Intention recognition matching method, device and equipment

Abstract

The application provides an intention recognition matching method, device and equipment. The method comprises the following steps: obtaining an intention created by an intention creator; obtaining criterion strategy related information matched with the intention; analyzing candidate intention knowledge according to the criterion strategy related information, and determining target intention knowledge; and sending the target intention knowledge to an intention execution module. In the embodiment of the application, the candidate intention knowledge is analyzed based on the criterion strategy related information, the target intention knowledge matched with the intention created by the intention creator is determined from the candidate intention knowledge, the criterion strategy related information includes a plurality of criterion factors affecting the target intention knowledge, the comprehensive judgment of the intention knowledge can be performed based on the plurality of factors, the system is more robust, and the intention recognition effect is improved.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to an intent recognition and matching method, apparatus and device. Background Technology

[0002] Intent is a service description method that abstractly defines network requirements and provides a set of expectations about the network or service without specifying concrete technical details. Intent-driven networking requires less specialized knowledge from users, has better universality, and can meet the network service needs of a wide range of people.

[0003] Existing methods for selecting intent knowledge mainly fall into two categories: one is rule-based intent recognition, which typically categorizes intent knowledge based on keywords. When a user intent is obtained, the intent text is matched with multiple sets of keywords to determine the corresponding intent knowledge. However, this method cannot adapt to the diversity and ambiguity of language logic, resulting in poor intent recognition performance.

[0004] Another type of approach is based on machine learning algorithms, which calculate the degree of matching between user intent and corresponding metrics, such as semantic matching degree and text edit distance, and outputs the intent knowledge with the highest matching degree based on the final score. This type of algorithm has high requirements for model accuracy and relies too much on the quality of the model itself. It cannot adaptively adjust according to historical records and specific user preferences during actual use, resulting in poor intent recognition performance. Summary of the Invention

[0005] The purpose of this invention is to provide an intent recognition and matching method, apparatus, and device that solves the problem of poor recognition performance in existing intent recognition methods.

[0006] To achieve the above objectives, embodiments of the present invention provide an intent recognition and matching method, comprising:

[0007] Retrieve the intent created by the intent creator;

[0008] Obtain relevant information on criteria and policies that match the stated intent;

[0009] Based on the relevant information of the aforementioned criteria and strategies, the candidate intent knowledge is analyzed to determine the target intent knowledge;

[0010] The target intent knowledge is sent to the intent execution module.

[0011] Optionally, obtaining criteria policy-related information matching the intent includes:

[0012] Based on the current system state and the characteristic information of the intent, a criterion strategy template is matched for the intent. The criterion strategy template includes criterion factors that influence the knowledge of the target intent.

[0013] Obtain the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge.

[0014] Optionally, obtaining the criterion factor weight vector of the criterion factors includes:

[0015] Based on the first influence weight of each criterion factor on the target intent knowledge, a criterion judgment matrix is ​​obtained; the column vector is obtained by summing the row vectors of the criterion judgment matrix, and the column vector is standardized to obtain the criterion factor weight vector.

[0016] or

[0017] The weight vector of the criteria factors is obtained from the selection report sent by the terminal.

[0018] Optionally, obtaining the intent knowledge weight vector of the candidate intent knowledge includes:

[0019] Based on the second influence weight of each candidate intent knowledge on each criterion factor, obtain the intent knowledge judgment matrix;

[0020] The row vectors of the intent knowledge judgment matrix are summed to obtain column vectors, and the column vectors are standardized to obtain the intent knowledge weight vector.

[0021] Optionally, the candidate intent knowledge is analyzed based on the criterion policy-related information to determine the target intent knowledge, including:

[0022] Based on the criterion factor weight vector and the intent knowledge weight vector, calculate the matching result of each candidate intent knowledge for the intent;

[0023] The target intent knowledge is determined based on the matching results.

[0024] Optionally, determining the target intent knowledge based on the matching result includes:

[0025] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is greater than or equal to the first threshold, then the intent knowledge corresponding to the highest matching result is determined to be the target intent knowledge.

[0026] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is at least less than a first threshold, then the intent knowledge selected by the intent creator is determined to be the target intent knowledge.

[0027] Optionally, after calculating the matching result of each candidate intent knowledge for the intent based on the criterion factor weight vector and the intent knowledge weight vector, the method further includes;

[0028] If the highest matching result in the matching results is 0, the intent is retrieved again.

[0029] Optionally, if the absolute value of the comparison between the highest matching result and a first matching result other than the highest matching result is less than at least a first threshold, the method further includes:

[0030] A selection reference report is sent to the terminal. The selection reference report includes: first intent knowledge corresponding to the first matching result whose absolute value is less than a first threshold, and second intent knowledge corresponding to the highest matching result, so that the intent creator can select the target intent knowledge from the first intent knowledge and the second intent knowledge.

[0031] The receiving terminal sends a selection report, which includes: the target intent knowledge selected by the intent creator and an updated criterion factor weight vector.

[0032] Optionally, the selection reference report may also include at least one of the following:

[0033] The matching results corresponding to the first intent knowledge and the second intent knowledge, respectively;

[0034] The criterion factor weight vector.

[0035] Optionally, after receiving the selection report sent by the receiving terminal, the method further includes:

[0036] Based on the selection report, the criterion strategy is self-learned and the criterion strategy template is updated.

[0037] Optionally, the criterion factors include at least one of the following:

[0038] Network resource metrics;

[0039] Business quality metrics;

[0040] Intent-knowledge matching degree;

[0041] Historical execution of intent knowledge;

[0042] Intent creator preference characteristics.

[0043] Optionally, if the criterion factors include the intent creator preference features, after matching the intent with the criterion strategy template, the method further includes:

[0044] The matched criterion policy template is updated, and the updated criterion policy template does not include the intent creator preference feature.

[0045] To achieve the above objectives, embodiments of the present invention provide an intent recognition and matching device, comprising:

[0046] The intent retrieval module is used to retrieve intents created by the intent creator.

[0047] The criteria data acquisition module is used to acquire criteria policy-related information that matches the intent;

[0048] The intent knowledge analysis module is used to analyze candidate intent knowledge based on the criteria and strategy information to determine the target intent knowledge;

[0049] The sending module is used to send the target intent knowledge to the intent execution module.

[0050] Optionally, the criterion data acquisition module includes:

[0051] The matching unit is used to match a criterion strategy template for the intent based on the current system state and the feature information of the intent. The criterion strategy template includes criterion factors that affect the knowledge of the target intent.

[0052] The first acquisition unit is used to acquire the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge.

[0053] Optionally, the first acquisition unit is specifically used for:

[0054] Based on the first influence weight of each criterion factor on the target intent knowledge, a criterion judgment matrix is ​​obtained; the column vector is obtained by summing the row vectors of the criterion judgment matrix, and the column vector is standardized to obtain the criterion factor weight vector.

[0055] or

[0056] The weight vector of the criteria factors is obtained from the selection report sent by the terminal.

[0057] Optionally, the first acquisition unit is specifically used to: acquire an intent knowledge judgment matrix based on the second influence weight of each candidate intent knowledge on each criterion factor;

[0058] The row vectors of the intent knowledge judgment matrix are summed to obtain column vectors, and the column vectors are standardized to obtain the intent knowledge weight vector.

[0059] Optionally, the intent knowledge analysis module includes:

[0060] The first calculation unit is used to calculate the matching result of each candidate intent knowledge for the intent based on the criterion factor weight vector and the intent knowledge weight vector.

[0061] The first determining unit is used to determine the target intent knowledge based on the matching result.

[0062] Optionally, the first determining unit is specifically used for:

[0063] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is greater than or equal to the first threshold, then the intent knowledge corresponding to the highest matching result is determined to be the target intent knowledge.

[0064] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is at least less than a first threshold, then the intent knowledge selected by the intent creator is determined to be the target intent knowledge.

[0065] Optionally, the device further includes;

[0066] The first acquisition module is used to reacquire the intent if the highest matching result in the matching results is 0.

[0067] Optionally, the device further includes:

[0068] A preference selection module is used to send a selection reference report to the terminal. The selection reference report includes: first intent knowledge corresponding to the first matching result whose absolute value is less than a first threshold, and second intent knowledge corresponding to the highest matching result, so that the intent creator can select the target intent knowledge from the first intent knowledge and the second intent knowledge.

[0069] The receiving module is used to receive a selection report sent by the terminal. The selection report includes: the target intent knowledge selected by the intent creator and an updated criterion factor weight vector.

[0070] Optionally, the selection reference report may also include at least one of the following:

[0071] The matching results corresponding to the first intent knowledge and the second intent knowledge, respectively;

[0072] The criterion factor weight vector.

[0073] Optionally, the device further includes:

[0074] The dynamic criterion strategy self-learning module is used to learn the criterion strategy and update the criterion strategy template based on the selection report.

[0075] Optionally, the criterion factors include at least one of the following:

[0076] Network resource metrics;

[0077] Business quality metrics;

[0078] Intent-knowledge matching degree;

[0079] Historical execution of intent knowledge;

[0080] Intent creator preference characteristics.

[0081] Optionally, if the criterion factors include the intent creator preference characteristics, the apparatus further includes:

[0082] The update module is used to update the matched criterion policy template, wherein the updated criterion policy template does not include the intent creator preference feature.

[0083] To achieve the above objectives, embodiments of the present invention provide an intent recognition and matching device, comprising: a transceiver and a processor;

[0084] The transceiver is used to: acquire intents created by intent creators; acquire criterion policy information that matches the intents;

[0085] The processor is configured to: analyze candidate intent knowledge based on the criteria policy information to determine target intent knowledge; and send the target intent knowledge to the intent execution module.

[0086] Optionally, the transceiver acquires criterion policy-related information matching the intent, specifically including:

[0087] Based on the current system state and the characteristic information of the intent, a criterion strategy template is matched for the intent. The criterion strategy template includes criterion factors that influence the knowledge of the target intent.

[0088] Obtain the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge.

[0089] Optionally, the transceiver acquires the criterion factor weight vector of the criterion factors, specifically including:

[0090] Based on the first influence weight of each criterion factor on the target intent knowledge, a criterion judgment matrix is ​​obtained; the column vector is obtained by summing the row vectors of the criterion judgment matrix, and the column vector is standardized to obtain the criterion factor weight vector.

[0091] or

[0092] The weight vector of the criteria factors is obtained from the selection report sent by the terminal.

[0093] Optionally, the transceiver acquires the intent knowledge weight vector of the candidate intent knowledge, specifically including:

[0094] Based on the second influence weight of each candidate intent knowledge on each criterion factor, obtain the intent knowledge judgment matrix;

[0095] The row vectors of the intent knowledge judgment matrix are summed to obtain column vectors, and the column vectors are standardized to obtain the intent knowledge weight vector.

[0096] Optionally, the processor analyzes the candidate intent knowledge based on the criterion policy-related information to determine the target intent knowledge, including:

[0097] Based on the criterion factor weight vector and the intent knowledge weight vector, calculate the matching result of each candidate intent knowledge for the intent;

[0098] The target intent knowledge is determined based on the matching results.

[0099] Optionally, the processor determines the target intent knowledge based on the matching result, including:

[0100] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is greater than or equal to the first threshold, then the intent knowledge corresponding to the highest matching result is determined to be the target intent knowledge.

[0101] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is at least less than a first threshold, then the intent knowledge selected by the intent creator is determined to be the target intent knowledge.

[0102] Optionally, after calculating the matching result of each candidate intent knowledge for the intent based on the criterion factor weight vector and the intent knowledge weight vector, the transceiver is further configured to;

[0103] If the highest matching result in the matching results is 0, the intent is retrieved again.

[0104] Optionally, if the absolute value of the comparison between the highest matching result and a first matching result other than the highest matching result is at least less than a first threshold, the transceiver is further configured to:

[0105] A selection reference report is sent to the terminal. The selection reference report includes: first intent knowledge corresponding to the first matching result whose absolute value is less than a first threshold, and second intent knowledge corresponding to the highest matching result, so that the intent creator can select the target intent knowledge from the first intent knowledge and the second intent knowledge.

[0106] The receiving terminal sends a selection report, which includes: the target intent knowledge selected by the intent creator and an updated criterion factor weight vector.

[0107] Optionally, the selection reference report may also include at least one of the following:

[0108] The matching results corresponding to the first intent knowledge and the second intent knowledge, respectively;

[0109] The criterion factor weight vector.

[0110] Optionally, after receiving the user selection report sent by the receiving terminal, the processor is further configured to:

[0111] Based on the selection report, the criterion strategy is self-learned and the criterion strategy template is updated.

[0112] Optionally, the criterion factors include at least one of the following:

[0113] Network resource metrics;

[0114] Business quality metrics;

[0115] Intent-knowledge matching degree;

[0116] Historical execution of intent knowledge;

[0117] Intent creator preference characteristics.

[0118] Optionally, if the criterion factors include the intent creator preference features, after matching the intent with a criterion strategy template, the processor is further configured to:

[0119] The matched criterion policy template is updated, and the updated criterion policy template does not include the intent creator preference feature.

[0120] To achieve the above objectives, embodiments of the present invention provide an electronic device, including: a transceiver, a processor, a memory, and a program or instructions stored in the memory and executable on the processor; when the processor executes the program or instructions, it implements the above-described intent recognition and matching method.

[0121] To achieve the above objectives, embodiments of the present invention provide a readable storage medium having a program or instructions stored thereon, which, when executed by a processor, implement the steps of the intent recognition and matching method described above.

[0122] The beneficial effects of the above-described technical solution of the present invention are as follows:

[0123] In embodiments of the present invention, candidate intent knowledge is analyzed based on criterion-strategy related information. Target intent knowledge matching the intent created by the intent creator is determined from the candidate intent knowledge. The criterion-strategy related information includes multiple criterion factors that influence the target intent knowledge. Comprehensive evaluation of intent knowledge can be performed based on multiple factors, making the system more robust and improving intent recognition performance. Attached Figure Description

[0124] Figure 1 This is one of the flowcharts illustrating the intent recognition and matching method according to an embodiment of the present invention;

[0125] Figure 2 This is one of the schematic diagrams of the dynamic hierarchical selection model of intent knowledge in an embodiment of the present invention;

[0126] Figure 3 This is a schematic diagram of the structure of the intent recognition and matching device according to an embodiment of the present invention;

[0127] Figure 4 This is a second schematic flowchart of the intent recognition and matching method according to an embodiment of the present invention;

[0128] Figure 5 This is the second schematic diagram of the dynamic hierarchical selection model of intent knowledge in an embodiment of the present invention;

[0129] Figure 6 This is a schematic diagram of the structure of an intent recognition and matching device according to an embodiment of the present invention;

[0130] Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0131] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0132] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0133] In various embodiments of the present invention, it should be understood that the sequence number of each process described below does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0134] In addition, the terms "system" and "network" are often used interchangeably in this article.

[0135] In the embodiments provided in this application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0136] like Figure 1 As shown, an embodiment of the present invention provides an intent recognition and matching method, including:

[0137] Step 11: Obtain the intent created by the intent creator.

[0138] The intent creator can be a user. In this embodiment, the intent created by the intent creator can be obtained through a user interface (UI), a voice device, or other system interfaces. For example, the intent creator can enter the content they want to search for at the location where a video search is performed.

[0139] Step 12: Obtain relevant information on the criteria strategy that matches the stated intent.

[0140] The criteria policy information may include: a criteria policy template, the content of the criteria policy template, and other parameters related to the criteria policy. The criteria policy can be influencing factors for intent recognition, such as: network resource indicators, service quality indicators, intent knowledge matching degree, historical execution status, intent creator preference characteristics, etc.

[0141] Step 13: Analyze the candidate intent knowledge based on the relevant information of the criterion strategy to determine the target intent knowledge;

[0142] Step 14: Send the target intent knowledge to the intent execution module.

[0143] In this embodiment, the candidate intent knowledge can be stored in an intent knowledge base. The intent knowledge base stores intent standard description files for each intent knowledge, where the description information includes, but is not limited to, intent-response execution flow and intent expression description. The candidate intent knowledge stored in the intent knowledge base is analyzed based on the criterion policy information to determine the target intent knowledge matching the user intent. This target intent knowledge is then sent to the intent execution module, which decides whether to execute the strategy or workflow mapped to the target intent knowledge. For example, if the intent execution module detects an intent conflict, it can choose not to execute the strategy or workflow mapped to the target intent knowledge.

[0144] In embodiments of the present invention, candidate intent knowledge is analyzed based on criterion-strategy related information. Target intent knowledge matching the intent created by the intent creator is determined from the candidate intent knowledge. The criterion-strategy related information includes multiple criterion factors that influence the target intent knowledge. Comprehensive evaluation of intent knowledge can be performed based on multiple factors, making the system more robust and improving intent recognition performance.

[0145] Optionally, the criterion-strategy related information for intent matching may include: a criterion-strategy template, a criterion factor weight vector, and an intent knowledge weight vector. Specifically, obtaining the criterion-strategy related information matching the intent may include:

[0146] Based on the current system state and the feature information of the intent, a matching criterion strategy template is generated for the intent. The criterion strategy template includes criterion factors that influence the target intent knowledge. The criterion factor weight vector and the intent knowledge weight vector of the candidate intent knowledge are obtained.

[0147] In this embodiment, the criterion strategy template can be stored in a dynamic criterion strategy library. The dynamic criterion strategy library stores multiple criterion strategy templates, and each criterion strategy template can include multiple criterion factors that influence the target intent knowledge. The influence weight of different criterion factors on the target intent knowledge may differ. The candidate intent knowledge can be stored in an intent knowledge base, and the influence weight of different candidate intent knowledge on different criterion factors may differ.

[0148] Optionally, when analyzing candidate intent knowledge based on the criterion strategy-related information, the analysis can be performed at different levels, for example, by establishing... Figure 2The dynamic hierarchical selection model of intent knowledge shown is as follows: a criterion layer is constructed using the criterion factors, and a scheme layer is constructed using candidate intent knowledge. Each criterion factor is associated with each candidate intent knowledge. The target intent knowledge is determined from the candidate intent knowledge through the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge. The target intent knowledge is the optimal intent knowledge selection.

[0149] Optionally, the criterion factors included in the criterion strategy template may include at least one of the following:

[0150] (1) Network resource indicators; for example: CPU utilization, virtual machine utilization, resource usage (referring to the comprehensive resource usage factors that combine multiple resource usage situations), etc., which are system resources that can be occupied after the intent knowledge is instantiated.

[0151] (2) Business quality indicators; for example, service quality factors provided after the instantiation of intent knowledge such as Quality of Service (QoS).

[0152] (3) Intent-knowledge matching degree;

[0153] (4) Historical execution status of intent knowledge; for example: number of historical executions, historical execution success rate, etc.

[0154] (5) Intent Creator Preference Characteristics. It should be noted that if an intent creator preference selection mechanism has been activated before the intent knowledge identification and matching process, the "intent creator preference characteristics" factor may be included in the criteria factors. The intent creator preference selection mechanism refers to: sending a selection reference report to the intent creator, enabling the intent creator to select the target intent knowledge from the candidate intent knowledge provided in the selection reference report.

[0155] In embodiments of the present invention, when analyzing candidate intent knowledge based on the criterion policy information to determine target intent knowledge, two execution scenarios may be included. The first scenario is when the intent creator preference selection mechanism has not been activated, in which case the criterion policy templates stored in the dynamic criterion policy library do not include the "intent creator preference feature" factor. The second scenario is when the intent creator preference selection mechanism has been activated, in which case the criterion policy templates stored in the dynamic criterion policy library include the "intent creator preference feature" factor.

[0156] As an optional embodiment, obtaining the criterion factor weight vector of the criterion factors includes:

[0157] Method 1: Obtain the criterion judgment matrix based on the first influence weight of each criterion factor on the target intent knowledge; sum the row vectors of the criterion judgment matrix to obtain the column vectors; standardize the column vectors to obtain the criterion factor weight vector.

[0158] or

[0159] Method 2: Obtain the weight vector of the criteria factors from the selection report sent by the terminal.

[0160] In this embodiment, there are two ways to obtain the criterion factor weight vector. The first method is to obtain a criterion judgment matrix based on the influence weights of the criterion factors, and then calculate the criterion weight vector based on the criterion judgment matrix. The second method is to obtain the criterion factor weight vector through the intention creator's selection report. For the second method, it indicates that the intention creator's preference selection mechanism has been activated, and the intention creator has selected preferred candidate intention knowledge. That is, the criterion strategy template includes the "intention creator preference characteristics" factor. In this case, the criterion factor weight vector can be directly generated based on the intention creator's selection.

[0161] As an optional embodiment, obtaining the intent knowledge weight vector of the candidate intent knowledge includes:

[0162] Based on the second influence weight of each candidate intent knowledge on each criterion factor, an intent knowledge judgment matrix is ​​obtained; the row vectors of the intent knowledge judgment matrix are summed to obtain column vectors, and the column vectors are standardized to obtain the intent knowledge weight vector.

[0163] The following example, using the method of obtaining the criterion factor weight vector through method one, illustrates the process of obtaining the criterion factor weight vector and the intent knowledge weight vector of the candidate intent knowledge.

[0164] Step (1): The first influence weights of the criterion factors on the target intent knowledge are respectively: a1, a2, ..., a i ;

[0165] The second influence weights of each candidate intent knowledge on the criterion factors are as follows: Influence weight of intent knowledge 1 on each criterion factor: b 11 ,b 12 ,...,b 1i The influence weight of intentional knowledge 2 on each criterion factor: b 21 ,b 22 ,...,b 2i And so on, the influence weight of intentional knowledge j on each criterion factor: b j1 ,b j2 ,...,b ji .

[0166] Step (2): Generate the criterion judgment matrix and the intent knowledge judgment matrix based on the above weight vector:

[0167] Criterion judgment matrix:

[0168] Intent knowledge judgment matrix:

[0169]

[0170] And so on.

[0171]

[0172] Step (3): Based on the above criterion judgment matrix and intention knowledge judgment matrix, calculate the criterion factor weight vector and the intention knowledge weight vector:

[0173] By summing the row vectors of the criterion judgment matrix and the intent knowledge judgment matrix respectively, the column vectors corresponding to the criterion judgment matrix and the intent knowledge judgment matrix are obtained. Then, the obtained column vectors are standardized to obtain the criterion factor weight vector ω. C And the intent knowledge weight vector:

[0174] As an optional embodiment, the candidate intent knowledge is analyzed based on the criterion policy-related information to determine the target intent knowledge, including:

[0175] Based on the criterion factor weight vector and the intent knowledge weight vector, calculate the matching result of each candidate intent knowledge for the intent; determine the target intent knowledge based on the matching result.

[0176] In this embodiment, the obtained matching result can be a matching score, where a higher score indicates a higher degree of matching with the intent created by the intent creator. Optionally, the matching result can be obtained by calculating the product of the criterion factor weight vector and the intent knowledge weight vector, for example:

[0177] Match score

[0178] Optionally, determining the target intent knowledge based on the matching result includes:

[0179] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is greater than or equal to the first threshold, then the intent knowledge corresponding to the highest matching result is determined to be the target intent knowledge.

[0180] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is at least less than a first threshold, then the intent knowledge selected by the intent creator is determined to be the target intent knowledge.

[0181] Optionally, after calculating the matching result of each candidate intent knowledge for the intent based on the criterion factor weight vector and the intent knowledge weight vector, the method further includes: if the highest matching result in the matching results is 0, re-acquiring the intent.

[0182] In this embodiment, the first matching result can be any one or more matching results other than the highest matching result. That is, the first matching result can include multiple results, each corresponding to different intent knowledge. The matching result can be a matching score.

[0183] When determining the target intent knowledge, if the current highest matching score is 0, the intent acquisition is repeated; otherwise, the scores S corresponding to each candidate intent knowledge (i.e., the intent knowledge corresponding to the first matching result) are determined, excluding the intent knowledge corresponding to the current highest matching score. k Matching the current highest score S max The relationship between the absolute values ​​of the comparisons and a set first threshold δ is defined. Specifically, when no absolute value is less than the first threshold (i.e., all values ​​are greater than or equal to the first threshold), the intent knowledge corresponding to the highest matching score is output as the target intent knowledge. If the absolute value of the comparison between the highest matching result and at least one of the first matching results is less than the first threshold, the intent knowledge and related data involved when the absolute value is less than the first threshold are sent to the intent creator as a selection reference report, enabling the intent creator's preference selection mechanism, allowing the intent creator to select the target intent knowledge based on the selection reference report.

[0184] When the absolute value of the comparison between the highest matching result and the first matching result is less than the first threshold, it can be said that the intent knowledge corresponding to the highest matching result is similar to the intent knowledge corresponding to the first matching result and the intent creator's intent. Therefore, the intent creator can select the target intent knowledge according to their own preferences.

[0185] Optionally, if the absolute value of the comparison between the highest matching result and a first matching result other than the highest matching result is less than at least a first threshold, the method further includes:

[0186] Send a selection reference report to the terminal, the selection reference report including: first intent knowledge corresponding to the first matching result whose absolute value is less than a first threshold, and second intent knowledge corresponding to the highest matching result, so that the intent creator selects the target intent knowledge from the first intent knowledge and the second intent knowledge; receive a selection report sent by the terminal, the selection report including: the target intent knowledge selected by the intent creator and an updated criterion factor weight vector.

[0187] The selection reference report may include at least one of the following:

[0188] The first intent knowledge and the matching results corresponding to the first intent knowledge;

[0189] The criterion factor weight vector.

[0190] This embodiment describes the implementation process of enabling the intent creator's preference selection mechanism. Specifically, when the absolute value of the comparison between the highest matching result and at least one of the other matching results (i.e., the first matching result) is less than a first threshold, a selection reference report is sent to the intent creator. This selection reference report includes second intent knowledge corresponding to the highest matching result, and also includes first intent knowledge corresponding to the first matching result whose absolute value is less than the first threshold. The intent creator can determine the target intent knowledge based on their own preferences, according to the provided intent knowledge and / or the matching scores of each intent knowledge, and generate a corresponding selection report.

[0191] For example, assuming the first threshold is Y, and candidate intent knowledge includes intent knowledge 1-5, where intent knowledge 1 has the highest matching result with the intent creator's intent, then the absolute values ​​of the comparisons between the matching results of intent knowledge 1 and the corresponding intent knowledge 2-4 can be calculated. If the absolute value of the comparison between the matching result of intent knowledge 1 and the matching results of intent knowledge 2 and 3 is less than Y, it indicates that the three are similar in degree of matching with the intent, and a selection reference report can be sent to the terminal. This selection reference report includes intent knowledge 1, intent knowledge 2, and intent knowledge 3, and may also include the matching results and criterion factor weight vectors corresponding to intent knowledge 1, intent knowledge 2, and intent knowledge 3 respectively. The intent creator can select target intent knowledge based on their own preferences, for example, selecting intent knowledge 1 with the highest matching result as the target intent knowledge.

[0192] Since the intent creator selects target intent knowledge based on their own preferences, the influence weights of candidate intent knowledge on the criterion factors change, and the criterion factor weight vector also changes. After the intent creator selects candidate intent knowledge, an updated criterion factor weight vector is generated. The selection report fed back by the terminal may include the target intent knowledge selected by the intent creator and the updated criterion factor weight vector. Upon receiving the selection report, the candidate intent knowledge selected by the intent creator is executed as the target intent knowledge.

[0193] Optionally, after receiving the selection report sent by the receiving terminal, the method further includes: performing self-learning on the criterion policy and updating the criterion policy template based on the selection report.

[0194] In this embodiment, upon receiving the selection report, the criterion strategy undergoes self-learning to obtain updated data, including but not limited to the latest dynamic criterion strategy template and related data. Specifically, the preference data of intent creators can be extracted from the selection report and stored in an intent creator preference database. During the self-learning process, the intent creator preference data stored in the database can be retrieved to update the current strategy's data and indicator parameters. Self-learning refers to determining whether the current criterion strategy needs adjustment based on rules. Adjustment methods include, but are not limited to, adding or deleting criterion factors in the criterion strategy, and adjusting the influence weight of criterion factors on the target layer.

[0195] As an optional embodiment, if the intent creator preference selection mechanism is activated, after performing criterion policy self-learning, the generated updated criterion policy template includes the "intent creator preference feature" factor. When the criterion factor includes the intent creator preference feature, after matching the criterion policy template for the intent, the method may further include: updating the matched criterion policy template (i.e., updating the matched criterion policy template without updating the policy templates in the criterion policy library), and the updated criterion policy template does not include the intent creator preference feature.

[0196] In this embodiment, if the criterion strategy template for intent matching created for the intent creator contains the "intent creator preference feature" factor, the "intent creator preference feature" can be hidden. The new current matching criterion strategy template formed after hiding the "intent creator preference feature" factor is used as the criterion strategy template for intent matching. In this updated criterion strategy template, since the selection report fed back by the intent creator during the intent creator preference selection mechanism contains the criterion factor weight vector, it is not necessary to calculate the criterion factor weight vector through the criterion judgment matrix when analyzing candidate intent knowledge. That is, the relevant operations on the criterion judgment matrix can be omitted, and the criterion weight vector can be directly used for corresponding calculation.

[0197] In embodiments of the present invention, when analyzing candidate intent knowledge based on the criterion policy-related information, the analysis can be performed at different levels, such as a dynamic hierarchical selection model for intent knowledge. Figure 2 As shown, the model is divided into three layers: a target layer, a criterion layer, and a solution layer. Specifically, the target layer represents the final goal of the current analysis and determines the target intent knowledge; the criterion layer represents the influencing factors involved in selecting the final target intent knowledge; and the solution layer represents alternative solutions, including multiple candidate intent knowledge. The following uses this dynamic hierarchical selection model of intent knowledge as an example to illustrate the implementation process of intent recognition and matching in this invention. Specifically, it includes:

[0198] 1): Create alternative sets of intent knowledge;

[0199] 2) Construct a dynamic hierarchical selection model based on intent knowledge. The model includes a target layer, a criterion layer, and a solution layer. The criterion layer includes multiple criterion factors 1-i contained in the criterion strategy template for intent matching. The solution layer includes multiple intent knowledge points 1-j from the intent knowledge base.

[0200] 3): The intention knowledge dynamic hierarchical selection model obtains the influence weights of each criterion factor on the target layer, namely a1, a2, ..., a i And obtain the influence weights of the scheme layer on the criterion factors, which are the influence weights of intention knowledge 1 on each criterion factor: b 11 ,b 12 ,...,b 1i The influence weight of intentional knowledge 2 on each criterion factor: b 21 ,b 22 ,...,b 2i The influence weight of intentional knowledge j on each criterion factor: b j1 ,b j2 ,...,b ji .

[0201] 4): Generate the criterion judgment matrix (i.e., the criterion-level judgment matrix) and the intent knowledge judgment matrix (i.e., the scheme-level judgment matrix) based on the above weight vector.

[0202] 5): Sum the row vectors of the criterion layer judgment matrix and the scheme layer judgment matrix respectively, and standardize the resulting column vectors to obtain the criterion factor weight vector (i.e., the criterion layer weight vector) and the intent knowledge weight vector (i.e., the scheme layer weight vector); calculate the matching score of each intent knowledge in the scheme layer based on the criterion layer weight vector and the scheme layer weight vector.

[0203] 6): Based on the matching score, determine the target intent knowledge of the target layer. The target intent knowledge is the optimal intent knowledge selection. The specific implementation steps are not described here.

[0204] 7): Execute the selected target intent knowledge.

[0205] In step 2) of this embodiment, the implementation process of the intent matching criterion strategy template is as follows.

[0206] For example: if, based on the current system state and the characteristic information of the intent, it is determined that the current resource utilization rate is high and the intent creator does not accept the downgraded service, then the criterion strategy template for matching the current intent includes three factors: resource utilization, intent knowledge matching degree, and historical execution count, which are used as criterion factors; if, based on the current system state and the characteristic information of the intent, it is determined that the current resource utilization rate is low and the user does not accept the downgraded service, then the criterion strategy template for matching the current intent includes three factors: intent knowledge matching degree, historical execution count, and historical execution success rate.

[0207] The above examples can be described using the Event-Condition-Action (ECA) strategy, and the corresponding strategy schemes can be implemented using general strategy engines (such as Apex, Drools, etc.). The ECA strategy is described as follows:

[0208] Strategy 1:

[0209] event:

[0210] Received the criteria layer construction request and related indicator information;

[0211] condition:

[0212] a) The current resource occupancy rate is greater than 80%;

[0213] b) The "QoS" value is greater than 8;

[0214] c) The value of the indicator "Accept Downgrade" is False;

[0215] action

[0216] a) Generate a criterion strategy template that combines three factors: "resource consumption", "intent knowledge matching degree" and "historical execution count".

[0217] Strategy 2:

[0218] event:

[0219] Received the criteria layer construction request and related indicator information;

[0220] condition:

[0221] a) The current resource utilization rate is less than 40%.

[0222] b) The "QoS" indicator value is greater than 8

[0223] c) The value of the indicator "Accept Downgrade" is False.

[0224] action:

[0225] a) Generate a criterion strategy template that combines three factors: "intent knowledge matching degree", "historical execution count", and "historical execution success rate".

[0226] Default policy:

[0227] event:

[0228] Received the criteria layer construction request and related indicator information;

[0229] condition:

[0230] a) All condition entries in the dynamic criterion strategy library that already have other strategies are evaluated as "False";

[0231] action:

[0232] a) Generate a criterion strategy template that combines the two factors of "intent knowledge matching degree" and "historical execution count".

[0233] It should be noted that the metrics and corresponding values ​​described in the policy example of this embodiment are illustrative examples and are for reference only in subsequent implementation, and do not have a mandatory meaning. For example, in this example, the QoS metric is considered to be represented by a number from 1 to 10, indicating the service requirements of the intent creator, with 1 being the lowest and 10 being the highest. These metrics can be expressed in other ways.

[0234] In embodiments of the present invention, candidate intent knowledge is analyzed based on criterion-strategy related information. Target intent knowledge matching the intent created by the intent creator is determined from the candidate intent knowledge. The criterion-strategy related information includes multiple criterion factors that influence the target intent knowledge. Comprehensive evaluation of intent knowledge can be performed based on multiple factors, making the system more robust and improving intent recognition performance.

[0235] The intent recognition and matching method of this invention can construct a criterion layer by matching corresponding strategies from a self-learning dynamic criterion strategy library based on system state and intent features, and perform hierarchical analysis of candidate intent knowledge sets. It can also perform criterion strategy self-learning based on intent creator preferences, thereby dynamically modifying the criterion strategy templates in the dynamic criterion strategy library. It considers the influence of multiple factors (such as resource consumption and user preference characteristics) on intent knowledge selection, and can establish a dynamic hierarchical intent knowledge selection model. This model can construct an optimal criterion layer by matching corresponding strategies from a self-learning dynamic criterion strategy library based on system state and intent features, and perform multi-element comprehensive evaluation, resulting in a more comprehensive and flexible consideration of the optimal intent knowledge selection.

[0236] Furthermore, when determining the target intent knowledge based on the matching results, the case where the scores of the selected intent knowledge are similar is considered, and an intent creator preference selection mechanism is introduced. This mechanism can update and self-learn the intent knowledge preference selection data and the criterion strategy library, making the system more robust.

[0237] It should be noted that the execution entity of the intent recognition and matching method provided in this application embodiment can be an intent recognition and matching device, or a control module in the intent recognition and matching device for the intent recognition and matching method. This application embodiment uses the execution of the intent recognition and matching method by an intent recognition and matching device as an example to illustrate the intent recognition and matching device provided in this application embodiment.

[0238] like Figure 3 As shown in the illustration, this application also provides an intent recognition and matching device 300, comprising:

[0239] The intent acquisition module 310 is used to acquire intents created by the intent creator.

[0240] The criterion data acquisition module 320 is used to acquire criterion policy-related information that matches the intent;

[0241] The intent knowledge analysis module 330 is used to analyze candidate intent knowledge based on the criteria and strategy related information to determine the target intent knowledge;

[0242] The sending module 340 is used to send the target intent knowledge to the intent execution module.

[0243] Optionally, the criterion data acquisition module includes:

[0244] The matching unit is used to match a criterion strategy template for the intent based on the current system state and the feature information of the intent. The criterion strategy template includes criterion factors that affect the knowledge of the target intent.

[0245] The first acquisition unit is used to acquire the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge.

[0246] Optionally, the first acquisition unit is specifically used for:

[0247] Based on the first influence weight of each criterion factor on the target intent knowledge, a criterion judgment matrix is ​​obtained; the column vector is obtained by summing the row vectors of the criterion judgment matrix, and the column vector is standardized to obtain the criterion factor weight vector.

[0248] or

[0249] The weight vector of the criteria factors is obtained from the selection report sent by the terminal.

[0250] Optionally, the first acquisition unit is specifically used to: acquire an intent knowledge judgment matrix based on the second influence weight of each candidate intent knowledge on each criterion factor;

[0251] The row vectors of the intent knowledge judgment matrix are summed to obtain column vectors, and the column vectors are standardized to obtain the intent knowledge weight vector.

[0252] Optionally, the intent knowledge analysis module includes:

[0253] The first calculation unit is used to calculate the matching result of each candidate intent knowledge for the intent based on the criterion factor weight vector and the intent knowledge weight vector.

[0254] The first determining unit is used to determine the target intent knowledge based on the matching result.

[0255] Optionally, the first determining unit is specifically used for:

[0256] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is greater than or equal to the first threshold, then the intent knowledge corresponding to the highest matching result is determined to be the target intent knowledge.

[0257] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is at least less than a first threshold, then the intent knowledge selected by the intent creator is determined to be the target intent knowledge.

[0258] Optionally, the device further includes;

[0259] The first acquisition module is used to reacquire the intent if the highest matching result in the matching results is 0.

[0260] Optionally, the device further includes:

[0261] A preference selection module is used to send a selection reference report to the terminal. The selection reference report includes: first intent knowledge corresponding to the first matching result whose absolute value is less than a first threshold, and second intent knowledge corresponding to the highest matching result, so that the intent creator can select the target intent knowledge from the first intent knowledge and the second intent knowledge.

[0262] A receiving module is used to receive a selection report sent by a terminal, the selection report including: the intent knowledge selected by the intent creator and an updated criterion factor weight vector.

[0263] Optionally, the selection reference report may also include at least one of the following:

[0264] The first intent and the second intent are respectively the corresponding matching results;

[0265] The criterion factor weight vector.

[0266] Optionally, the device further includes:

[0267] The dynamic criterion strategy self-learning module is used to learn the criterion strategy and update the criterion strategy template based on the selection report.

[0268] Optionally, the criterion factors include at least one of the following:

[0269] Network resource metrics;

[0270] Business quality metrics;

[0271] Intent-knowledge matching degree;

[0272] Historical execution of intent knowledge;

[0273] Intent creator preference characteristics.

[0274] Optionally, if the criterion factors include the intent creator preference characteristics, the apparatus further includes:

[0275] The update module is used to update the matched criterion policy template, wherein the updated criterion policy template does not include the intent creator preference feature.

[0276] The following describes the process by which the intent recognition and matching device of an embodiment of the present invention implements the intent recognition and matching method, such as... Figure 4 As shown, it specifically includes:

[0277] 1: Intent Acquisition. The intent acquisition module can acquire the intent created by the intent creator through the UI interface, voice device or other system interfaces.

[0278] 2: Intent Knowledge Selection. The following describes the repository and related intent modules involved in the intent recognition and matching device described in this embodiment.

[0279] (1) You can create an intent knowledge base and store intent standard description files in the intent knowledge base.

[0280] (2) Intent knowledge analysis module, which may include Figure 2 The intent knowledge dynamic hierarchical selection model shown here, the process of intent recognition and matching implemented by the intent knowledge dynamic hierarchical selection model will not be described in detail here.

[0281] The dynamic hierarchical selection model for intent knowledge obtains candidate intent knowledge through an intent knowledge base.

[0282] (3) Preference selection module, used to execute the intention creator preference selection mechanism. In the intention creator preference selection mechanism, the system sends a selection reference report to the intention creator. After the intention creator selects candidate intention knowledge, it returns a selection report to the system. The information contained in the report includes, but is not limited to, the intention creator selection result and the weight vector of the criterion factors of the intention creator's preference.

[0283] (4) Preference database, used to store data related to the preference selection of intent creators, including but not limited to information on the selection report in the preference selection mechanism of intent creators.

[0284] (5) The dynamic criterion strategy self-learning module can be used to obtain the preference data of the intent creator stored in the preference database, and perform self-learning and data update of the indicator parameters of the current strategy.

[0285] (6) Dynamic Criterion Strategy Library, used to store criterion strategy templates. The dynamic criterion strategy library can obtain updated data after the dynamic criterion strategy self-learning module has learned, including but not limited to the latest dynamic criterion strategy templates and related data.

[0286] When performing intent knowledge hierarchy analysis for the current intent, it is necessary to interact with the criterion data acquisition module to match the corresponding criterion strategy template based on the current system state and intent feature information, in order to construct the criterion layer in the dynamic hierarchical selection model of intent knowledge.

[0287] (7) Criterion data acquisition module, including but not limited to the following capabilities:

[0288] a: It can interact with the dynamic criterion strategy library to obtain the current system status information and intent feature information, and interact with the dynamic criterion strategy library to match the most suitable criterion strategy template. The system status information includes, but is not limited to, system resource usage, idle / busy time, and the degree of service reduction acceptable to the intent creator.

[0289] b: It can interact with the preference database, including: if the currently matched criterion strategy template contains the "intent creator preference feature" factor, checking the intent creator preference data and directly generating the criterion layer weight vector of the corresponding factor in the criterion strategy template.

[0290] c: It can interact with the intent knowledge analysis module, including:

[0291] (i) When the currently matched criterion strategy template does not contain the "intent creator preference feature" factor, the currently matched criterion strategy template and related data are sent to the intent knowledge analysis module, including but not limited to criterion factors, the influence weight of criterion factors on the target layer of the dynamic hierarchical selection model of intent knowledge (i.e., the influence weight of criterion factors on the target intent knowledge), and the influence weight of the scheme layer of the dynamic hierarchical selection model of intent knowledge on the criterion factors (i.e., the influence weight of each candidate intent knowledge on the criterion factors).

[0292] (ii) When the currently matched criterion strategy template contains the "intent creator preference feature" factor, the "intent creator preference feature" factor is hidden, a new currently matched criterion strategy template is formed, and the criterion strategy template and related data are sent to the intent knowledge analysis module, including but not limited to criterion factors, the criterion layer weight vector of the corresponding factor in the criterion strategy template, and the influence weight of the scheme layer on the criterion factors. At the same time, the intent knowledge analysis module is notified that when performing hierarchical analysis, since the criterion layer judgment matrix is ​​missing, the relevant operations on the criterion layer judgment matrix are omitted, and the criterion layer weight vector is used directly for the corresponding calculation.

[0293] Specifically, the intent knowledge analysis module may include a judgment unit, used to determine whether the candidate intent knowledge can be used as the target intent knowledge based on the matching results of the candidate intent knowledge with the intent. Wherein:

[0294] If the matching score of the highest-scoring intent knowledge is 0, then intent acquisition is performed again;

[0295] Otherwise, determine the matching score S of each candidate intent, excluding the intent knowledge corresponding to the highest matching score. k Matching the current highest score S max The relationship between the absolute value of the comparison and the set first threshold δ is used to determine |S k -S max |<δ,1≤k≤j,k≠max。When there is no case where the absolute value of the comparison is less than the first threshold, the intent knowledge corresponding to the current highest matching score is output as the target intent knowledge; otherwise, the intent knowledge and related data involved when the absolute value of the comparison is less than the first threshold are selected and sent to the intent creator to enable the intent creator preference selection mechanism.

[0296] 3: Intent Execution. Execute the selected target intent knowledge mapping strategy or workflow. It should be noted that this step is optional; the intent execution module can determine whether to execute the target intent knowledge mapping strategy or workflow.

[0297] The following specific example illustrates the execution process of the intent recognition and matching device in implementing the intent recognition and matching method.

[0298] Taking the following scenario as an example: An intent creator requests a high-quality video live streaming service and does not accept service degradation; after obtaining the intent created by the intent creator, the intent knowledge base is used as the intent knowledge set. Based on this, a scheme layer of the dynamic hierarchical selection model for intent knowledge is created; the criterion data acquisition module obtains that the current system resource utilization is high and the intent creator does not accept service degradation. Based on the current system status information and intent feature information, it interacts with the dynamic criterion strategy library to match the intent created by the current intent creator with a criterion strategy template. The resource utilization, intent knowledge matching degree, and historical execution count of this template are used as the criterion layer; the target layer is the optimal intent knowledge selection, i.e., the obtained target intent knowledge. Resource utilization refers to a comprehensive indicator of the utilization of various resources in the system. The constructed dynamic hierarchical selection model for intent knowledge is as follows: Figure 5 As shown.

[0299] based on Figure 5 The model shown requires the criterion data acquisition module to send the influence weights of criterion factors on the target layer to the intent knowledge analysis module, namely, resource consumption level a1, intent knowledge matching degree a2, and historical execution count a3; and the influence weights of the criterion factors on the solution layer, namely, the influence weights of intent knowledge 1 on resource consumption, intent knowledge matching degree, and historical execution count b. 11 ,b 12 ,b 13The weight b of the influence of intent knowledge 2 on resource consumption, intent knowledge matching degree, and historical execution count. 21 ,b 22 ,b 23 .

[0300] Based on the above data, the intent-knowledge dynamic hierarchical selection model generates a criterion-level judgment matrix:

[0301]

[0302] And the scheme layer judgment matrix:

[0303]

[0304] By summing the row vectors of the above judgment matrix and standardizing the resulting column vectors, we can obtain the criterion layer weight vector ω. C Scheme layer weight vector The final scores for each plan are:

[0305]

[0306] Current highest match score S max When |S = 0, it means that the current intent knowledge does not match the intent creator's intent at all, and the schematic creator needs to re-enter the intent; when |S k -S max When |<δ, 1≤k≤2, and k≠max, the intent knowledge and related data involved when the absolute value is less than the threshold δ need to be selected and sent to the intent creator. The intent creator selects the preferred intent knowledge based on the report and their own preferences and outputs it to the intent execution module, and returns a selection report to the system, which is then stored in the preference database. Otherwise, the intent knowledge with the highest current score is output to the intent execution module.

[0307] The dynamic criterion strategy self-learning module learns from the current system state and intent characteristics based on data in the preference database. For example, in this embodiment, if the currently matched criterion strategy template does not contain intent creator preference characteristics, the dynamic strategy self-learning module can add these characteristics to the criterion strategy template after learning based on the rules. Once the intent creator preference data is complete, the likelihood of the system subsequently calling the intent creator preference selection mechanism is relatively low.

[0308] In embodiments of the present invention, candidate intent knowledge is analyzed based on criterion-strategy related information. Target intent knowledge matching the intent created by the intent creator is determined from the candidate intent knowledge. The criterion-strategy related information includes multiple criterion factors that influence the target intent knowledge. Comprehensive evaluation of intent knowledge can be performed based on multiple factors, making the system more robust and improving intent recognition performance.

[0309] In embodiments of this invention, a criterion layer can be constructed by matching corresponding strategies from a self-learning dynamic criterion strategy library based on system state and intent characteristics, and by performing hierarchical analysis of candidate intent knowledge sets. Criterion strategies can be self-learned based on intent creator preferences, thereby dynamically modifying the criterion strategy templates in the dynamic criterion strategy library. The invention considers the influence of multiple factors (such as resource consumption and user preference characteristics) on intent knowledge selection, and can establish a dynamic hierarchical intent knowledge selection model. This model can construct an optimal criterion layer by matching corresponding strategies from a self-learning dynamic criterion strategy library based on system state and intent characteristics, and perform multi-element comprehensive evaluation, resulting in a more comprehensive and flexible consideration of the optimal intent knowledge selection.

[0310] Furthermore, when determining the target intent knowledge based on the matching results, the system considers cases where the scores of the selected intent knowledge are similar and introduces a user preference selection mechanism. This mechanism can update and self-learn the intent knowledge preference selection data and the criterion strategy library, making the system more robust.

[0311] It should be noted that the intent recognition and matching device provided in this embodiment of the invention can implement all the method steps implemented in the above intent recognition and matching method embodiment, and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0312] To achieve the above objectives, such as Figure 6 As shown, an intent recognition and matching device 600 according to an embodiment of the present invention includes a processor 610 and a transceiver 620, wherein,

[0313] The transceiver 620 is used to: acquire an intent created by an intent creator; acquire criterion policy-related information matching the intent;

[0314] The processor 610 is configured to: analyze candidate intent knowledge based on the criterion policy related information to determine target intent knowledge; and send the target intent knowledge to the intent execution module.

[0315] Optionally, the transceiver acquires criterion policy-related information matching the intent, specifically including:

[0316] Based on the current system state and the characteristic information of the intent, a criterion strategy template is matched for the intent. The criterion strategy template includes criterion factors that influence the knowledge of the target intent.

[0317] Obtain the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge.

[0318] Optionally, the transceiver acquires the criterion factor weight vector of the criterion factors, specifically including:

[0319] Based on the first influence weight of each criterion factor on the target intent knowledge, a criterion judgment matrix is ​​obtained; the column vector is obtained by summing the row vectors of the criterion judgment matrix, and the column vector is standardized to obtain the criterion factor weight vector.

[0320] or

[0321] The weight vector of the criteria factors is obtained from the selection report sent by the terminal.

[0322] Optionally, the transceiver acquires the intent knowledge weight vector of the candidate intent knowledge, specifically including:

[0323] Based on the second influence weight of each candidate intent knowledge on each criterion factor, obtain the intent knowledge judgment matrix;

[0324] The row vectors of the intent knowledge judgment matrix are summed to obtain column vectors, and the column vectors are standardized to obtain the intent knowledge weight vector.

[0325] Optionally, the processor analyzes the candidate intent knowledge based on the criterion policy-related information to determine the target intent knowledge, including:

[0326] Based on the criterion factor weight vector and the intent knowledge weight vector, calculate the matching result of each candidate intent knowledge for the intent;

[0327] The target intent knowledge is determined based on the matching results.

[0328] Optionally, the processor determines the target intent knowledge based on the matching result, including:

[0329] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is greater than or equal to the first threshold, then the intent knowledge corresponding to the highest matching result is determined to be the target intent knowledge.

[0330] If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is at least less than a first threshold, then the intent knowledge selected by the intent creator is determined to be the target intent knowledge.

[0331] Optionally, after calculating the matching result of each candidate intent knowledge for the intent based on the criterion factor weight vector and the intent knowledge weight vector, the transceiver is further configured to;

[0332] If the highest matching result in the matching results is 0, the intent is retrieved again.

[0333] Optionally, if the absolute value of the comparison between the highest matching result and other matching results is less than a first threshold, the transceiver is further configured to:

[0334] A selection reference report is sent to the terminal. The selection reference report includes: first intent knowledge corresponding to the first matching result whose absolute value is less than a first threshold, and second intent knowledge corresponding to the highest matching result, so that the intent creator can select the target intent knowledge from the first intent knowledge and the second intent knowledge.

[0335] The receiving terminal sends a selection report, which includes: the target intent knowledge selected by the intent creator and an updated criterion factor weight vector.

[0336] Optionally, the selection reference report includes at least one of the following:

[0337] The matching results corresponding to the first intent knowledge and the second intent knowledge, respectively;

[0338] The criterion factor weight vector.

[0339] Optionally, after receiving the selection report sent by the receiving terminal, the processor is further configured to:

[0340] Based on the selection report, the criterion strategy is self-learned and the criterion strategy template is updated.

[0341] Optionally, the criterion factors include at least one of the following:

[0342] Network resource metrics;

[0343] Business quality metrics;

[0344] Intent-knowledge matching degree;

[0345] Historical execution of intent knowledge;

[0346] Intent creator preference characteristics.

[0347] Optionally, if the criterion factors include the intent creator preference features, after matching the intent with a criterion strategy template, the processor is further configured to:

[0348] The matched criterion policy template is updated, and the updated criterion policy template does not include the intent creator preference feature.

[0349] It should be noted that the intent recognition and matching device provided in the embodiments of the present invention can implement all the method steps implemented in the above intent recognition and matching method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0350] Another embodiment of the electronic device of the present invention, such as Figure 7 As shown, it includes a transceiver 710, a processor 700, a memory 720, and a program or instructions stored in the memory 720 and executable on the processor 700; when the processor 700 executes the program or instructions, it implements the above-mentioned intent recognition and matching method.

[0351] The transceiver 710 is used to receive and send data under the control of the processor 700.

[0352] Among them, Figure 7 In this context, the bus architecture may include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 700) and memory (memory 720). The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 710 may be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. The processor 700 is responsible for managing the bus architecture and general processing, and the memory 720 may store data used by the processor 700 during operation.

[0353] An embodiment of the present invention provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the steps in the intent recognition and matching method described above and achieve the same technical effect. To avoid repetition, the details will not be repeated here.

[0354] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0355] It should be further noted that the electronic devices described in this specification include, but are not limited to, smartphones, tablets, etc., and many of the described functional components are referred to as modules in order to more specifically emphasize the independence of their implementation.

[0356] In this embodiment of the invention, the module can be implemented in software so that it can be executed by various types of processors. For example, an identified executable code module may include one or more physical or logical blocks of computer instructions, which may be constructed as objects, procedures, or functions. Nevertheless, the executable code of the identified module does not need to be physically located together, but may include different instructions stored in different bits, which, when logically combined, constitute the module and achieve the module's intended purpose.

[0357] In practice, an executable code module can be a single instruction or many instructions, and can even be distributed across multiple different code segments, different programs, and across multiple memory devices. Similarly, operational data can be identified within the module and can be implemented in any suitable form and organized within any suitable type of data structure. This operational data can be collected as a single dataset or distributed across different locations (including different storage devices), and can exist, at least in part, solely as electronic signals within the system or network.

[0358] When a module can be implemented using software, considering the current level of hardware technology, modules that can be implemented in software can be implemented using hardware circuits by those skilled in the art to achieve the corresponding functions, without considering cost. These hardware circuits include conventional very-large-scale integrated circuits (VLSI) or gate arrays, as well as existing semiconductors such as logic chips and transistors, or other discrete components. Modules can also be implemented using programmable hardware devices, such as field-programmable gate arrays, programmable array logic, and programmable logic devices.

[0359] The exemplary embodiments described above are with reference to the accompanying drawings. Many different forms and embodiments are feasible without departing from the spirit and teachings of the invention. Therefore, the invention should not be construed as limiting the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided to make the invention complete and convey the scope of the invention to those skilled in the art. In these drawings, component dimensions and relative dimensions may be exaggerated for clarity. The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. As used herein, unless clearly indicated otherwise, the singular forms “a,” “an,” and “the” are intended to include all such forms. It will be further understood that the terms “comprising” and / or “including”, when used in this specification, indicate the presence of the stated features, integers, steps, operations, components, and / or elements, but do not exclude the presence or addition of one or more other features, integers, steps, operations, components, and / or groups thereof. Unless otherwise indicated, when stated, a range of values ​​includes the upper and lower limits of the range and any subranges in between.

[0360] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An intent recognition and matching method, characterized in that, include: Retrieve the intent created by the intent creator; Obtain relevant information on criteria and policies that match the stated intent; Based on the relevant information of the aforementioned criteria and strategies, the candidate intent knowledge is analyzed to determine the target intent knowledge; Send the target intent knowledge to the intent execution module; The acquisition of criteria policy-related information matching the intent includes: Based on the current system state and the characteristic information of the intent, a criterion strategy template is matched for the intent. The criterion strategy template includes criterion factors that influence the knowledge of the target intent. Obtain the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge.

2. The method according to claim 1, characterized in that, Obtaining the criterion factor weight vector of the criterion factors includes: Based on the first influence weight of each criterion factor on the target intent knowledge, a criterion judgment matrix is ​​obtained; the column vector is obtained by summing the row vectors of the criterion judgment matrix, and the column vector is standardized to obtain the criterion factor weight vector. or The weight vector of the criteria factors is obtained from the selection report sent by the terminal.

3. The method according to claim 1, characterized in that, Obtaining the intent knowledge weight vector of the candidate intent knowledge includes: Based on the second influence weight of each candidate intent knowledge on each criterion factor, obtain the intent knowledge judgment matrix; The row vectors of the intent knowledge judgment matrix are summed to obtain column vectors, and the column vectors are standardized to obtain the intent knowledge weight vector.

4. The method according to claim 1, characterized in that, Based on the relevant information of the aforementioned criteria and strategies, candidate intent knowledge is analyzed to determine target intent knowledge, including: Based on the criterion factor weight vector and the intent knowledge weight vector, calculate the matching result of each candidate intent knowledge for the intent; The target intent knowledge is determined based on the matching results.

5. The method according to claim 4, characterized in that, Determining the target intent knowledge based on the matching result includes: If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is greater than or equal to the first threshold, then the intent knowledge corresponding to the highest matching result is determined to be the target intent knowledge. If the highest matching result in the matching results is not 0, and the absolute value of the comparison between the highest matching result and the first matching result other than the highest matching result is at least less than a first threshold, then the intent knowledge selected by the intent creator is determined to be the target intent knowledge.

6. The method according to claim 4, characterized in that, After calculating the matching result of each candidate intent knowledge for the intent based on the criterion factor weight vector and the intent knowledge weight vector, the method further includes: If the highest matching result in the matching results is 0, the intent is retrieved again.

7. The method according to claim 5, characterized in that, If, in the case that at least one of the absolute values ​​of the comparison between the highest matching result and the first matching result other than the highest matching result is less than a first threshold, the method further includes: A selection reference report is sent to the terminal. The selection reference report includes: first intent knowledge corresponding to the first matching result whose absolute value is less than a first threshold, and second intent knowledge corresponding to the highest matching result, so that the intent creator can select the target intent knowledge from the first intent knowledge and the second intent knowledge. The receiving terminal sends a selection report, which includes: the target intent knowledge selected by the intent creator and an updated criterion factor weight vector.

8. The method according to claim 7, characterized in that, The selection reference report also includes at least one of the following: The matching results corresponding to the first intent knowledge and the second intent knowledge, respectively; The criterion factor weight vector.

9. The method according to claim 7, characterized in that, After receiving the selection report sent by the receiving terminal, the method further includes: Based on the selection report, the criterion strategy is self-learned and the criterion strategy template is updated.

10. The method according to claim 1, characterized in that, The criteria factors include at least one of the following: Network resource metrics; Business quality metrics; Intent-knowledge matching degree; Historical execution of intent knowledge; Intent creator preference characteristics.

11. The method according to claim 10, characterized in that, When the criterion factors include the intent creator preference characteristics, after matching the intent with a criterion strategy template, the method further includes: The matched criterion policy template is updated, and the updated criterion policy template does not include the intent creator preference feature.

12. An intent recognition and matching device, characterized in that, include: The intent retrieval module is used to retrieve intents created by the intent creator. The criteria data acquisition module is used to acquire criteria policy-related information that matches the intent; The intent knowledge analysis module is used to analyze candidate intent knowledge based on the criteria and strategy information to determine the target intent knowledge; The sending module is used to send the target intent knowledge to the intent execution module; The criterion data acquisition module includes: The matching unit is used to match a criterion strategy template for the intent based on the current system state and the feature information of the intent. The criterion strategy template includes criterion factors that affect the knowledge of the target intent. The first acquisition unit is used to acquire the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge.

13. An intent recognition and matching device, characterized in that, include: Transceiver and processor; The transceiver is used to: acquire intents created by the intent creator; Obtain relevant information on criteria and policies that match the stated intent; The processor is configured to: analyze candidate intent knowledge based on the criterion policy-related information to determine target intent knowledge; The transceiver is also used to: send the target intent knowledge to the intent execution module; The transceiver acquires criterion policy information related to the intent, specifically including: Based on the current system state and the characteristic information of the intent, a criterion strategy template is matched for the intent. The criterion strategy template includes criterion factors that influence the knowledge of the target intent. Obtain the criterion factor weight vector of the criterion factors and the intent knowledge weight vector of the candidate intent knowledge.

14. An electronic device comprising: A transceiver, a processor, a memory, and a program or instructions stored in the memory and executable on the processor; characterized in that, when the processor executes the program or instructions, it implements the intent recognition matching method as described in any one of claims 1-11.

15. A readable storage medium having a program or instructions stored thereon, characterized in that, When the program or instructions are executed by the processor, they implement the steps of the intent recognition and matching method as described in any one of claims 1-11.

Images