Time sequence construction method and device, equipment, medium and program product

Abstract

The embodiment of the invention provides a time sequence construction method and device, equipment, a medium and a program product, and relates to the technical field of data processing. The method comprises the following steps: identifying key nodes in dialogue data to be analyzed, and extracting user feedback content corresponding to each key node; inputting each user feedback content into a pre-trained feature recognition model to obtain doubtful attitude information corresponding to each user feedback content output by the feature recognition model; and integrating the doubtful attitude information with preset dialogue strategy information corresponding to the to-be-analyzed dialogue data, and constructing a dialogue feature time sequence corresponding to the to-be-analyzed dialogue data based on the integrated data. According to the embodiment of the invention, the doubtful attitude information of the user feedback content is extracted, and the doubtful attitude information and the preset dialogue content are integrated to generate the time sequence, so that the flexibility of a time sequence construction scheme and the comprehensiveness of information extraction are effectively improved, and the method is suitable for the field of natural language dialogues.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a time series construction method, device, equipment, medium and program product. Background Art

[0002] Time series analysis is a key technology in customer interaction scenarios such as telephone sales. Time series reflects characteristic information such as the trend of observed variables changing over time. By analyzing time series, companies can track customer behavior, predict sales trends, and optimize interaction strategies.

[0003] At present, traditional time series construction methods mainly include fixed time interval sampling and event-driven sampling. Among them, fixed time interval sampling is to collect data points at preset fixed time intervals (such as every hour, every day, etc.), and finally form a time series; this method is suitable for data collection with strong regularity, but it may miss emergencies at critical moments, affecting the integrity of the data and the accuracy of the analysis. Event-driven sampling collects data based on predefined events. In fields such as financial market analysis, this method can effectively capture the data characteristics of key events (such as transaction execution or price reaching a certain threshold, etc.), but this method relies on clearly defined events and rules, and lacks flexibility when processing unstructured data (such as natural language conversations).

[0004] In summary, the existing time series construction methods have the defects of incomplete information extraction and lack of flexibility, and are not suitable for the field of natural language dialogue. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a time series construction method, apparatus, device, medium and program product, so as to improve the applicability of the time series construction scheme in the field of natural language dialogue.

[0006] In a first aspect, an embodiment of the present application provides a method for constructing a time series, including:

[0007] Identifying key nodes in the conversation data to be analyzed, and extracting user feedback content corresponding to each key node from the conversation data to be analyzed;

[0008] Input each of the user feedback contents into a pre-trained feature recognition model to obtain the doubt attitude information corresponding to each of the user feedback contents output by the feature recognition model; wherein the feature recognition model is trained based on pre-collected user feedback content samples and their corresponding doubt attitude annotation information;

[0009] The doubt attitude information is integrated with the preset dialogue strategy information corresponding to the dialogue data to be analyzed, and a dialogue feature time series corresponding to the dialogue data to be analyzed is constructed based on the integrated data.

[0010] In an embodiment of the present application, user feedback content is extracted according to dialogue nodes and the corresponding questionable attitude information is identified using a pre-trained model, and the questionable attitude information is integrated with the preset dialogue content to generate a time series, thereby effectively improving the flexibility of the time series construction scheme and the comprehensiveness of information extraction, which is suitable for the field of natural language dialogue.

[0011] In some possible embodiments, identifying key nodes in the conversation data to be analyzed includes:

[0012] Determine the sales stage type to which the conversation data to be analyzed belongs, and perform matching based on a pre-stored recognition strategy library to obtain a target recognition strategy corresponding to the sales stage type;

[0013] According to the target identification strategy, key nodes in the conversation data to be analyzed are identified.

[0014] In the embodiment of the present application, the key nodes of the conversation data to be analyzed are identified by acquiring corresponding identification strategies according to different sales stages, thereby further improving the flexibility of constructing the time series.

[0015] In some possible embodiments, the inputting of each user feedback content into a pre-trained feature recognition model to obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model includes:

[0016] Based on the pre-stored recognition model library, matching is performed to obtain a pre-trained feature recognition model corresponding to the sales stage type, and each user feedback content is input into the feature recognition model respectively to obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model.

[0017] In the embodiment of the present application, the flexibility of constructing the time series is further improved by acquiring corresponding feature recognition models according to different sales stages to identify the doubtful attitude information corresponding to the user feedback content.

[0018] In some possible embodiments, the inputting of each user feedback content into a pre-trained feature recognition model to obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model includes:

[0019] Acquire a pre-trained feature recognition model corresponding to each key node, input each user feedback content into the corresponding feature recognition model, and obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model;

[0020] Each of the feature recognition models is trained based on user feedback content samples pre-collected according to key node types and their corresponding questionable attitude annotation information.

[0021] In the embodiment of the present application, corresponding feature recognition models are trained according to different key node types to identify questionable attitude information of user feedback content of corresponding key nodes, thereby further improving the flexibility of constructing time series.

[0022] In some possible embodiments, the key nodes include dialogue nodes and operation nodes;

[0023] The step of integrating each of the doubt attitude information with the preset dialogue strategy information corresponding to the dialogue data to be analyzed includes:

[0024] Extracting preset machine dialogue content corresponding to each dialogue node based on the dialogue data to be analyzed as the preset dialogue strategy information corresponding to the dialogue data to be analyzed;

[0025] The doubt attitude information is integrated with the preset dialogue strategy information.

[0026] In the embodiment of the present application, the flexibility of constructing the time series is further improved by identifying the dialogue nodes in the dialogue data and obtaining the corresponding machine dialogue content as the dialogue strategy information integrated with the question attitude information.

[0027] In some possible embodiments, before identifying the key nodes in the conversation data to be analyzed, the method further includes:

[0028] Acquire an initial conversation data set, and select target conversation data that meets preset conditions from the initial conversation data set as the conversation data to be analyzed;

[0029] Among them, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set whose conversation duration exceeds the preset threshold; or, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set with a preset target key node type; or, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set whose key node number exceeds the preset number threshold.

[0030] In the embodiment of the present application, by first preliminarily screening all conversation data according to preset rules before analyzing the conversation data, unnecessary data analysis processes are reduced, thereby further improving the flexibility of constructing time series.

[0031] In a second aspect, an embodiment of the present application provides a time series construction device, including:

[0032] A content extraction module, used to identify key nodes in the conversation data to be analyzed, and extract user feedback content corresponding to each key node from the conversation data to be analyzed;

[0033] An information recognition module, used to input each of the user feedback contents into a pre-trained feature recognition model, and obtain the doubt attitude information corresponding to each of the user feedback contents output by the feature recognition model; wherein the feature recognition model is trained based on pre-collected user feedback content samples and their corresponding doubt attitude annotation information;

[0034] The sequence construction module is used to integrate each of the doubt attitude information with the preset dialogue strategy information corresponding to the dialogue data to be analyzed, and construct a dialogue feature time sequence corresponding to the dialogue data to be analyzed based on the integrated data.

[0035] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the method described in any embodiment of the first aspect when executing the program.

[0036] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in any embodiment of the first aspect can be implemented.

[0037] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a computer program, wherein when the computer program is executed by a processor, the method described in any embodiment of the first aspect can be implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1A schematic diagram of a flow chart of a method for constructing a time series provided in an embodiment of the present application;

[0040] Figure 2 A schematic diagram of the structure of a time series construction device provided in an embodiment of the present application;

[0041] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0043] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0044] It should be noted that existing time series construction schemes usually adopt fixed time interval sampling, event-driven sampling and random sampling methods, but these schemes either rely on data sets with strong regularity or rely on clearly defined events and rules, so they are not applicable to the field of natural language dialogue. Specifically, they have the following limitations: 1. Sampling at fixed time intervals and predefined events may not be able to capture the conversation dynamics that have a significant impact on sales results in a timely manner. 2. Important non-preset events (such as changes in customer emotions, the raising of key questions, etc.) may not be recorded in the time series. 3. When processing unstructured data such as telephone conversations, it is impossible to effectively identify and utilize all the key information in the conversation.

[0045] In response to the problems existing in the above-mentioned prior art, an embodiment of the present application provides a method for constructing a time series, which constructs a time series by training a large model to identify user doubts and attitude information in the conversation content, and integrating it with the preset machine conversation, thereby effectively improving the applicability of the time series process construction scheme in the field of natural language conversation.

[0046] like Figure 1 As shown, the embodiment of the present application provides a method for constructing a time series, which may include the steps of:

[0047] S1. Identify key nodes in the conversation data to be analyzed, and extract user feedback content corresponding to each key node from the conversation data to be analyzed.

[0048] It is understandable that the embodiments of the present application are mainly aimed at human-computer dialogue scenarios, especially automated dialogue systems in telephone sales. Specifically, the dialogue data to be analyzed is dialogue data generated by a call between an automated dialogue system and a certain customer. The dialogue data to be analyzed can be in text form and include timestamp information corresponding to the text content of the dialogue.

[0049] For the currently acquired conversation data to be analyzed, the key nodes are first identified. It can be understood that the key nodes are pre-set specific process nodes in the conversation process, such as "send insurance", "open WeChat", "search box" and "public account", etc. These nodes refer to the specific links that the AI ​​robot guides the user to reach during the conversation process.

[0050] After acquiring the conversation data formed by the automatic conversation, keyword matching can be performed according to the text content in the conversation data to be analyzed to identify the key content therein, and the timestamp corresponding to the key content can be marked as a key node.

[0051] Then, the corresponding user feedback content is extracted based on the timestamp of the key node. In the formed human-computer dialogue data, there will be a sentence belonging to the customer's reply at each key node, such as the content answered at a key node or the question raised at a key node, etc. These dialogue contents generated by the customer are the user feedback content corresponding to each key node. Exemplarily, according to the timestamp of each key node, one or more dialogue contents generated by the customer after the key node can be determined as the corresponding user feedback content; or one or more dialogue contents generated by the customer between two key nodes can be determined as the corresponding user feedback content.

[0052] S2. Input each user feedback content into a pre-trained feature recognition model to obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model; wherein the feature recognition model is trained based on the pre-collected user feedback content samples and their corresponding doubt attitude annotation information;

[0053] Specifically, a pre-trained feature recognition model is used to perform feature recognition on each segment of extracted user feedback content to obtain corresponding doubt attitude information (reflecting the user's doubt and attitude at the corresponding key nodes).

[0054] Among them, the feature recognition model can be obtained by training the large model with labeled data; the labeled data can be obtained by using the historical conversation data to extract the user feedback content corresponding to each node as samples in the same way, and manually annotating these user feedback content samples with user doubts and attitudes based on business experience to obtain labeled data for training the model.

[0055] For example, in a certain conversation data, at a node where an insurance planner introduces an insurance product, the user feedback content is "What operation is required for this?", then the user attitude of the user feedback content can be marked as "neutral", and the user doubt can be marked as "asking how to operate".

[0056] S3. Integrate each doubt attitude information with preset dialogue strategy information corresponding to the dialogue data to be analyzed, and construct a dialogue feature time series corresponding to the dialogue data to be analyzed based on the integrated data.

[0057] Specifically, first obtain the preset dialogue strategy information corresponding to the current dialogue data to be analyzed. The preset dialogue strategy information can be the dialogue content automatically executed by the robot according to preset rules, or the dialogue strategy corresponding to the dialogue content (the dialogue content is matched with the set dialogue strategy).

[0058] It can be understood that the question attitude information corresponding to each key node is essentially a time series with a timestamp, and each section of preset dialogue strategy information also carries timestamp information corresponding to the key node in the dialogue data to be analyzed.

[0059] Then, based on the call start time, the doubt attitude information and the dialogue strategy information are integrated in chronological order to obtain integrated data. Finally, based on the integrated data, a dialogue feature time series corresponding to the dialogue data to be analyzed is constructed according to the timestamp information.

[0060] In the embodiment of the present application, user feedback content is extracted according to the dialogue node and the corresponding question attitude information is identified using a pre-trained model. Then, the question attitude information is integrated with the preset dialogue content to generate a time series, thereby effectively improving the flexibility of the time series construction scheme and the comprehensiveness of information extraction, which is suitable for the field of natural language dialogue.

[0061] In some possible embodiments, in step S1, identifying key nodes in the conversation data to be analyzed includes:

[0062] S101, determining the sales stage type to which the conversation data to be analyzed belongs, and matching based on a pre-stored recognition strategy library to obtain a target recognition strategy corresponding to the sales stage type;

[0063] S102: Identify key nodes in the conversation data to be analyzed according to the target identification strategy.

[0064] It should be noted that in the field of telephone sales, different sales stages can be pre-defined according to actual needs, and different sales strategies and links can be adopted for different sales stages. For example, the following three sales stages can be designed in the insurance product sales scenario: 1. "0call Rubik's Cube" stage: This stage is the first contact with the user, with the main sales purpose of giving free insurance to the user; 2. "1call Rubik's Cube": This stage is to contact the users retained by "0call Rubik's Cube" for the second time, with the main purpose of selling short-term insurance Rubik's Cube products; 3. "2call Rubik's Cube": This stage is to resell to high-quality users who are quantitatively screened after 1call Rubik's Cube.

[0065] It is understandable that the content of telephone conversations at different sales stages and types will have different key node types, so different key node identification strategies can be used to identify nodes in the conversation data to be analyzed.

[0066] Specifically, first determine the sales stage type to which the conversation data to be analyzed belongs, and then match the target recognition strategy corresponding to the sales stage type in the pre-stored recognition strategy library, where the recognition strategy library stores recognition strategies corresponding to different sales stage types; finally, identify the key nodes of the conversation data to be analyzed according to the matched target recognition strategy.

[0067] In some possible embodiments, step S2, inputting each user feedback content into a pre-trained feature recognition model to obtain the question attitude information corresponding to each user feedback content output by the feature recognition model, may include:

[0068] S201, matching based on a pre-stored recognition model library and obtaining a pre-trained feature recognition model corresponding to the sales stage type, and inputting each user feedback content into the feature recognition model respectively, to obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model.

[0069] It should be noted that since the sales conversation content of different sales stage types has different key nodes, and the user feedback content corresponding to different key nodes will also have different characteristics, different feature recognition models can be trained to identify doubtful attitude information in the user feedback content of conversation data belonging to different sales stages.

[0070] Specifically, we first collect corresponding conversation data according to the sales stage type, identify key nodes and extract user feedback content from these conversation data, then annotate these user feedback contents with information about user doubts and attitudes, and finally train the model based on the annotated information and user feedback content to obtain feature recognition models corresponding to each sales stage type, and store these feature recognition models in the recognition model library.

[0071] In actual application, according to the type of sales stage to which the data to be analyzed belongs, a corresponding feature recognition model can be matched from the pre-stored recognition model library to respectively analyze the doubt attitude information of the user feedback content of the current dialogue data.

[0072] Based on this, by acquiring corresponding feature recognition models according to different sales stages to identify the doubtful attitude information corresponding to the user feedback content, the accuracy of the doubtful attitude information recognition can be further improved.

[0073] In some possible embodiments, step S2, inputting each user feedback content into a pre-trained feature recognition model to obtain the question attitude information corresponding to each user feedback content output by the feature recognition model, may include:

[0074] S211, obtaining a pre-trained feature recognition model corresponding to each key node, inputting each user feedback content into the corresponding feature recognition model, and obtaining the doubt attitude information corresponding to each user feedback content output by the feature recognition model;

[0075] Among them, each feature recognition model is trained based on user feedback content samples collected in advance according to key node types and their corresponding question attitude annotation information.

[0076] It should be noted that for different key node types, the corresponding user feedback content will have different characteristics. Therefore, feature recognition models can be trained for different key nodes to identify doubtful attitude information.

[0077] Specifically, different conversation data are first collected and key nodes are identified. Then, the corresponding user feedback content is collected according to the key node type. Then, the user doubt attitude information is annotated on these user feedback contents. Finally, model training is performed based on the annotated information and the corresponding user feedback content to obtain the feature recognition models corresponding to each key node type, and these feature recognition models are stored in the recognition model library.

[0078] In actual application, according to the different key node types identified by the data to be analyzed, the feature recognition model corresponding to each key node can be matched from the pre-stored recognition model library, and the user feedback content of the key node can be used to identify the doubtful attitude information using the matched feature recognition model.

[0079] Based on this, by training corresponding feature recognition models according to different key node types, the accuracy of doubtful attitude information recognition can be further improved by using them to identify the doubtful attitude information of user feedback content of the corresponding key nodes.

[0080] In some possible embodiments, key nodes may include dialog nodes and operation nodes;

[0081] In step S3, each piece of doubt attitude information is integrated with the preset dialogue strategy information corresponding to the dialogue data to be analyzed, which may include:

[0082] S301, extracting preset machine dialogue content corresponding to each dialogue node based on the dialogue data to be analyzed as preset dialogue strategy information corresponding to the dialogue data to be analyzed;

[0083] S302: Integrate each question attitude information with preset dialogue strategy information.

[0084] It should be noted that in the sales process, the key nodes identified based on dialogue data can be further divided into dialogue nodes and operation nodes. Among them, the operation node refers to the node where a specific operation is performed through dialogue with the user, such as "enter activation code", "submit order", "payment" and "upgrade", etc.; the dialogue node refers to the node where the AI ​​robot executes a specific dialogue based on the preset dialogue strategy.

[0085] Based on this, when identifying the key nodes of the dialogue data to be analyzed, the node can also be distinguished and identified as a dialogue node or an operation node. For the identified dialogue nodes, the preset machine dialogue content corresponding to each dialogue node can be extracted from the dialogue data to be analyzed as the preset dialogue strategy information corresponding to the dialogue data to be analyzed.

[0086] In some possible embodiments, before step S1, the following steps may also be included:

[0087] S100, obtaining an initial conversation data set, and screening out target conversation data that meets preset conditions from the initial conversation data set as conversation data to be analyzed;

[0088] Among them, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set whose conversation duration exceeds the preset threshold; or, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set with a preset target key node type; or, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set whose key node number exceeds the preset number threshold.

[0089] It should be noted that before identifying key nodes, the conversation data may be screened first according to needs to reduce the amount of data that needs to be processed.

[0090] Specifically, the conditions for filtering data can be set according to needs, and exemplarily may include the following: 1. Setting a conversation duration threshold, if the conversation duration threshold is exceeded, it is determined to meet the preset conditions; 2. Setting a specific key node type, if the conversation data is identified to contain the corresponding key node type, it is determined to meet the preset conditions; 3. Setting a key node quantity threshold, if the number of key nodes identified in the conversation data exceeds the quantity threshold, it is determined to meet the preset conditions.

[0091] Based on the currently acquired initial conversation data set, each initial conversation data is judged according to each preset condition, and the conversation data that meets the preset conditions is screened out as the conversation data to be analyzed that needs to be processed, so as to construct a corresponding time series for the conversation data to be analyzed based on the method of the above steps S1 to S3.

[0092] It should be noted that after constructing a time series for a large amount of dialogue data, using the collected time series as the basis for data analysis, further in-depth analysis and processing can identify the impact of different dialogue strategies on user doubt attitudes, thereby continuously improving and optimizing dialogue strategies and processes.

[0093] Please refer to Figure 2 , Figure 2 FIG. 1 is a block diagram showing a time series construction device provided in some embodiments of the present application. It should be understood that the time series construction device is similar to the above-mentioned Figure 1 Corresponding to the method embodiment, it is able to execute each step involved in the above method embodiment. The specific functions of the time series construction device can be found in the description above. To avoid repetition, the detailed description is appropriately omitted here.

[0094] Figure 2 The time series construction device includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the time series construction device, and the time series construction device includes:

[0095] The content extraction module 210 is used to identify key nodes in the conversation data to be analyzed, and extract user feedback content corresponding to each key node from the conversation data to be analyzed;

[0096] The information recognition module 220 is used to input each user feedback content into a pre-trained feature recognition model to obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model; wherein the feature recognition model is trained based on the pre-collected user feedback content samples and their corresponding doubt attitude annotation information;

[0097] The sequence construction module 230 is used to integrate each of the doubt attitude information with the preset dialogue strategy information corresponding to the dialogue data to be analyzed, and construct a dialogue feature time sequence corresponding to the dialogue data to be analyzed based on the integrated data.

[0098] It can be understood that the above-mentioned device item embodiment corresponds to the method item embodiment of the present invention. A time series construction device provided by the embodiment of the present invention can implement the time series construction method provided by any method item embodiment of the present invention.

[0099] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method, and will not be described in detail here.

[0100] like Figure 3 As shown, some embodiments of the present application provide an electronic device 300, which includes: a memory 310, a processor 320, and a computer program stored in the memory 310 and executable on the processor 320, wherein the processor 320 reads the program from the memory 310 through the bus 330 and executes the program to implement the method of any embodiment included in the above-mentioned method for constructing a time series.

[0101] Processor 320 can process digital signals and can include various computing structures, such as complex instruction set computer structure, reduced instruction set computer structure, or a structure that implements a combination of multiple instruction sets. In some examples, processor 320 can be a microprocessor.

[0102] The memory 310 may be used to store instructions executed by the processor 320 or data related to the execution of instructions. These instructions and / or data may include codes for implementing some or all functions of one or more modules described in the embodiments of the present application. The processor 320 of the disclosed embodiment may be used to execute instructions in the memory 310 to implement the method shown above. The memory 310 includes a dynamic random access memory, a static random access memory, a flash memory, an optical memory, or other memory known to those skilled in the art.

[0103] Some embodiments of the present application further provide a computer-readable storage medium having a computer program stored thereon. The computer program is executed by a processor to execute the method described in the method embodiment.

[0104] Some embodiments of the present application further provide a computer program product, which, when executed on a computer, enables the computer to execute the method described in the method embodiment.

[0105] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0106] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0107] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0108] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as 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.

[0109] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0110] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0111] It should be noted that, in this article, relational terms such as first and second, etc. are only used 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", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including 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. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for constructing a time series, characterized in that: include: Identifying key nodes in the conversation data to be analyzed, and extracting user feedback content corresponding to each key node from the conversation data to be analyzed; Input each of the user feedback contents into a pre-trained feature recognition model to obtain the doubt attitude information corresponding to each of the user feedback contents output by the feature recognition model; wherein the feature recognition model is trained based on pre-collected user feedback content samples and their corresponding doubt attitude annotation information; The doubt attitude information is integrated with the preset dialogue strategy information corresponding to the dialogue data to be analyzed, and a dialogue feature time series corresponding to the dialogue data to be analyzed is constructed based on the integrated data.

2. The method for constructing a time series according to claim 1, characterized in that: The identifying key nodes in the conversation data to be analyzed includes: Determine the sales stage type to which the conversation data to be analyzed belongs, and perform matching based on a pre-stored recognition strategy library to obtain a target recognition strategy corresponding to the sales stage type; According to the target identification strategy, key nodes in the conversation data to be analyzed are identified.

3. The method for constructing a time series according to claim 2, characterized in that: The step of inputting each of the user feedback contents into a pre-trained feature recognition model to obtain the doubt attitude information corresponding to each of the user feedback contents output by the feature recognition model includes: Based on the pre-stored recognition model library, matching is performed to obtain a pre-trained feature recognition model corresponding to the sales stage type, and each user feedback content is input into the feature recognition model respectively to obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model.

4. The method for constructing a time series according to claim 1, characterized in that: The step of inputting each of the user feedback contents into a pre-trained feature recognition model to obtain the doubt attitude information corresponding to each of the user feedback contents output by the feature recognition model includes: Acquire a pre-trained feature recognition model corresponding to each key node, input each user feedback content into the corresponding feature recognition model, and obtain the doubt attitude information corresponding to each user feedback content output by the feature recognition model; Each of the feature recognition models is trained based on user feedback content samples pre-collected according to key node types and their corresponding questionable attitude annotation information.

5. The method for constructing a time series according to claim 1, characterized in that: The key nodes include dialogue nodes and operation nodes; The step of integrating each of the doubt attitude information with the preset dialogue strategy information corresponding to the dialogue data to be analyzed includes: Extracting preset machine dialogue content corresponding to each dialogue node based on the dialogue data to be analyzed as the preset dialogue strategy information corresponding to the dialogue data to be analyzed; The doubt attitude information is integrated with the preset dialogue strategy information.

6. The method for constructing a time series according to claim 1, characterized in that: Before identifying the key nodes in the conversation data to be analyzed, the method further includes: Acquire an initial conversation data set, and select target conversation data that meets preset conditions from the initial conversation data set as the conversation data to be analyzed; Among them, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set whose conversation duration exceeds the preset threshold; or, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set with a preset target key node type; or, the target conversation data that meets the preset conditions is the conversation data in the initial conversation data set whose key node number exceeds the preset number threshold.

7. A time series construction device, characterized in that: include: A content extraction module, used to identify key nodes in the conversation data to be analyzed, and extract user feedback content corresponding to each key node from the conversation data to be analyzed; An information recognition module, used to input each of the user feedback contents into a pre-trained feature recognition model, and obtain the doubt attitude information corresponding to each of the user feedback contents output by the feature recognition model; wherein the feature recognition model is trained based on pre-collected user feedback content samples and their corresponding doubt attitude annotation information; The sequence construction module is used to integrate each of the doubt attitude information with the preset dialogue strategy information corresponding to the dialogue data to be analyzed, and construct a dialogue feature time sequence corresponding to the dialogue data to be analyzed based on the integrated data.

8. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor can implement the time series construction method described in any one of claims 1 to 6 when executing the program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing a time series according to any one of claims 1 to 6 is executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method for constructing a time series according to any one of claims 1 to 6 is implemented.

Images