Intention label determination method and device, text processing method and device, equipment and medium

Abstract

The invention provides an intention label determination method and device, a text processing method and device, equipment and a medium. The method comprises the steps of obtaining a dialogue text set and historical dialogue texts of multiple dialogue texts in the dialogue text set; according to a historical dialogue text of each dialogue text, determining a target intention label of each dialogue text and a representation vector of the target intention label; classifying the multiple target intention tags according to the representation vectors of the multiple target intention tags to obtain multiple intention classification tag groups; grouping intention labels corresponding to all the intention classification label groups are determined, the grouping intention labels are used for updating intention label system data of an intention recognition model, and the intention label system data are used for conducting intention recognition on the user dialogue text. According to the embodiment of the invention, the accuracy of determining the target intention tag can be improved, the manual annotation cost is reduced, and the multi-level intention tag system data is constructed.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to an intent label determination method, a text processing method and apparatus, an electronic device, and a computer-readable storage medium. Background Technology

[0002] With the rapid development of artificial intelligence (AI) technology, large language models (LLMs) have become one of the core technologies driving the widespread application of AI. Training large language models often requires a large amount of training data. In scenarios involving interactive dialogue with users, large language models are typically used as intent recognition models. These models need to identify the user's intent in the input text and respond accordingly. Before applying an intent recognition model, a large amount of user intent data needs to be collected beforehand to train the model. This ensures the trained model can accurately identify the user's intent in the input text and respond accordingly. However, predefined user intents often fail to meet user needs. Therefore, it is necessary to mine new intents from unannotated user dialogue text to train the intent recognition model. Summary of the Invention

[0003] This disclosure provides an intent label determination method, a text processing method and apparatus, an electronic device, a computer-readable storage medium, and a computer program product.

[0004] Firstly, this disclosure provides a method for determining intent labels, including:

[0005] Obtain the dialogue text set and the historical dialogue text of multiple dialogue texts in the dialogue text set;

[0006] Based on the historical dialogue text of each dialogue text, determine the target intent label and the representation vector of the target intent label for each dialogue text;

[0007] Based on the representation vectors of the multiple target intent labels, the multiple target intent labels are classified to obtain multiple intent classification label groups;

[0008] Each intent classification label group is determined separately, wherein the group intent label is used to update the intent label system data of the intent recognition model, and the intent label system data is used to perform intent recognition on user dialogue text.

[0009] Secondly, this disclosure provides a text processing method, including:

[0010] Obtain the user's dialogue text and the historical user dialogue text of the user's dialogue text;

[0011] The user dialogue text, the historical user dialogue text, and the current intent tag system data are input into the intent recognition model to obtain the user dialogue intent, wherein the intent tag system data is determined according to the intent tag determination method described above.

[0012] Based on the user's dialogue intent, generate a response text to the user's dialogue text.

[0013] Thirdly, this disclosure provides an intent tag determination apparatus, comprising:

[0014] The acquisition module is configured to acquire a dialogue text set and historical dialogue texts of multiple dialogue texts in the dialogue text set;

[0015] The first determining module is configured to determine the target intent tag and the representation vector of the target intent tag for each dialogue text based on the historical dialogue text of each dialogue text.

[0016] The classification module is configured to classify the multiple target intent labels based on the representation vectors of the multiple target intent labels to obtain multiple intent classification label groups;

[0017] The second determining module is configured to determine the grouped intent label corresponding to each of the intent classification label groups, wherein the grouped intent label is used to update the intent label system data of the intent recognition model, and the intent label system data is used to perform intent recognition on the user dialogue text.

[0018] Fourthly, this disclosure provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores one or more computer programs executable by the at least one processor, the one or more computer programs being executed by the at least one processor to enable the at least one processor to perform the aforementioned intent tag determination method or text processing method.

[0019] Fifthly, this disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described intent tag determination method or text processing method.

[0020] In a sixth aspect, this disclosure provides a computer program product that includes computer-readable code or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the aforementioned intent tag determination method or text processing method.

[0021] The intent tag determination method provided in this disclosure can acquire a user's dialogue text set and historical dialogue texts of multiple dialogue texts within the dialogue text set. It determines the target intent tag and its representation vector for each dialogue text using the historical dialogue text. Based on the representation vector, it classifies the target intent tag, obtaining multiple intent category tag groups. Furthermore, it determines the grouped intent tags corresponding to each intent category tag group. This method not only determines the target intent tag for each dialogue text but also mines the grouped intent tags corresponding to the intent category tag group to which the target intent tag belongs. By combining the historical dialogue text of the dialogue text to determine the target intent tag, it achieves the goal of completing the target intent tag of the dialogue text with the context of the dialogue text, improving the accuracy of target intent tag determination. It eliminates the need for manual annotation of each dialogue text, reducing manual annotation costs. It can directly perform intent recognition on unannotated dialogue text sets, achieving intent tag mining from scratch, and further determining the grouped intent tags of intent category tag groups. This method can be used to construct a multi-level intent tag system data, enabling granular mining between different intent tags.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0023] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the embodiments of the present disclosure to explain the disclosure and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the detailed description of exemplary embodiments with reference to the accompanying drawings, in which:

[0024] Figure 1 This is a flowchart of an intent label determination method provided in an embodiment of this disclosure;

[0025] Figure 2 This is a flowchart illustrating how to determine a target intent label and a representation vector, as provided in an embodiment of this disclosure.

[0026] Figure 3This is a schematic diagram illustrating how to determine whether a first prompt word meets the conditions of an input intent recognition model, as provided in an embodiment of this disclosure.

[0027] Figure 4 This is a flowchart of an intent tag determination method provided in an embodiment of this disclosure;

[0028] Figure 5 This is a block diagram of an intent label determination device provided in an embodiment of this disclosure;

[0029] Figure 6 This is a flowchart of a text processing method provided in an embodiment of this disclosure;

[0030] Figure 7 This is a block diagram of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0031] To enable those skilled in the art to better understand the technical solutions of this disclosure, exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of this disclosure to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0032] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0033] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0034] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded. Words such as “connected” or “linked” are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect.

[0035] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined herein.

[0036] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information in this technical solution comply with relevant laws and regulations and do not violate public order and good morals. The use of user data in this technical solution follows relevant national laws and regulations (e.g., the "Information Security Technology - Personal Information Security Specification"). For example, appropriate measures are taken for personal information access control; restrictions are imposed on the display of personal information; the purpose of using personal information does not exceed the scope of direct or reasonable association; and explicit identity targeting is eliminated when using personal information to avoid precisely identifying specific individuals.

[0037] Currently, the training of large-scale language models often requires a large amount of training data. In interactive dialogue scenarios, large-scale language models are typically used as intent recognition models. These models need to identify the user's intent in the input text and respond accordingly. Before applying an intent recognition model, a large amount of user intent data needs to be collected beforehand to train the model, ensuring it can accurately identify the user's intent in the input text and respond accordingly. However, predefined user intents often fail to meet user needs. Intent mining based solely on single sentences of dialogue text results in incomplete intents. Furthermore, manual annotation of the dialogue text requires a large amount of text, leading to high manual annotation costs.

[0038] Based on this, the present disclosure provides an intent label determination method, an intent label determination device, a text processing method, an electronic device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0039] Both the intent tag determination method and the text processing method according to embodiments of this disclosure can be executed by electronic devices such as terminal devices or servers. The terminal device can be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, wearable device, etc. The server can be an independent physical server, a server cluster composed of multiple physical servers, or a cloud server capable of cloud computing. This method can be implemented by a processor calling computer-readable program instructions stored in memory.

[0040] See Figure 1 , Figure 1 A flowchart of an intent label determination method according to an embodiment of the present disclosure is shown, which specifically includes the following steps:

[0041] Step 102: Obtain the dialogue text set and the historical dialogue texts of multiple dialogue texts in the dialogue text set.

[0042] The dialogue text set refers to the collection of dialogue texts between users and intelligent robots, intelligent customer service, etc., without intent tags; the dialogue text set includes multiple dialogue texts. Historical dialogue texts refer to the dialogue texts of all previous dialogue rounds preceding each dialogue text. For example, if dialogue text S1 is a dialogue text in dialogue text set S, and the user and intelligent customer service conducted 5 rounds of dialogue in a complete conversation, then dialogue text S1 is the dialogue text of the 5th round in this complete conversation, and the dialogue texts of the 1st to 4th rounds are the historical dialogue texts of dialogue text S1.

[0043] Specifically, it retrieves the unlabeled dialogue text set and the historical dialogue text for each dialogue text in the dialogue text set.

[0044] Step 104: Based on the historical dialogue text of each dialogue text, determine the target intent tag and the representation vector of the target intent tag for each dialogue text.

[0045] After obtaining the dialogue text set and the historical dialogue text of each dialogue text in the dialogue text set, context analysis can be performed on the historical dialogue text of each dialogue text to determine the target intent tag for each dialogue text. The target intent tag refers to the intent tag of the dialogue text, used to represent the purpose of the user's input dialogue text. The representation vector is the vectorized expression of the target intent tag.

[0046] Specifically, based on the acquired dialogue text and its historical dialogue text, prompt words can be constructed to determine the target intent label of the dialogue text. Then, an intent recognition model is used to identify the target intent label and its representation vector based on the constructed prompt words. This eliminates the need for manual annotation to determine the target intent label of the dialogue text, reducing the cost of manual annotation. Using an intent recognition model to determine the target intent label of the dialogue text improves the accuracy and efficiency of determining the target intent label.

[0047] Based on this, in a specific embodiment provided in this disclosure, the target intent tag and the representation vector of the target intent tag are determined for each dialogue text according to the historical dialogue text of each dialogue text, including:

[0048] Get the first prompt word template;

[0049] Each dialogue text and its historical dialogue text are filled into the first prompt word template to generate a first prompt word corresponding to each dialogue text.

[0050] Each of the first prompt words is input into the intent recognition model to determine the target intent label and the representation vector of the target intent label for each dialogue text.

[0051] The first prompt word template refers to the template used to construct the prompt word for determining the target intent label of the dialogue text; the first prompt word refers to the prompt word used to determine the target intent label of the dialogue text; the intent recognition model is used to identify the target intent label of the dialogue text, identify grouped intent labels with commonalities among multiple target intent labels, and update the intent label system data; the intent recognition model can be a large-scale language model (LLM), a BERT model (Bidirectional Encoder Representations from Transformers), a Transformer model, etc., in practical applications, and can be determined according to the actual application situation. This disclosure does not limit the type of intent recognition model.

[0052] See Figure 2 , Figure 2A flowchart illustrating a method for determining a target intent label and a representation vector according to an embodiment of this disclosure is shown. Specifically, a pre-configured first prompt word template is obtained. Each dialogue text and its historical dialogue text are filled into the corresponding positions in the first prompt word template to generate a first prompt word for each dialogue text. The first prompt word is input into an intent recognition model. During the processing of the input first prompt word, the intent recognition model generates a representation vector for the target intent label. At this point, the representation vector of the target intent label can be obtained. After processing the representation vector, the intent recognition model outputs the text content of the representation vector, i.e., the target intent label. Thus, the target intent label and the representation vector of the target intent label for each dialogue text can be obtained.

[0053] The following is an example of a first prompt word template:

[0054] #Complete Cold Start Prompt

[0055] ---------System Prompt---------

[0056] #roledefinition

[0057] You are a conversation detection expert, skilled at analyzing and detecting the conversation content between User A and User B.

[0058] #task definition

[0059] Your task is to refer to historical dialogue texts and learn to analyze the intentions behind User B's utterances in the current dialogue text.

[0060] #data definition

[0061] 1. <context>It stores the historical dialogue text of the current dialogue text;

[0062] 2. <conversation>< / conversation> The text contains UserB's statements in the current dialogue, and we need to analyze UserB's intentions based on this text.

[0063] #task requirements

[0064] 1. The output UserB intent should not contain a specific name; instead, it should be referred to by its entity category.

[0065] 2. The UserB intent output should not contain a specific time; use "time" to refer to it. ...

[0067] ---------User Prompt---------

[0068] <context>

[0069] UserA:xxxxx

[0070] UserB:xxxxx

[0071] UserA:xxxxx

[0072] < / context>

[0073] <conversation>

[0074] UserB:xxxxx

[0075] < / conversation>

[0076] #generate mark

[0077] refer to <context>< / context> From the historical dialogue text, we can know <conversation>< / conversation> The intent of UserB in the current dialogue text is:

[0078] In practical applications, the obtained dialogue text is filled into the aforementioned first prompt word template. <conversation>< / conversation> In the middle, the historical dialogue text of the dialogue text is filled into the above-mentioned first prompt word template. <context>< / context> The first prompt word can be generated in the process, and the intent of User B output by the intent recognition model is the target intent tag. It should be noted that the above first prompt word template is only an example given in the embodiments of this disclosure. In actual applications, different first prompt word templates can be pre-configured according to the actual application situation.

[0079] This embodiment of the disclosure analyzes and processes the historical dialogue text to determine the target intent tag of the dialogue text, and identifies the target intent tag of the dialogue text through an intent recognition model based on the construction of the first prompt word, thereby improving the accuracy of identifying the target intent tag.

[0080] Furthermore, to ensure the accuracy of the target intent label output by the intent recognition model, the first prompt word can be tested after generating each dialogue text to determine whether the target intent label output by the intent recognition model based on the first prompt word can achieve the expected result.

[0081] Therefore, in one specific embodiment provided in this disclosure, after generating the first prompt word corresponding to each of the dialogue texts, the method further includes:

[0082] The first prompt words of multiple dialogue texts are input into the discrimination model to obtain test results of multiple first prompt words, wherein the test results are used to characterize the accuracy and generalization of the first prompt words;

[0083] Based on multiple test results, multiple effect test values ​​for the first prompt words are determined, wherein the effect test values ​​are used to measure the accuracy of the target intent label determined based on the multiple first prompt words;

[0084] If the effect test value reaches a preset threshold, multiple first prompt words are determined to meet the conditions for inputting the intent recognition model.

[0085] The system comprises: a discriminant model, used to determine the accuracy and generalization of the first prompt word; a test result, which is the processing result of the discriminant model on the first prompt word, used to characterize the accuracy and generalization of the first prompt word, including the accuracy test result and the generalization test result, specifically represented by 0 or 1; an effect test value, which is the result after weighting the test results of multiple first prompt words, used to measure the accuracy of the target intent label determined based on multiple first prompt words, specifically represented by a percentage or decimal form; and a preset threshold, which is the lower limit of the expected effect test value to be achieved.

[0086] Specifically, the generated first prompt word is input into the discriminant model. The model processes the first prompt word and outputs its test results, including accuracy (1 or 0) and generalization (1 or 0). An accuracy result of 1 indicates that the first prompt word's accuracy meets expectations (high accuracy), while an accuracy result of 0 indicates that the first prompt word's accuracy does not meet expectations (low accuracy). Similarly, a generalization result of 1 indicates that the first prompt word's generalization meets expectations (high generalization), while a generalization result of 0 indicates that the first prompt word's generalization does not meet expectations (low generalization). The test results with a value of 1 in both the accuracy and generalization tests are statistically analyzed. In practical applications, the accuracy and generalization test results are pre-set with corresponding weights. The statistically analyzed accuracy and generalization test results are then weighted according to these pre-set weights to obtain the overall performance test value. The effect test value is compared with a preset threshold. If the effect test value reaches the preset threshold, the first prompt word is determined to meet the conditions of the input intent recognition model, and can be input into the intent recognition model for subsequent processing. If the effect test value does not reach the preset threshold, the first prompt word is determined to not meet the conditions of the input intent recognition model, and cannot be input into the intent recognition model for subsequent processing. In this case, the first prompt word needs to be regenerated until the effect test value of the first prompt word reaches the preset threshold.

[0087] See Figure 3 , Figure 3 This diagram illustrates a method for determining whether a first prompt word satisfies the conditions of an input intent recognition model according to an embodiment of this disclosure. The illustration is given using an example of n first prompt words. Figure 3 As shown, the first prompt word 1, ..., the first prompt word n ​​are input into the discriminant model to obtain the test results corresponding to the first prompt word 1, ..., the first prompt word n, respectively, including the accuracy test results and the generalization test results. The test results with a value of 1 among the accuracy test results of the n first prompt words are counted, and the accuracy test value is determined to be 70%. Figure 3 Taking 70% as an example, the test results with a generalization value of 1 among the n first prompt words are counted, and the generalization test value is determined to be 60%. Figure 3 (Taking 60% as an example). Based on a weight of 0.7 for the accuracy test result and 0.3 for the generalization test result, a weighted average of the accuracy and generalization test values ​​is calculated, resulting in an effectiveness test value of 67%. Further, it is determined whether the effectiveness test value reaches a preset threshold. If the effectiveness test value reaches the preset threshold, it is determined that the first prompt word meets the conditions for the input intent recognition model, and the first prompt word can be input into the intent recognition model for subsequent processing. If the effectiveness test value does not reach the preset threshold, it is determined that the first prompt word does not meet the conditions for the input intent recognition model, and a new first prompt word needs to be generated.

[0088] This embodiment of the disclosure can test the accuracy and generalization of the first prompt words after generating multiple first prompt words. If the accuracy and generalization of the first prompt words reach a preset threshold, the first prompt words are then input into the intent recognition model for processing. This further improves the accuracy of the determined target intent label and avoids problems that occur during the processing of the first prompt words by the intent recognition model, thereby improving processing efficiency.

[0089] Step 106: Classify the multiple target intent labels according to the representation vectors of the multiple target intent labels to obtain multiple intent classification label groups.

[0090] After identifying the target intent tags of multiple dialogue texts, to further mine multi-level intent tags among different target intent tags, the identified target intent tags can be classified according to their representation vectors, resulting in multiple intent classification tag groups. These groups are then used to further determine the grouped intent tags for each intent classification tag group in subsequent processes. Here, an intent classification tag group refers to a group of intent tags of the same category obtained by classifying the target intent tags. The specific implementation method for classifying target intent tags to obtain multiple intent classification tag groups is as follows:

[0091] In one specific embodiment provided in this disclosure, multiple target intent tags are classified according to their representation vectors to obtain multiple intent classification tag groups, including:

[0092] Multiple representation vectors are clustered to obtain multiple clusters;

[0093] Select the cluster center of each cluster and determine the center intent label corresponding to each cluster center;

[0094] Based on the mapping relationship between the representation vectors and target intent labels in the multiple clusters, the multiple target intent labels are classified to obtain multiple intent classification label groups, wherein each intent classification label group includes the central intent label and the target intent label of the corresponding cluster.

[0095] The central intent label refers to the intent label corresponding to the cluster center of each cluster. The central intent label can be the target intent label corresponding to a certain representation vector, or it can be a newly generated intent label.

[0096] Specifically, multiple representation vectors are clustered. For example, a density-based spatial clustering of applications with noise (DBSCAN) algorithm can be used to cluster the representation vectors, resulting in multiple clusters. The mean of the data points (representation vectors) within each cluster is calculated, and the data point with the smallest mean is selected as the cluster center. In practical applications, either K-means clustering or K-mediods clustering can be used to select the cluster center. It should be noted that if K-means clustering is used to select the cluster center, the final selected cluster center may not be an actual data point in the cluster, but a newly generated data point. In this case, it is necessary to determine the central intent label corresponding to the cluster center. If the selected cluster center is an actual data point in the cluster, then the target intent label corresponding to that data point is directly determined as the central intent label of the cluster center.

[0097] Furthermore, after determining the central intent label corresponding to each cluster center, multiple target intent labels are classified based on the mapping relationship between the representation vector and the target intent label in each cluster, resulting in multiple intent classification label groups.

[0098] This embodiment first clusters the representation vectors corresponding to the target intent tags to obtain multiple clusters. Then, it selects the cluster centers of each cluster and determines the central intent tag corresponding to each cluster center. Finally, based on the mapping relationship between the representation vectors and the target intent tags, it classifies the multiple target intent tags to obtain multiple intent classification tag groups. Using clustering based on representation vectors to classify target intent tags can reduce information loss and improve the accuracy of target intent tag classification.

[0099] Step 108: Determine the group intent label corresponding to each intent classification label group.

[0100] Among them, grouped intent labels refer to the parent labels of the central intent label and multiple target intent labels in the intent classification label group, which are used to update the intent label system data of the intent recognition model; intent label system data refers to the system data with a hierarchical structure composed of several intent labels, which is used to perform intent recognition on user dialogue text; user dialogue text refers to the dialogue text input by the user in real time in actual applications.

[0101] Specifically, after obtaining multiple intent classification label groups, we can further mine the group intent labels corresponding to each intent classification label group, that is, mine the parent labels that have commonalities between the central intent label and the target intent label in the intent classification label group, so that the existing intent label system data can be updated based on the group intent labels.

[0102] In one specific embodiment provided in this disclosure, the group intent label corresponding to each intent classification label group is determined, including:

[0103] Get the second prompt word template;

[0104] For any one of the intent classification label groups, a candidate intent label is selected from the intent classification label group;

[0105] The candidate intent tags are filled into the second prompt word template to generate the second prompt word corresponding to the intent category tag group;

[0106] The second prompt word is input into the intent recognition model to determine the grouped intent label corresponding to the intent classification label group.

[0107] The second prompt word template refers to the template used to construct prompt words for determining the grouped intent labels corresponding to the intent classification label group; the candidate intent labels include the central intent label and the target intent labels selected in the intent classification label group that are similar to the central intent label; the second prompt word refers to the prompt word used to determine the grouped intent labels corresponding to the intent classification label group.

[0108] Specifically, a pre-configured second prompt word template is obtained. Candidate intent labels are selected from each intent category label group, and these candidate intent labels are filled into the corresponding positions in the second prompt word template to generate a second prompt word for each intent category label group. The second prompt word is then input into the intent recognition model, which outputs the grouped intent label for each intent category label group based on the second prompt word.

[0109] The following is an example of a second prompt word template:

[0110] ---------System Prompt---------

[0111] #roledefinition

[0112] You are an intent tag mining expert, skilled at extracting highly generalizable intent tags from a massive pool of candidate intent tags.

[0113] #task definition

[0114] Your task is to learn and analyze the provided candidate intent labels and summarize an intent label that can represent these intents.

[0115] #data definition

[0116] 1. <data>< / data> It stores K candidate intent labels, and these candidate intent labels have commonalities in the intent dimension; ...

[0118] #task requirements

[0119] 1. The extracted intent tags need to be concise and generalizable, and able to be interpreted. <data>< / data> All candidate intent tags;

[0120] 2. Currently, the identified intent tags are [a, b, c...]. If <data>< / data> If the candidate intent labels in the list can be interpreted using the intent labels in the list, then the existing labels are output.

[0121] 3. Output according to the following JSON format

[0122] ```

[0123] {{

[0124] "intent_label":<string,need be simple and general>

[0125] }}

[0126] ```

[0127] ---------User Prompt---------

[0128] <data>

[0129] Candidate Intent Label 1:

[0130] Candidate Intent Label 2: ...

[0132] Candidate intent label K:

[0133] < / data>

[0134] #generate mark

[0135] The extracted intent tags are output in JSON format:

[0136] In practical applications, candidate intent tags are populated into the second cue word template mentioned above. <data>< / data> The second prompt word can be generated from the input, and the intent label output by the intent recognition model is the... <data>< / data> The candidate intent tags correspond to the grouped intent tags, that is, the grouped intent tags corresponding to each intent category tag group. It should be noted that the above-mentioned second prompt word template is only an example given in the embodiments of this disclosure. In actual applications, different second prompt word templates can be pre-configured according to the actual application situation.

[0137] In this embodiment of the disclosure, a central intent label and a target intent label that is close to the central intent label can be selected as candidate intent labels in each intent classification label group. The second prompt word constructed based on the candidate intent labels is processed by the intent recognition model to determine the group intent label corresponding to each intent classification label group.

[0138] Furthermore, the specific implementation method for selecting candidate intent tags from the intent classification tag group is explained below. In a specific embodiment provided in this disclosure, selecting candidate intent tags from the intent classification tag group includes:

[0139] Determine the distance between each representation vector in the cluster corresponding to the intent classification label group and the cluster center of the cluster;

[0140] Based on the distance between each representation vector and the cluster center, a preset number of target intent labels are selected from the intent classification label group;

[0141] The preset number of target intent tags and the central intent tag of the cluster are determined as candidate intent tags.

[0142] The preset quantity refers to the number of target intent tags to be selected from the intent classification tag group. The preset quantity can be set according to the actual application situation, and this disclosure does not limit it.

[0143] Specifically, for each intent classification label group, the distance between each representation vector in the cluster corresponding to the intent classification label group and the cluster center is determined. Based on the distance between each representation vector and the cluster center, a preset number of target intent labels that are closest to the cluster center are selected. The selected target intent labels and the central intent label corresponding to the cluster center are determined as candidate intent labels.

[0144] For example, if it is necessary to select K candidate intent tags from the intent classification tag group, the preset number can be set to K-1. After determining the distance between each representation vector in the cluster corresponding to the intent classification tag group and the cluster center, select K-1 target intent tags that are closest to the cluster center, and then determine the K-1 target intent tags and the central intent tag of the cluster center as the candidate intent tags of the intent classification tag group.

[0145] This embodiment calculates the distance between each representation vector in a cluster and the cluster center, and selects the target intent label based on the distance, ensuring the similarity between the selected target intent label and the central intent label, thus making the grouped intent labels determined by the intent recognition model more accurate.

[0146] After mining the group intent tags corresponding to each intent category tag group, the existing intent tag system data can be updated based on the group intent tags, or the mined group intent tags and target intent tags can be integrated into a new intent tag system data based on the existing intent tag system data, ensuring the integrity and timeliness of the intent tag system data.

[0147] Based on this, in a specific embodiment provided in this disclosure, after determining the grouped intent label corresponding to each intent classification label group, the method further includes:

[0148] Obtain third-party prompt word templates and existing first-intention tag system data;

[0149] The third prompt word template is filled with multiple group intent tags and the first intent tag system data to generate the third prompt word;

[0150] The third prompt word is input into the intent recognition model to obtain the updated second intent label system data.

[0151] The third prompt word template refers to the prompt word used to build and update the intent tag system data; the first intent tag system data refers to the existing intent tag system data, which can be a tree structure or other hierarchical structure; the third prompt word refers to the prompt word used to update the intent tag system data; and the second intent tag system data refers to the intent tag system data updated from the first intent tag system data, or the new intent tag system data generated based on the integration of grouped intent tags and target intent tags.

[0152] Specifically, the pre-configured third prompt word template and existing first intent label system data are obtained. Multiple grouped intent labels and first intent label system data are filled into the corresponding positions in the third prompt word template to generate the third prompt word. The third prompt word is then input into the intent recognition model to obtain the second intent label system data output by the intent recognition model.

[0153] The following is Example 1 of the third prompt word template:

[0154] ---------System Prompt---------

[0155] #roledefinition

[0156] You are an intent tag mining expert, skilled at adjusting existing intent tag system data to make the intent tag system data more reasonable.

[0157] #task definition

[0158] Your task is to learn and analyze the provided intent tagging system data, and combine it with... <labels>< / labels> The grouped intent tags in the data summarize a more reasonable n-level intent tag system.

[0159] #data definition

[0160] 1. <data>< / data> It stores existing intent labeling system data;

[0161] 2. <labels>< / labels> It contains m group intent tags;

[0162] #task requirements

[0163] 1. Intent tags in the intent tagging system data need to be concise and generalizable;

[0164] 2. The adjusted intent tag system data ensures that intent tags at the same level do not overlap.

[0165] ---------User Prompt---------

[0166] <data>

[0167] {Existing intent tagging system data}

[0168] < / data>

[0169] <labels>

[0170] Grouping Intent Label 1:

[0171] Grouping Intent Label 2:

[0172] …

[0173] Grouping intent label m:

[0174] < / labels>

[0175] #generate mark

[0176] Adjusted intent tagging system data:

[0177] Example 1 above is a third prompt word template that adjusts and updates the existing intent tag system data based on the mined grouped intent tags. In practical applications, the existing intent tag system data is populated into the third prompt word template of Example 1 above. <data>< / data> In the middle, populate the group intent tags into the third prompt word template of Example 1 above. <labels>< / labels> This will generate a third prompt word for updating the intent tag system data.

[0178] In practical applications, it is also possible to configure a third prompt word template that integrates the mined grouped intent tags and target intent tags with existing intent tag system data to generate new intent tag system data, as shown in Example 2 below:

[0179] ---------System Prompt---------

[0180] #roledefinition

[0181] You are an intent tag mining expert, skilled at integrating existing intent tag system data to generate new intent tag system data, making the newly generated intent tag system data more reasonable.

[0182] #task definition

[0183] Your task is to learn and analyze the provided intent labeling system data, and <labels>< / labels> The group intent tags and target intent tags are integrated into a reasonable n-level intent tag system data.

[0184] #data definition

[0185] 1. <data>< / data> It stores existing intent labeling system data;

[0186] 2. <labels>< / labels> It contains m1 group intent labels and m2 target intent labels;

[0187] #task requirements

[0188] 1. Intent tags in the intent tagging system data need to be concise and generalizable;

[0189] 2. Integrate the generated intent tag system data, ensuring that intent tags at the same level do not overlap.

[0190] ---------User Prompt---------

[0191] <data>

[0192] {Existing intent tagging system data}

[0193] < / data>

[0194] <labels>

[0195] Grouping Intent Label 1:

[0196] Grouping Intent Label 2:

[0197] …

[0198] Grouping intent label m1:

[0199] Target Intent Label 1:

[0200] Target Intent Label 2:

[0201] …

[0202] Target intent label m2:

[0203] < / labels>

[0204] #generate mark

[0205] Integrate the generated intent tag system data:

[0206] Example 2 above illustrates how to generate a new intent tagging system data third prompt template by integrating mined grouped intent tags and target intent tags based on existing intent tagging system data. In practical applications, existing intent tagging system data is then populated into the third prompt template of Example 2. <data>< / data> In the middle, populate the group intent label and target intent label into the third prompt word template of Example 2 above. <labels>< / labels> This generates a third prompt word for updating the intent tag system data. It should be noted that the third prompt word templates in Example 1 and Example 2 are merely examples provided in this disclosure; in practical applications, different third prompt word templates can be pre-configured according to the actual application situation.

[0207] This embodiment of the disclosure can update the intent tag system data as needed according to different configured third prompt word templates, thereby improving the flexibility of generating intent tag system data.

[0208] The intent tag determination method provided in this disclosure includes: acquiring a dialogue text set and historical dialogue texts of multiple dialogue texts in the dialogue text set; determining a target intent tag and a representation vector of the target intent tag for each dialogue text based on the historical dialogue text of each dialogue text; classifying the multiple target intent tags according to the representation vectors of the multiple target intent tags to obtain multiple intent classification tag groups; and determining a grouped intent tag corresponding to each intent classification tag group, wherein the grouped intent tags are used to update the intent tag system data of the intent recognition model, and the intent tag system data is used to perform intent recognition on user dialogue texts.

[0209] This disclosure implements a method to determine the target intent tag and its representation vector for each dialogue text by analyzing its historical dialogue text. Based on the representation vector, the target intent tag is classified into multiple intent category tag groups. Furthermore, the grouped intent tags corresponding to each intent category tag group are determined. This method not only identifies the target intent tag for each dialogue text but also mines the grouped intent tags corresponding to the intent category tag group to which the target intent tag belongs. Determining the target intent tag by combining the historical dialogue text with the dialogue text allows for contextual completion of the target intent tag, improving accuracy. It eliminates the need for manual annotation of each dialogue text, reducing manual annotation costs. It enables direct intent recognition from unannotated dialogue text sets, allowing for zero-based mining of intent tags and further determination of grouped intent tags for intent category tag groups. This can be used to construct a multi-level intent tag system data, enabling granular mining of different intent tags.

[0210] The following combination Figure 4 The intent tag determination method provided in the embodiments of this disclosure will be further explained and described. Figure 4 A flowchart illustrating an intent label determination method according to an embodiment of this disclosure is shown, such as... Figure 4 As shown, the intent tag determination method provided in this embodiment can be divided into three parts: generating representation vectors, classifying target intent tags, and mining grouped intent tags. First, a dialogue text set and historical dialogue texts of multiple dialogue texts within the dialogue text set are acquired. A first prompt word is constructed based on the dialogue text and its historical dialogue text. The first prompt word is processed by an intent recognition model to obtain the target intent tag and its representation vector for each dialogue text. Then, multiple representation vectors are clustered to obtain multiple clusters. Based on the mapping relationship between each representation vector in the cluster and the target intent tag, multiple target intent tags are classified to obtain multiple intent classification tag groups. Further, a preset number of target intent tags are selected from each intent classification tag group. The selected target intent tags and the central intent tag corresponding to the cluster center of each cluster are determined as candidate intent tags for each intent classification tag group. A second prompt word is constructed based on the candidate intent tags. The second prompt word is processed by an intent recognition model to determine the grouped intent tags corresponding to each intent classification tag group. Finally, a third prompt word is constructed based on the grouped intent tags and the existing first intent tag system data. This third prompt word is then processed by the intent recognition model, causing the model to update the existing first intent tag system data, resulting in an updated second intent tag system. This updated second intent tag system data can be used to identify the intent of user-inputted dialogue text in practical applications.

[0211] This disclosure achieves the following: by combining historical dialogue texts of the dialogue text, the target intent tag of the dialogue text is determined, thereby improving the accuracy of determining the target intent tag; thus eliminating the need for manual annotation of each dialogue text, reducing the cost of manual annotation; further, the grouped intent tags of intent classification tag groups are determined, which can be used to construct or update multi-level intent tag system data, and realize the granular mining between different intent tags.

[0212] It is understood that the various method embodiments mentioned above in this disclosure can be combined with each other to form combined embodiments without violating the principle and logic. Due to space limitations, this disclosure will not elaborate further. Those skilled in the art will understand that in the above methods of specific implementation, the specific execution order of each step should be determined by its function and possible internal logic.

[0213] In addition, this disclosure also provides an intent tag determination apparatus, an electronic device, a computer-readable storage medium, and a computer program product, all of which can be used to implement any of the intent tag determination methods provided in this disclosure. The corresponding technical solutions and descriptions are described in the corresponding section of the method and will not be repeated here.

[0214] Figure 5 A block diagram of an intent label determination apparatus according to an embodiment of the present disclosure is shown. See also Figure 5 This disclosure provides an intent label determination device, which includes:

[0215] The acquisition module 502 is configured to acquire a dialogue text set and historical dialogue texts of multiple dialogue texts in the dialogue text set;

[0216] The first determining module 504 is configured to determine the target intent tag and the representation vector of the target intent tag for each dialogue text based on the historical dialogue text of each dialogue text.

[0217] The classification module 506 is configured to classify the multiple target intent labels according to the representation vectors of the multiple target intent labels to obtain multiple intent classification label groups;

[0218] The second determining module 508 is configured to determine the grouped intent label corresponding to each of the intent classification label groups, wherein the grouped intent label is used to update the intent label system data of the intent recognition model, and the intent label system data is used to perform intent recognition on user dialogue text.

[0219] Optionally, the first determining module 504 is further configured to:

[0220] Get the first prompt word template;

[0221] Each dialogue text and its historical dialogue text are filled into the first prompt word template to generate a first prompt word corresponding to each dialogue text.

[0222] Each of the first prompt words is input into the intent recognition model to determine the target intent label and the representation vector of the target intent label for each dialogue text.

[0223] Optionally, the device further includes a discrimination module configured to:

[0224] The first prompt words of multiple dialogue texts are input into the discrimination model to obtain test results of multiple first prompt words, wherein the test results are used to characterize the accuracy and generalization of the first prompt words;

[0225] Based on multiple test results, multiple effect test values ​​for the first prompt words are determined, wherein the effect test values ​​are used to measure the accuracy of the target intent label determined based on the multiple first prompt words;

[0226] If the effect test value reaches a preset threshold, multiple first prompt words are determined to meet the conditions for inputting the intent recognition model.

[0227] Optionally, the classification module 506 is further configured to:

[0228] Multiple representation vectors are clustered to obtain multiple clusters;

[0229] Select the cluster center of each cluster and determine the center intent label corresponding to each cluster center;

[0230] Based on the mapping relationship between the representation vectors and target intent labels in the multiple clusters, the multiple target intent labels are classified to obtain multiple intent classification label groups, wherein each intent classification label group includes the central intent label and the target intent label of the corresponding cluster.

[0231] Optionally, the second determining module 508 is further configured to:

[0232] Get the second prompt word template;

[0233] For any one of the intent classification label groups, a candidate intent label is selected from the intent classification label group;

[0234] The candidate intent tags are filled into the second prompt word template to generate the second prompt word corresponding to the intent category tag group;

[0235] The second prompt word is input into the intent recognition model to determine the grouped intent label corresponding to the intent classification label group.

[0236] Optionally, the second determining module 508 is further configured to:

[0237] Determine the distance between each representation vector in the cluster corresponding to the intent classification label group and the cluster center of the cluster;

[0238] Based on the distance between each representation vector and the cluster center, a preset number of target intent labels are selected from the intent classification label group;

[0239] The preset number of target intent tags and the central intent tag of the cluster are determined as candidate intent tags.

[0240] Optionally, the device further includes an update module configured to:

[0241] Obtain third-party prompt word templates and existing first-intention tag system data;

[0242] The third prompt word template is filled with multiple group intent tags and the first intent tag system data to generate the third prompt word;

[0243] The third prompt word is input into the intent recognition model to obtain the updated second intent label system data.

[0244] The intent tag determination apparatus provided in this disclosure includes: an acquisition module configured to acquire a dialogue text set and historical dialogue texts of multiple dialogue texts in the dialogue text set; a first determination module configured to determine a target intent tag and a representation vector of the target intent tag for each dialogue text based on the historical dialogue text of each dialogue text; a classification module configured to classify the multiple target intent tags based on the representation vectors of the multiple target intent tags to obtain multiple intent classification tag groups; and a second determination module configured to determine a grouped intent tag corresponding to each intent classification tag group, wherein the grouped intent tags are used to update the intent tag system data of the intent recognition model, and the intent tag system data is used to perform intent recognition on user dialogue texts.

[0245] This disclosure implements a method to determine the target intent tag and its representation vector for each dialogue text by analyzing its historical dialogue text. Based on the representation vector, the target intent tag is classified into multiple intent category tag groups. Furthermore, the grouped intent tags corresponding to each intent category tag group are determined. This method not only identifies the target intent tag for each dialogue text but also mines the grouped intent tags corresponding to the intent category tag group to which the target intent tag belongs. Determining the target intent tag by combining the historical dialogue text with the dialogue text allows for contextual completion of the target intent tag, improving accuracy. It eliminates the need for manual annotation of each dialogue text, reducing manual annotation costs. It enables direct intent recognition from unannotated dialogue text sets, allowing for zero-based mining of intent tags and further determination of grouped intent tags for intent category tag groups. This can be used to construct a multi-level intent tag system data, enabling granular mining of different intent tags.

[0246] The modules in the aforementioned intent label determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can invoke and execute the operations corresponding to each module.

[0247] See Figure 6 , Figure 6 A flowchart of a text processing method according to an embodiment of this disclosure is shown, which specifically includes the following steps:

[0248] Step 602: Obtain the user dialogue text and the historical user dialogue text of the user dialogue text.

[0249] Step 604: Input the user dialogue text, the historical user dialogue text, and the current intent tag system data into the intent recognition model to obtain the user dialogue intent, wherein the intent tag system data is determined according to the intent tag determination method described above.

[0250] Step 606: Generate a response text to the user's dialogue text based on the user's dialogue intent.

[0251] Among them, user dialogue text refers to the dialogue text entered by the user that requires a response; historical user dialogue text refers to the dialogue text of all rounds before the current user dialogue text in the current dialogue process; user dialogue intent refers to the actual intent of the user dialogue text, that is, the purpose of the user entering the user dialogue text; and response text refers to the dialogue text generated to reply to the user based on the identified user dialogue intent.

[0252] Specifically, the system acquires the user's real-time input dialogue text, as well as the historical user dialogue text during the current dialogue. The acquired user dialogue text, historical user dialogue text, and current intent tagging system data are input into the intent recognition model to obtain the user dialogue intent of the user dialogue text output by the intent recognition model. Based on the user dialogue intent, a reply text is generated to respond to the user, and the reply text is then fed back to the user.

[0253] The embodiments disclosed herein can identify the user's dialogue intent in the user's input dialogue text through the current intent tagging system data, thereby improving the accuracy of the identified user dialogue intent, and thus making the response text generated based on the user's dialogue intent more accurate and improving the user experience.

[0254] It should be noted that the intent tag system data generated based on the above intent tag determination method can be applied not only to the scenarios corresponding to the above text processing methods, but also to other scenarios, such as model training scenarios. Specifically, the intent tags in the intent tag system data can be used as sample intent tags for sample dialogue text for model training.

[0255] Figure 7 This is a block diagram of an electronic device provided in an embodiment of the present disclosure.

[0256] See Figure 7 This disclosure provides an electronic device 700, which includes: at least one processor 701; at least one memory 702; and one or more I / O interfaces 703 connected between the processor 701 and the memory 702; wherein the memory 702 stores one or more computer programs that can be executed by the at least one processor 701, and the one or more computer programs are executed by the at least one processor 701 to enable the at least one processor 701 to perform the above-described intent tag determination method or text processing method.

[0257] The modules in the aforementioned electronic devices can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0258] This disclosure also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor / processing core, implements the aforementioned intent tag determination method or text processing method. The computer-readable storage medium may be volatile or non-volatile.

[0259] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described intent tag determination method or text processing method.

[0260] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0261] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable program instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0262] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0263] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as "C" or similar languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions to implement various aspects of this disclosure.

[0264] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0265] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0266] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0267] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0268] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0269] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in connection with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in connection with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.< / context>

Claims

1. A method for determining intent labels, characterized in that, include: Obtain the dialogue text set and the historical dialogue text of multiple dialogue texts in the dialogue text set; Based on the historical dialogue text of each dialogue text, determine the target intent label and the representation vector of the target intent label for each dialogue text; Based on the representation vectors of the multiple target intent labels, the multiple target intent labels are classified to obtain multiple intent classification label groups; Each intent classification label group is determined separately, wherein the group intent label is used to update the intent label system data of the intent recognition model, and the intent label system data is used to perform intent recognition on user dialogue text.

2. The method as described in claim 1, characterized in that, Based on the historical dialogue text of each dialogue text, determine the target intent label and the representation vector of the target intent label for each dialogue text, including: Get the first prompt word template; Each dialogue text and its historical dialogue text are filled into the first prompt word template to generate a first prompt word corresponding to each dialogue text. Each of the first prompt words is input into the intent recognition model to determine the target intent label and the representation vector of the target intent label for each dialogue text.

3. The method as described in claim 2, characterized in that, After generating the first prompt word corresponding to each of the dialogue texts, the method further includes: The first prompt words of multiple dialogue texts are input into the discrimination model to obtain test results of multiple first prompt words, wherein the test results are used to characterize the accuracy and generalization of the first prompt words; Based on multiple test results, multiple effect test values ​​for the first prompt words are determined, wherein the effect test values ​​are used to measure the accuracy of the target intent label determined based on the multiple first prompt words; If the effect test value reaches a preset threshold, multiple first prompt words are determined to meet the conditions for inputting the intent recognition model.

4. The method as described in claim 1, characterized in that, Based on the representation vectors of the multiple target intent labels, the multiple target intent labels are classified to obtain multiple intent classification label groups, including: Multiple representation vectors are clustered to obtain multiple clusters; Select the cluster center of each cluster and determine the center intent label corresponding to each cluster center; Based on the mapping relationship between the representation vectors and target intent labels in the multiple clusters, the multiple target intent labels are classified to obtain multiple intent classification label groups, wherein each intent classification label group includes the central intent label and the target intent label of the corresponding cluster.

5. The method as described in claim 4, characterized in that, Determine the group intent label corresponding to each of the intent classification label groups, including: Get the second prompt word template; For any one of the intent classification label groups, a candidate intent label is selected from the intent classification label group; The candidate intent tags are filled into the second prompt word template to generate the second prompt word corresponding to the intent category tag group; The second prompt word is input into the intent recognition model to determine the group intent label corresponding to the intent classification label group.

6. The method as described in claim 5, characterized in that, Candidate intent labels are selected from the intent classification label group, including: Determine the distance between each representation vector in the cluster corresponding to the intent classification label group and the cluster center of the cluster; Based on the distance between each representation vector and the cluster center, a preset number of target intent labels are selected from the intent classification label group; The preset number of target intent tags and the central intent tag of the cluster are determined as candidate intent tags.

7. The method as described in claim 1, characterized in that, After determining the grouped intent label corresponding to each intent classification label group, the method further includes: Obtain third-party prompt word templates and existing first-intention tag system data; The third prompt word template is filled with multiple group intent tags and the first intent tag system data to generate the third prompt word; The third prompt word is input into the intent recognition model to obtain the updated second intent label system data.

8. A text processing method, characterized in that, include: Obtain the user's dialogue text and the historical user dialogue text of the user's dialogue text; The user dialogue text, the historical user dialogue text, and the current intent tag system data are input into the intent recognition model to obtain the user dialogue intent, wherein the intent tag system data is determined according to the method described in any one of claims 1-7. Based on the user's dialogue intent, generate a response text to the user's dialogue text.

9. An intent tag determination device, characterized in that, include: The acquisition module is configured to acquire a dialogue text set and historical dialogue texts of multiple dialogue texts in the dialogue text set; The first determining module is configured to determine the target intent tag and the representation vector of the target intent tag for each dialogue text based on the historical dialogue text of each dialogue text. The classification module is configured to classify the multiple target intent labels based on the representation vectors of the multiple target intent labels to obtain multiple intent classification label groups; The second determining module is configured to determine the grouped intent label corresponding to each of the intent classification label groups, wherein the grouped intent label is used to update the intent label system data of the intent recognition model, and the intent label system data is used to perform intent recognition on the user dialogue text.

10. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores one or more computer programs that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1-8.

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

Images