call speech and strategy can be quickly built; and this method can perform comprehensive and automatic annotation from the conversation structure, etc., can realize multi-level annotation, can more quickly and comprehensively disassemble the key information in the conversation, and can better assist the business in formulating the conversation strategy; and the big model can be used to uniformly process outbound call conversations in different scenarios and large quantities, and automatically annotate the key information of the outbound call conversation; after annotation, the annotation results are automatically counted, and the conversation strategy of the BOT is further optimized, so as to improve the interactive experience with users and the conversion rate.

[0077] In order to make the present invention easier to understand, please refer to Figure 4 , Figure 4 An example diagram of an automatic labeling process provided by an embodiment of the present invention may specifically include:

[0078] Convert the telephone-style invitation outbound call conversation into text form, use the big model to mark the key points of the conversation in the invitation dialogue text, and statistically analyze the distribution of the key points of the conversation. Based on the statistical results, output the invitation dialogue strategy to BOT, so that BOT can optimize the invitation outbound call based on the invitation dialogue strategy.

[0079] You can also refer to Figure 5 , Figure 5 An example diagram of interaction provided by an embodiment of the present invention, the interaction process between the BOT management module and the large model module may specifically include:

[0080] The BOT management module inputs the historical outbound call conversations (samples to be labeled) stored in itself into the big model. The big model module receives the samples to be labeled, and uses the prompt words and matching rules entered by the user to decompose the samples to be labeled (i.e., the disassembly process), and outputs various labels under all different dimensions (greeting methods, opening words, invitation strategies, guidance after rejection, and guidance operation difficulty) according to the content of each part of the conversation after the split. The distribution of the focus of the conversation output by the big model is counted, the labels are counted, and the conversation strategy is output to the BOT management module to update the outbound call conversations of the BOT management module. Among them, the split nodes are not limited to the greeting method, invitation purpose, invitation process, invitation question type, and driver question type. This implementation example directly uses the big model to fully and automatically label the outbound call conversation, and counts the optimal strategy in the final result to optimize the BOT.

[0081] The following is an introduction to a session generation device provided in an embodiment of the present invention. The session generation device described below and the session generation method described above can be referenced to each other.

[0082] Please refer to Figure 6 , Figure 6 A structural diagram of a session generation device provided by an embodiment of the present invention may include:

[0083] The annotation module 100 is used to annotate the conversation to be annotated using the large model to obtain an annotation result;

[0084] A statistics module 200 is used to collect statistics on the annotation results and determine a conversation strategy according to the statistical results;

[0085] The generating module 300 is used to generate an outbound call session according to the session strategy.

[0086] Based on the above embodiment, the marking module 100 may include:

[0087] A labeling unit is used to input the conversation to be labeled, matching rules and prompt words into the large model, so that the large model structurally decomposes the conversation to be labeled according to the prompt words, matches each part of the decomposed conversation content with the corresponding knowledge base, and determines the labeling result of each part of the conversation content according to the matching rules; wherein the conversation to be labeled is a historical invitation conversation in text form in the same conversation scenario.

[0088] Based on the above embodiment, the session generating device may further include:

[0089] The knowledge base construction module is used to analyze the historical conversation text and establish the knowledge base according to the analysis result. The key of the knowledge base is the tag, and the value of the knowledge base is the content in the historical conversation text.

[0090] Based on the above embodiment, the marking unit may include:

[0091] A first disassembly subunit is used to use the large model to preliminarily disassemble the to-be-annotated session according to the first-level disassembly node to obtain a preliminary disassembly result;

[0092] The second disassembly subunit is used to use the large model to perform secondary disassembly on the preliminary disassembly result according to the secondary disassembly node; wherein the secondary disassembly node is a child node of a disassembly node.

[0093] Based on the above embodiment, the marking unit may include:

[0094] The disassembly subunit is used to use the large model to split the conversation to be annotated into a greeting part, an opening part, an invitation strategy part, a post-rejection guidance part and a guidance operation difficulty part, wherein the invitation strategy part includes an invitation purpose part, an invitation process part and an invitation question type part.

[0095] Based on any of the above embodiments, the statistics module 200 may include:

[0096] A marking result acquisition unit, used to acquire the marking result of a target session; the target session is a session that has been successfully invited among the sessions to be marked;

[0097] A statistical unit, used to perform distributed statistics on the annotation results of the target session according to the split nodes, and obtain statistical results under each split node;

[0098] A session strategy determination unit is used to use the annotation result with the largest proportion in the statistical results as the session strategy.

[0099] Based on the above embodiment, the session generating device may further include:

[0100] A judgment module, used for judging whether the proportion of the current largest annotation result is greater than a preset threshold value when the largest annotation result in the statistical results changes;

[0101] The session strategy updating module is used to update the session strategy according to the annotation result with the largest current proportion if it is greater than.

[0102] It should be noted that the order of the modules and units in the above-mentioned session generation device can be changed without affecting the logic.

[0103] The conversation generation device provided by the embodiment of the present invention is applied, through the annotation module 100, used to use the big model to annotate the conversation to be annotated to obtain the annotation result; the statistical module 200 is used to count the annotation result and determine the conversation strategy according to the statistical result; the generation module 300 is used to generate the outbound call conversation according to the conversation strategy. This device uses the big model technology to automatically mark the content in the conversation with the corresponding annotation to obtain the annotation result; and gives the conversation strategy by counting all the annotation results, and generates the outbound call conversation according to the conversation strategy. This device realizes automatic annotation through the big model, releases the manpower cost of manual annotation and analysis of statistical strategies, and can use this device to annotate conversations in different scenarios, automatically generate the optimal conversation strategy according to the annotated statistical results, and can quickly build and optimize the conversation robot, thereby realizing intelligent outbound calls.

[0104] The following is an introduction to a session generation device provided in an embodiment of the present invention. The session generation device described below and the session generation method described above can refer to each other.

[0105] Please refer to Figure 7 , Figure 7 A structural diagram of a session generation device provided in an embodiment of the present invention may include:

[0106] A memory 10, used for storing computer programs;

[0107] The processor 20 is used to execute a computer program to implement the above-mentioned session generation method.

[0108] The memory 10 , the processor 20 , and the communication interface 31 all communicate with each other via the communication bus 32 .

[0109] In the embodiment of the present invention, the memory 10 is used to store one or more programs, and the program may include program code, and the program code includes computer operation instructions. In the embodiment of the present invention, the memory 10 may store programs for implementing the following functions:

[0110] Use the large model to annotate the conversation to be annotated and obtain the annotation result;

[0111] Collect statistics on the annotation results and determine the conversation strategy based on the statistical results;

[0112] Generate an outbound call session based on the session policy.

[0113] In a possible implementation, the memory 10 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function, etc.; the data storage area may store data created during use.

[0114] In addition, the memory 10 may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include an NVRAM. The memory stores an operating system and operating instructions, executable modules or data structures, or a subset thereof, or an extended set thereof, wherein the operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and processing hardware-based tasks.

[0115] The processor 20 may be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic device, a microprocessor or any conventional processor, etc. The processor 20 may call a program stored in the memory 10 .

[0116] The communication interface 31 may be an interface of a communication module, and is used to connect to other devices or systems.

[0117] Of course, it should be noted that Figure 7 The structure shown does not constitute a limitation on the session generating device in the embodiment of the present invention. In actual applications, the session generating device may include Figure 7 More or fewer components than shown, or combinations of certain components.

[0118] The following is an introduction to the computer-readable storage medium provided in an embodiment of the present invention. The computer-readable storage medium described below and the session generation method described above can be referenced to each other.

[0119] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned session generation method are implemented.

[0120] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0121] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0122] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0123] Finally, it should be noted that, in this article, relationships such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0124] The above is a detailed introduction to a session generation method, device, equipment and computer-readable storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for a person skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for generating a session, characterized in that: include: Use the large model to annotate the conversation to be annotated and obtain the annotation result; Performing statistics on the annotation results, and determining a conversation strategy according to the statistical results; An outbound call session is generated according to the session strategy.

2. The session generation method according to claim 1, characterized in that: Use the big model to annotate the session to be annotated and obtain the annotation results, including: Inputting the conversation to be annotated, the matching rules and the prompt words into the large model, so that the large model structurally decomposes the conversation to be annotated according to the prompt words, matches each part of the conversation content after the decomposition with the corresponding knowledge base, and determines the annotation result of each part of the conversation content according to the matching rules; The conversation to be annotated is a historical invitation conversation in text form in the same conversation scenario.

3. The session generation method according to claim 2, characterized in that: Before matching the disassembled conversation contents with the corresponding knowledge base, it also includes: The historical conversation text is analyzed, and the knowledge base is established according to the analysis result, wherein the key of the knowledge base is a tag, and the value of the knowledge base is the content in the historical conversation text.

4. The session generation method according to claim 2, characterized in that: The large model performs structural decomposition of the conversation to be annotated according to the prompt word, including: Using the large model, preliminarily disassemble the to-be-annotated session according to the first-level disassembly node to obtain a preliminary disassembly result; Using the large model, performing secondary disassembly on the preliminary disassembly result according to the secondary disassembly node; The second-level disassembly node is a child node of a first-level disassembly node.

5. The method for generating a session according to claim 2, wherein: The large model performs structural decomposition of the conversation to be annotated according to the prompt word, including: Using the large model, the conversation to be annotated is divided into a greeting part, an opening part, an invitation strategy part, a guidance part after rejection, and a guidance operation difficulty part. The invitation strategy part includes an invitation purpose part, an invitation process part, and an invitation question type part.

6. The session generation method according to any one of claims 1 to 5, characterized in that: The annotation results are counted, and a conversation strategy is determined according to the statistical results, including: Obtaining the marking result of a target session; the target session is a session in which the invitation is successfully completed among the sessions to be marked; According to the splitting nodes, distributed statistics are performed on the annotation results of the target session to obtain statistical results under each splitting node; The labeled result with the largest proportion in the statistical results is used as the conversation strategy.

7. The session generation method according to claim 6, characterized in that: After taking the largest number of the annotation results in the statistical results as the session strategy, the method further includes: When the largest-proportioned annotation result in the statistical results changes, it is determined whether the proportion of the largest-proportioned annotation result currently is greater than a preset threshold; If it is greater, the session strategy is updated according to the annotation result with the largest current proportion.

8. A conversation generating device, characterized in that: include: Annotation module, used to annotate the session to be annotated using the large model to obtain the annotation result; A statistical module, used to collect statistics on the annotation results and determine the conversation strategy according to the statistical results; A generating module is used to generate an outbound call session according to the session strategy.

9. A conversation generating device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the session generation method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by the processor, the session generation method according to any one of claims 1 to 7 is implemented.

Images