Task processing method and electronic device

Abstract

This application discloses a task processing method and an electronic device. The method includes: in response to obtaining a task command, determining target task features of a target task indicated by the task command; determining target guidance information from at least one guidance information based on the target task features, wherein the guidance information is generated based on a historical command cluster, and the historical command cluster includes at least one historical task command belonging to the same intent category; the historical command cluster corresponding to the target guidance information includes: historical task commands whose corresponding task features match the target task features; and processing the target task indicated by the task command based on the task command and the target guidance information.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a task processing method and an electronic device. Background Technology

[0002] With the continuous development of artificial intelligence technology, the application of AI-based intelligent agents is becoming increasingly widespread. However, tasks performed by intelligent agents can sometimes result in errors. Therefore, reducing these errors is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0003] On the one hand, this application provides a task processing method, including:

[0004] In response to receiving a task command, determine the target task characteristics of the target task indicated by the task command;

[0005] Based on the target task characteristics, target guidance information is determined from at least one guidance information, wherein the guidance information is generated based on a historical command cluster, and the historical command cluster includes at least one historical task command belonging to the same intent category; the historical command cluster corresponding to the target guidance information includes: historical task commands whose corresponding task characteristics match the target task characteristics;

[0006] Based on the task command and the target guidance information, process the target task indicated by the task command.

[0007] In one possible implementation, determining the target task characteristics of the target task indicated by the task command includes:

[0008] Determine the target index tag corresponding to the task command, wherein the target index tag is used to characterize the summary information of the target task indicated by the task command;

[0009] The step of determining target guidance information from at least one guidance information based on the target task characteristics includes:

[0010] Based on the index tag set of each of the at least one type of guidance information and the target index tag, a target guidance information is determined from the at least one type of guidance information, wherein the index tag set includes at least one index tag, and the index tag set of the target guidance information includes index tags that match the target index tag.

[0011] In yet another possible implementation, the task processing method also includes:

[0012] Determine the target intent category corresponding to the command intent of the task command;

[0013] Determine the guidance information set corresponding to the target intent category, the guidance information set including: at least one guidance information and a set of index tags corresponding to the guidance information.

[0014] In another possible implementation, the guidance information is strategy information used to avoid task execution errors;

[0015] The historical command cluster includes at least one first historical task command that was executed incorrectly, and at least one second historical task command that is semantically similar to the first historical task command and was executed correctly.

[0016] The historical command cluster is associated with error description information, which is generated based on at least a portion of the first historical task commands in the historical command cluster and is used to describe the error cause information of the first historical task commands in the historical command cluster.

[0017] The guidance information is generated based on the error description information corresponding to the historical command cluster, as well as each historical task command in the historical command cluster, the actual processing result of the historical task command, and the correct processing result of the historical task command marked.

[0018] In yet another possible implementation, the guidance information is obtained in the following manner:

[0019] Based on the error description information corresponding to the historical command cluster, as well as at least some historical task commands in the historical command cluster, the actual processing results of the at least some historical task commands, and the labeled correct processing results, candidate guidance information is determined using a model. The candidate guidance information is strategy information for avoiding the execution errors of the historical task commands.

[0020] Based on the candidate guidance information, each historical task command in the historical command cluster is processed to obtain the test processing result of the historical task command;

[0021] Based on the test processing results and the marked correct processing results of the historical task commands, the candidate guidance information is optimized, and the optimized candidate guidance information is determined as the guidance information corresponding to the historical command cluster.

[0022] In another possible implementation, the candidate guidance information is optimized based on the test processing results and the labeled correct processing results of the historical task commands, and the optimized candidate guidance information is determined as the guidance information corresponding to the historical command cluster, including:

[0023] Based on the test processing results and the marked correct processing results of the historical task commands, determine the actual proportion of historical task commands that are currently executing correctly in the historical command cluster, as well as at least one candidate historical task command that is currently executing incorrectly.

[0024] If the actual proportion is less than a set threshold or the number of iterations is less than a set number of iterations, based on the at least one candidate historical task command and the test processing results and labeled correct processing results of the candidate historical task command, the candidate guidance information is optimized using the model, and the number of iterations is incremented by one.

[0025] Based on the optimized candidate guidance information, the operation of processing each historical task command in the historical command cluster using the model is returned until the actual proportion is not less than the set threshold or the number of iterations is not less than the set number.

[0026] In another possible implementation, the set of index tags for the guidance information includes at least one index tag selected from the candidate index tags of each historical task command in the historical command cluster, which meets the requirements for recall and precision.

[0027] The candidate index tags of the historical task command are used to characterize the summary information corresponding to the historical task indicated by the historical task command;

[0028] The recall rate is related to the first number of times that the candidate index tags of each historical task command in the historical command cluster hit the set of index tags.

[0029] The accuracy is related to the first hit count and the second hit count. The second hit count is the number of candidate index tags of each reference historical task command in at least one reference historical command cluster that hit the set of index tags. The reference historical command cluster and the historical command cluster correspond to the same intent category.

[0030] In another possible implementation, the set of index tags for the guidance information is obtained as follows:

[0031] Determine the candidate index tags of each historical task command in the historical command cluster and the candidate index tags of each reference historical task command in the reference historical command cluster.

[0032] Each candidate index label corresponding to the historical command cluster is determined as an index label in the index label set;

[0033] Based on the candidate index tags of each historical task command in the historical command cluster and the set of index tags, determine the first number of times the candidate index tags of each historical task command in the historical command cluster hit the set of index tags.

[0034] Based on the candidate index tags of each reference task command in the reference history command cluster and the set of index tags, determine the second number of times the candidate index tags of each reference history task command in the reference history command cluster hit the set of index tags.

[0035] If the recall and precision rates determined based on the first and second hit counts do not meet the set requirements, the index tags in the index tag set are optimized based on the candidate index tags in the reference historical command cluster that hit the reference historical task command of the index tag set until the recall and precision rates meet the set requirements.

[0036] In another possible implementation, the optimization of the index tags in the index tag set based on the reference historical task command that matches the index tag set in the corresponding candidate index tag in the reference historical command cluster includes:

[0037] Based on the reference historical task command that matches the index tag set in the reference historical command cluster with the corresponding candidate index tag, the index tags in the index tag set are corrected to obtain the corrected index tag set.

[0038] Based on the corrected set of index tags, the missed historical task commands in the historical command cluster are determined. The missed historical task commands are historical task commands whose candidate index tags cannot match the corrected set of index tags.

[0039] The candidate index tags corresponding to the missed detection historical task commands are updated, and the updated candidate index tags are added to the corrected index tag set.

[0040] In another aspect, this application also provides an electronic device, including: an input unit and a processor;

[0041] The input unit is used to obtain task commands;

[0042] The processor is configured to, in response to receiving a task command, determine the target task characteristics of the target task indicated by the task command; based on the target task characteristics, determine target guidance information from at least one guidance information, the guidance information being generated based on a historical command cluster, the historical command cluster including at least one historical task command belonging to the same intent category; the historical command cluster corresponding to the target guidance information includes: historical task commands whose corresponding task characteristics match the target task characteristics; and process the target task indicated by the task command based on the task command and the target guidance information. Attached Figure Description

[0043] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0044] Figure 1 A flowchart illustrating the task processing method provided in this application;

[0045] Figure 2 Another flowchart illustrating the task processing method provided in this application;

[0046] Figure 3 This is a schematic diagram illustrating one implementation process for constructing a historical command cluster in this application;

[0047] Figure 4 This is a schematic diagram illustrating one implementation process of generating boot information based on historical command clusters in this application;

[0048] Figure 5 This is a schematic diagram illustrating another implementation process for generating boot information for historical command clusters in this application;

[0049] Figure 6 A schematic diagram illustrating one implementation process for generating an index tag set for the boot information of this historical command cluster in this application;

[0050] Figure 7 A schematic diagram of the component architecture of the electronic device provided in this application. Detailed Implementation

[0051] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is only for explaining specific embodiments and is not intended to limit the application. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0052] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0053] like Figure 1 The diagram illustrates an implementation flow of the task processing method provided in this application. The method of this embodiment can be applied to electronic devices, such as terminal devices, hosts, servers, or device nodes in cloud systems.

[0054] The method in this embodiment may include:

[0055] S101, in response to receiving a task command, determine the target task characteristics of the target task indicated by the task command.

[0056] The task command is used to instruct the target task to be executed. The task command can be in the form of text or voice, without any restrictions.

[0057] Among them, the target task features are the characteristics of the target task indicated by the task command. For example, the target task features can be the encoding features of the target task, the content features of the target task, or the semantic features of the target task.

[0058] S102, Based on the characteristics of the target task, determine the target guidance information from at least one guidance information.

[0059] In this application, the guidance information is generated based on historical command clusters and is used to guide task processing. Different guidance information is generated based on different historical command clusters.

[0060] Depending on the task processing scenario, the guiding role and specific content of the guidance information will also vary.

[0061] For example, guidance information can be strategy information used to avoid task execution errors. For instance, guidance information can suggest tools or models to avoid when handling the task, and recommend at least one tool or model to use.

[0062] For example, guidance information can be processing strategy information used to guide task execution, ensuring the task is processed reasonably or correctly. This could include instructions on the tools or programs to be invoked to process the task, or prompts and optimization strategies applicable to the task.

[0063] The historical command cluster includes at least one historical task command belonging to the same intent category. A historical task command is a task command obtained before the current moment (i.e., in the past). For example, a historical command cluster includes at least one historical task command belonging to the same intent category and having similar semantic features.

[0064] Understandably, if the guidance information is determined based on only a single historical task command, the guidance information may be significantly biased and unable to effectively guide the intelligent program to correctly process the historical task command. Moreover, the guidance information determined based on a single historical task command is only applicable to the case of processing the historical task command again, and its scope of application is relatively small.

[0065] In this application, guidance information is generated based on historical task commands that belong to the same intent category and belong to the same historical command cluster. This allows for a more comprehensive and accurate processing of guidance information suitable for tasks related to the historical command cluster, by combining at least one historical task command. It also improves the applicability of the guidance information.

[0066] In this application, the target guidance information is the guidance information among the at least one guidance information that matches the target task feature. Based on this, the historical command cluster corresponding to the target guidance information includes: historical task commands whose corresponding task features match the target task feature.

[0067] S103 processes the target task indicated by the task command based on the task command and target guidance information.

[0068] Depending on the task processing scenario, the specific implementation of task commands based on target-guided information processing will also vary.

[0069] For example, in electronic devices where intelligent programs such as intelligent agents can be deployed, the electronic device needs to determine the model or other program tools to process the task command based on the intelligent program. If the electronic device determines the wrong model or program tool to be called, it may lead to errors in the execution of the task command. Based on this, this application can determine the target model or target program tool to process the task command based on the target guidance information, and use the target model or target program tool to process the target task indicated by the task command.

[0070] For example, after the electronic device determines the model to process the task command, if the meaning of the target task indicated by the task command is unclear or the processing capability of the model is limited, prompt words can be generated based on the target guidance information and the task command. Based on the prompt words, the model can be used to process the target task to improve the success rate of the model in processing the target task.

[0071] As can be seen from the above, after obtaining a task command, this application can determine target guidance information from at least one guidance information based on the target task characteristics indicated by the task command. Since the target guidance information is generated based on a historical command cluster consisting of at least one historical task command belonging to the same intent category, the generated target guidance information is more accurate and comprehensive. It is applicable not only to guiding the execution of each historical task command in the historical command cluster but also to other task commands that match the task characteristics of the historical task commands in the historical command cluster. Therefore, combining this target guidance information to process the target task indicated by the task command can improve the reliability and accuracy of target task execution and reduce task execution errors.

[0072] In this application, there are many possible implementations of the target task characteristics of the target task indicated by the task command, and no specific restrictions are imposed.

[0073] In one possible implementation, to ensure that the task features accurately represent the target task and facilitate matching with guidance information, this application can determine the task features based on the summary information of the task indicated by the task command. Based on this, determining the target task features can be achieved by: determining the target index label corresponding to the task command, whereby the target index label is used to represent the summary information of the target task indicated by the task command.

[0074] There are several possibilities for the specific implementation of determining the target index label corresponding to the task command.

[0075] For example, a summary extraction model can be used to extract summary information of the task command, and the extracted summary information can be determined as the target index label corresponding to the target task indicated by the task command.

[0076] For example, multiple tag extraction templates can be predefined. Based on these predefined templates, tags can be extracted from task commands to obtain the target index tags. For instance, assuming the task command is "Recommend a certain laptop, with a budget of 4000-6000", based on at least one predefined tag extraction template, the target index tags for this task command can be extracted as "Recommend [product], [demand]".

[0077] Accordingly, when determining the target guidance information, the target guidance information can be determined from the at least one guidance information based on the index tag set of each of the at least one guidance information and the target index tag corresponding to the task command.

[0078] The set of index tags for the guidance information includes at least one index tag.

[0079] The set of index tags for the target guidance information includes index tags that match the target index tag.

[0080] The index tags in the guidance information's index tag set can be generated based on the index tags of each historical task command in the historical command cluster corresponding to the guidance information. For example, the guidance information's index tag set may include the index tags of at least some of the historical task commands in the historical command cluster corresponding to the guidance information. The index tags of the historical task commands are used to characterize the summary information of the historical task indicated by the historical task command.

[0081] Therefore, since an index tag in the index tag set can represent the summary information of a historical task command in the historical command cluster corresponding to the target guidance information, if there is an index tag in the index tag set of the target guidance information that matches the target index tag, it means that there is a historical task command similar to the task command in the historical command cluster corresponding to the target guidance information. Based on this, since the target guidance information is generated based on the historical command cluster and can guide the execution of historical task commands in the historical command cluster, the target guidance information is naturally also applicable to guiding the execution of the task command, provided that there is a historical task command similar to the task command in the historical command cluster corresponding to the target guidance information.

[0082] Furthermore, in practical applications, the number of historical task commands is large, and the intent categories to which they can be classified are also numerous. Moreover, historical task commands within the same intent category may be categorized into different historical command clusters. Therefore, in order to efficiently and accurately determine the appropriate target guidance information for the task command, this application can determine the target intent category corresponding to the task command before determining the target guidance information. Then, the target guidance information is determined from at least one guidance information corresponding to the target intent category.

[0083] The following is combined Figure 2 Please provide an explanation. For example... Figure 2 This illustrates another flowchart of the task processing method provided in this application. The method in this embodiment includes:

[0084] S201, in response to receiving a task command, determine the target intent category corresponding to the command intent of the task command.

[0085] There are no restrictions on the specific implementation of the target intent category corresponding to the task command.

[0086] For example, by performing intent recognition on task commands, the target intent category to which the command intent belongs can be determined.

[0087] S202, determine the guidance information set corresponding to the target intent category.

[0088] The guidance information set includes: at least one guidance information and a set of index tags corresponding to that guidance information.

[0089] The guidance information set corresponding to the target intent category includes: at least one guidance information generated based on at least one historical command cluster under the target intent category, wherein one guidance information is generated based on one historical command cluster, and different guidance information in the guidance information set corresponds to different historical command clusters.

[0090] Within the guidance information set, each guidance information's corresponding historical command cluster includes at least one historical task command belonging to that target intent category. For example, the guidance information's corresponding historical command cluster includes at least one historical task command belonging to that target intent category and having similar semantics.

[0091] S203, Determine the target index label corresponding to this task command.

[0092] The target index label is used to characterize the summary information of the target task indicated by the task command.

[0093] The order in which step S203 and step S201 determine the target intent category can be interchanged or performed simultaneously.

[0094] S204, Based on the set of index tags for each of at least one set of guidance information in the guidance information set and the target index tag, determine the target guidance information from at least one set of guidance information.

[0095] The set of index tags for the guidance information includes at least one index tag.

[0096] In this application, the set of index tags for target guidance information includes index tags that match the target index tag.

[0097] S205, Based on the task command and the target guidance information, process the target task indicated by the task command.

[0098] The steps S203 to S205 above can be found in the relevant descriptions of the previous embodiments, and will not be repeated here.

[0099] In this embodiment, considering that only historical task commands belonging to the same intent category are likely to be applicable to the same guidance information, this application first determines the target intent category to which the task command belongs, and then determines the guidance information set of the target intent category. This naturally allows for the filtering of guidance information sets that may be applicable to guiding the task command. Based on this, this application only needs to determine the target guidance information matching the target index tag of the task command from at least one guidance information set of the target intent category, which can reduce the amount of data for guidance information matching and improve matching efficiency and accuracy.

[0100] In any of the above embodiments of this application, in order to guide the processing of task commands as accurately as possible and reduce execution errors of the target tasks indicated by the task commands, the guidance information in this application can be strategy information for avoiding task execution errors. In this case, the historical task commands in the historical command cluster corresponding to the guidance information include at least one first historical task command that has an execution error, so as to determine the guidance information required to avoid the execution errors of each first historical task command based on each first historical task command that has an execution error. Based on this, the guidance information can also be called reflection information, that is, reflection information generated based on each first historical task command that has an execution error, used to avoid task command execution errors.

[0101] Furthermore, considering that generating guidance information based solely on historical task commands that have been executed incorrectly may result in guidance deviations, which could cause task commands that should have been executed correctly to be executed incorrectly due to inappropriate guidance information, in order to generate guidance information more accurately based on the historical command cluster, in this application, the historical command cluster further includes at least one second historical task command that is semantically similar to the first historical task command and is executed correctly.

[0102] In particular, since one historical command cluster corresponds to one boot information, in order for the boot information to be effectively used to avoid execution errors of each first task command in the historical command cluster, in this application, at least one first historical task command in the historical command cluster that has an execution error has the same error type, that is, the historical command cluster includes at least one first task command that has an execution error and the same error type.

[0103] It is understood that this application requires first identifying at least one historical command cluster under different intent categories, and then generating guidance information for each historical command cluster. The specific implementation for identifying the historical command cluster can have multiple possibilities, and no restriction is imposed.

[0104] The following example demonstrates one approach to constructing a historical command cluster, combined with... Figure 3 Please provide an explanation. For example... Figure 3This illustration shows a schematic diagram of one implementation process for constructing a historical command cluster in this application. This embodiment may include:

[0105] S301, Based on the intent categories of each historical task command in the dataset, determine at least one set of commands corresponding to each intent category.

[0106] The dataset may include the collected historical task commands, as well as the actual processing results and the labeled correct processing results for each historical task command.

[0107] Specifically, if the actual processing result of a historical task command is inconsistent with the marked correct processing result, then the historical task command is considered an incorrectly executed historical task command, i.e., the first historical task command mentioned earlier. Conversely, if the actual processing result of a historical task command is consistent with the marked correct processing result, then the historical task command is considered correctly executed (or executed correctly), i.e., the second historical task command mentioned earlier.

[0108] The command set corresponding to the intent category includes at least one historical task command belonging to that intent category.

[0109] S302, for each intent category, based on the error type of at least a portion of the first historical task commands that err in the command set of the intent category, determine at least one negative sample cluster under that intent category.

[0110] There are various ways to determine the error type of the first historical task command that was executed incorrectly, and this application does not impose any restrictions on this. For example, the first historical task command that was executed incorrectly can be input into a pre-trained large model or other error recognition model to determine the error type of the first task command.

[0111] Each negative sample cluster includes at least one first historical task command with the same error type.

[0112] In this application, there are multiple ways to determine the negative sample cluster.

[0113] For example, the error types of all first historical task commands can be identified separately, and the first task category with the same error type can be identified as a negative sample cluster.

[0114] For example, each first historical task command can be input into a large model or other model that can be used for error classification. Based on this, the model can be used to determine the negative sample clusters composed of first historical task commands belonging to the same error type, thus obtaining at least one negative sample cluster.

[0115] In particular, the above two methods require a large amount of data for analyzing and processing the first historical task commands. In order to reduce the amount of data processing, this application can also randomly select a specified number of first historical task commands from the command set of the intent category, and use a large model or other models that can be used for error classification to identify the target error type that contains the most first historical task commands among the specified number of first historical task commands. The first historical task commands that belong to the target error type among the specified number of first historical task commands are then grouped into a negative sample cluster to obtain a negative sample cluster.

[0116] Furthermore, after identifying the negative sample cluster, in order to avoid having too few first historical task commands in the negative sample cluster, which would result in a smaller range of applicable error types for the first historical task commands in the negative sample cluster, and consequently a smaller range of applicable guidance information generated from the historical command cluster constructed based on the negative sample cluster, and a lower probability of being matched by subsequent task commands, the negative sample cluster can be discarded if the number of first historical task commands in the negative sample cluster is lower than a set threshold.

[0117] S303, for each negative sample cluster, at least one candidate historical task command that is not yet classified in the command set of the intent category and is semantically similar to the first historical task command in the negative sample cluster is classified into the negative sample cluster, thus obtaining the classified historical command cluster.

[0118] It is understood that there may be one or more negative sample clusters generated in step S302. The negative sample cluster may only include a portion of the first historical task commands in the command set of the intent category. Therefore, the historical task commands classified into the negative sample cluster in step S303 may include either the first historical task command that was executed incorrectly or the second historical task command that was executed correctly.

[0119] Understandably, after executing step S303, if the number of unclassified historical task commands in the command set of the intent category exceeds the set threshold, it is still possible to return to step S302 to continue building the historical command cluster under that intent category until the number of iterations reaches the set number or the number of unclassified historical task commands in the command set of the intent category does not exceed the set threshold.

[0120] In any of the above embodiments, when the guidance information is strategy information for avoiding task execution errors, the historical command cluster on which the guidance information is generated may include: at least one first historical task command that executed incorrectly, and at least one second historical task command that is semantically similar to the first historical task command and executed correctly. In this case, there may be multiple possibilities for generating guidance information based on each historical command cluster.

[0121] For example, in one possible implementation, in order to ensure the accuracy of the guidance information generated based on the historical command cluster, and to ensure that the guidance information can be reliably used to guide the historical command cluster and the task commands similar to the historical task commands in the historical command cluster, this application can also determine the error description information corresponding to the historical command cluster, so that the historical command cluster is associated with the error description information.

[0122] The error description information is generated based on at least a portion of the first historical task commands in the historical command cluster, and is used to describe the error cause information of the first historical task commands in the historical command cluster.

[0123] Correspondingly, the guidance information is generated based on the error description information corresponding to the historical command cluster, as well as each historical task command in the historical command cluster, the actual processing result of the historical task command, and the labeled correct processing result of the historical task command. Among them, the labeled correct processing result of the historical task command can be manually labeled or determined by the model, and is the execution result corresponding to the correct execution of the historical task command.

[0124] The following example illustrates the process by which this application generates boot information based on historical command clusters. Figure 4 This illustration shows a schematic diagram of an implementation process for generating boot information based on historical command clusters in this application. The method in this example may include:

[0125] S401, based on the error description information corresponding to the historical command cluster, as well as at least some historical task commands in the historical command cluster, the actual processing results of the at least some historical task commands, and the labeled correct processing results, candidate guidance information is determined using a model.

[0126] As mentioned earlier, the error description information corresponding to the historical command cluster can be generated based on at least a portion of the first historical task commands in that historical command cluster. This error description information is used to describe the error cause information of the first historical task commands in the historical command cluster. For example, the error description information can describe the problem that occurred in the first historical task command in the historical command cluster and the cause of the problem.

[0127] In particular, each first historical task command in a historical command cluster can belong to the same error type, and the error description information can describe the reason why the historical command cluster corresponds to the error type.

[0128] The error description information corresponding to the historical command cluster can be determined by using a large model or other models that can be used to achieve error classification, based on at least some of the first historical task commands in the historical command cluster.

[0129] In particular, if during the construction of a historical command cluster, a specified number of first historical task commands are extracted, and the target error type with the largest number of first historical task commands is determined using a large model or other models, the large model or other models can also synchronously output the error description information of the target error type. In this case, it is not necessary to generate the error description information corresponding to the historical command cluster separately.

[0130] The candidate guidance information is strategy information for avoiding errors in the execution of historical task commands. For example, guidance information is generated using a trained reflective generative model based on at least a portion of the historical task commands in the historical command cluster, and this guidance information is used as candidate guidance information for that historical command cluster. The number of first and second historical task commands in the at least a portion of the historical task commands should be as equal as possible to improve the accuracy and applicability of the generated candidate guidance information.

[0131] In addition, in order to reduce the amount of data processing and improve the efficiency of generating candidate boot information, candidate boot information can be generated by sampling only a portion of historical task commands from the historical command cluster.

[0132] The model that generates the candidate guidance information can be selected according to actual needs. For example, the model can be a trained large language model or other models that can process and optimize text, without any restrictions.

[0133] S402, based on candidate boot information, processes each historical task command in the historical command cluster to obtain the test processing results of the historical task commands.

[0134] This step is actually based on the candidate guidance information to backtest each historical task command in the historical command cluster using the model, so as to retest the processing results obtained by reprocessing the historical task commands using the model with the candidate guidance information.

[0135] In order to distinguish it from the actual processing results of historical task commands, the processing results obtained by using the model to process historical task commands based on candidate guidance information are called test processing results.

[0136] For example, suppose the original method for processing historical task commands in a historical command cluster was to first call a search tool to perform a search, and then call model A to process the historical task commands based on the search results. However, the generated candidate guidance information indicates that model B is suitable to be called after the search is completed. Therefore, when processing historical task commands in a historical command cluster based on this candidate guidance information, model B can be called to process each historical task command based on the search results, and the test processing results can be obtained.

[0137] S403, based on the test processing results and the marked correct processing results of each historical task command in the historical command cluster, optimize the candidate guidance information and determine the optimized candidate guidance information as the guidance information corresponding to the historical command cluster.

[0138] Understandably, by combining the test processing results and the marked correct processing results of each historical task command in the historical command cluster, it is possible to determine whether each historical task command in the historical command cluster is correctly executed when the candidate guidance information is used to process the historical task commands. This allows us to determine whether the candidate guidance information can accurately guide the correct execution of the historical task commands, and naturally reflects the accuracy of the candidate guidance information. Therefore, when the accuracy of the candidate guidance information is low, we can optimize the candidate guidance information.

[0139] For example, in one possible implementation, the actual percentage of currently correctly executed historical task commands in the historical command cluster and at least one candidate historical task command currently executing incorrectly can be determined based on the test processing results and the labeled correct processing results of historical task commands. For ease of distinction, the historical task commands that execute incorrectly in the historical command cluster during backtesting are referred to as candidate historical task commands.

[0140] Based on this, if the actual percentage of correctly executed historical task commands is less than a set threshold, or the number of iterations is less than a set number, the candidate guidance information can be optimized using the model based on at least one candidate historical task command, the test processing results of the candidate historical task command, and the labeled correct processing results, and the number of iterations is incremented by one. Then, based on the optimized candidate guidance information, the execution step S402 is returned to re-execute the actual percentage of currently correctly executed historical task commands and the candidate historical task commands that are still incorrect, and this process is repeated until the actual percentage is not less than the set threshold or the number of iterations is not less than the set number.

[0141] Specifically, when the actual proportion is not less than a set threshold or the number of iterations is not less than a set number, the candidate guidance information with the highest actual proportion can be determined as the final optimized candidate guidance information, i.e., the optimal candidate guidance information, and the optimal candidate guidance information can be determined as the guidance information corresponding to the historical command cluster.

[0142] It is understandable that after reprocessing each historical task command in the historical command cluster based on the candidate guidance information, the candidate historical task commands that still execute incorrectly in the historical command cluster can effectively characterize the deficiencies or errors in the candidate guidance information. Therefore, the test processing results based on the candidate historical task commands and the labeled correct execution results can provide a more reliable basis for optimizing the candidate guidance information, thereby improving the accuracy of the optimized candidate guidance information.

[0143] Furthermore, even after determining the guidance information for the historical command cluster by controlling the candidate guidance information based on the actual proportion and the number of iterations, some historical task commands within the historical command cluster may still fail to execute when processed based on this guidance information. Therefore, to ensure that the guidance information is universally applicable to all historical task commands within the historical command cluster, and applicable to task commands similar to those in the historical command cluster during the inference phase, this application can also identify target historical task commands in the historical command cluster that are still executing incorrectly based on the guidance information, and delete these target historical task commands from the historical command cluster, resulting in an optimized historical command cluster.

[0144] To facilitate understanding, the following will be combined with... Figure 5 Please provide an explanation, such as Figure 5 This illustration shows another implementation flow diagram of generating boot information for historical command clusters according to this application. The method in this embodiment may include:

[0145] S501, based on the error description information corresponding to the historical command cluster, as well as at least some historical task commands in the historical command cluster, the actual processing results of the at least some historical task commands, and the labeled correct processing results, candidate guidance information is determined using a model.

[0146] This step involves the generation of initial boot information.

[0147] S502, based on candidate boot information, processes each historical task command in the historical command cluster to obtain the test processing results of the historical task commands.

[0148] This step involves backtesting and evaluating historical task commands in the historical command cluster based on candidate guidance information.

[0149] S503, based on the test processing results of historical task commands and the labeled correct processing results, determine the actual proportion of historical task commands currently executing correctly in the historical command cluster, and at least one candidate historical task command currently executing incorrectly.

[0150] Among them, a candidate historical task command that executes incorrectly refers to a historical task command whose test processing result obtained when processing the historical task command based on the candidate guidance information is different from the labeled correct processing result.

[0151] S504, check whether the actual proportion is less than the set threshold or the number of iterations is less than the set number. If yes, proceed to step S505; otherwise, proceed to step S506.

[0152] The actual percentage is the execution success rate of each historical task command in the historical command cluster. Based on this, if the actual percentage is less than the set threshold, it means that the execution success rate is less than the set threshold.

[0153] S505: Based on at least one candidate historical task command, the test processing results of the candidate historical task command, and the labeled correct processing results, optimize the candidate guidance information using the model, increment the iteration number by one, and return to execution step S502 based on the optimized candidate guidance information.

[0154] The initial value of the iteration count can be 0. The iteration count is incremented by one each time step S505 is executed.

[0155] Depend on Figure 5 It can be seen that the above steps S502 to S505 belong to the process of guiding information loop optimization.

[0156] S506, determine the target candidate guidance information that has the largest actual proportion, and determine the target candidate guidance information as the guidance information corresponding to the historical command cluster.

[0157] Understandably, if the actual proportion is not less than the set threshold, then the currently determined candidate guidance information is the target candidate guidance information with the largest actual proportion (that is, the one that maximizes the execution success rate). If the current iteration count has reached the set number, then the target candidate guidance information that maximizes the actual proportion (i.e., the one that maximizes the execution success rate) can be determined during the candidate guidance information iterative optimization process.

[0158] S507, at least one candidate historical task command that was determined to have an execution error based on the target candidate guidance information is deleted from the historical command cluster and added to the command set corresponding to the intent category to which the historical command cluster belongs.

[0159] Among them, the at least one candidate historical task command that is determined to be an execution error based on the target candidate guidance information is the at least one candidate historical task command that is determined to be an execution error when processing each historical task command in the historical command cluster based on the target guidance information.

[0160] It is understandable that, since the target candidate guidance information is ultimately determined to be the guidance information of the historical command cluster, when backtesting is performed based on the target candidate guidance information, the historical task commands that are determined to be executed incorrectly are not actually suitable for task processing using the guidance information. Therefore, they need to be deleted from the historical command cluster and added back to the command set corresponding to the intent category to which the historical command cluster belongs, so that they can be subsequently classified into the appropriate historical command cluster.

[0161] S508, if the number of historical task commands in the historical command cluster is greater than the set number, then store the historical command cluster and the corresponding boot information.

[0162] One optional step is to determine whether the number of historical task commands in the historical command cluster is greater than a set number. The purpose is to avoid the number of historical task commands in the historical command cluster being too small, which would make the applicable scope of the guidance information corresponding to the historical task command cluster too narrow and unable to be effectively utilized in the subsequent online task processing stage.

[0163] In any of the above embodiments of this application, after determining the historical command cluster and the guidance information corresponding to the historical command cluster, in order to determine the appropriate guidance information for the task command based on the obtained task command in the subsequent online task processing stage, this application also needs to generate an index tag set for the guidance information corresponding to the historical command cluster based on the historical task commands that the historical command cluster has.

[0164] As mentioned earlier, candidate index tags from some or all of the historical task commands in the historical command cluster can be used as index tags in the set of index tags corresponding to the guidance information of that historical command cluster. However, this approach may result in some index tags being unsuitable, causing task commands similar to historical task commands in the historical command cluster to fail to reliably match the index tags, thus preventing the matching of the guidance information.

[0165] Based on this, in order to more reasonably determine the index tags in the index tag set of guidance information and improve the accuracy of locating guidance information based on index tags, in one possible implementation, the index tag set of guidance information includes: at least one index tag selected from the candidate index tags of each historical task command in the historical command cluster, which meets the requirements of recall and precision.

[0166] The candidate index tags for historical task commands are used to represent the summary information corresponding to the historical task indicated by the command. For ease of distinction, the index tags of historical task commands are referred to as candidate index tags. The process of generating candidate index tags for historical task commands is similar to that of generating target index tags for task commands, and will not be elaborated further.

[0167] The recall rate is related to the first number of hits of the candidate index tags of each historical task command in the historical command cluster that match the set of index tags of the historical command cluster.

[0168] The accuracy is related to the first hit count and the second hit count. The second hit count is the number of times a candidate index tag of each reference historical task command in at least one reference historical command cluster hits the index tag set. The reference historical command cluster corresponds to the same intent category as the historical command cluster. For example, if the intent category to which the historical command cluster belongs corresponds to a command set, the reference historical command cluster can include historical task commands other than the historical command cluster itself from the command set corresponding to that intent category.

[0169] To distinguish them from historical task commands within the same historical command family, historical task commands within the same historical command family are referred to as reference historical task commands. Based on this, the candidate index tags of reference historical task commands are also used to characterize the summary information of the historical commands they represent.

[0170] In this context, "the candidate index tag of a historical task command (or a reference to a historical task command) hitting the set of index tags" means that there exists an index tag in the set of index tags that matches the candidate index tag. In other words, the candidate index tag based on the historical task command can be matched with an index tag in the set of index tags.

[0171] Furthermore, to make the applicability of index tags in the index tag set more reasonable, this application can continuously optimize some candidate index tags during the process of selecting index tags from the candidate index tags of each historical task command in the historical command cluster. The following is combined with... Figure 6 Please provide an explanation. For example... Figure 6 This illustration shows a schematic diagram of an implementation process for generating an index tag set for boot information of historical command clusters according to this application. The method of this embodiment may include:

[0172] S601, determine the candidate index labels of each historical task command in the historical command cluster and the candidate index labels of each reference historical task command in the reference historical command cluster.

[0173] S602, determine each candidate index label corresponding to the historical command cluster as the index label in the index label set.

[0174] In this embodiment, the candidate index tags corresponding to each historical task command in the historical command cluster are all used as index tags in the index tag set to obtain the initial index tag set.

[0175] S603, based on the candidate index tags of each historical task command in the historical command cluster and the set of index tags, determine the first number of times the candidate index tags of each historical task command in the historical command cluster hit the set of index tags.

[0176] Understandably, before the index tag set is optimized, since each candidate index tag in the historical command cluster is used as an index tag in the index tag set, the candidate index tags of each historical task command in the historical command cluster can be matched with the index tags in the index tag set. That is, the candidate index tags of all historical task commands in the historical command cluster can hit the index tag set.

[0177] However, during the iterative optimization of the index tag set, there may be situations where candidate index tags for some historical task commands in the historical command cluster cannot be matched with the index tag set.

[0178] S604, based on the candidate index tags of each reference task command in the reference history command cluster and the set of index tags, determine the second number of times the candidate index tags of each reference history task command in the reference history command cluster hit the set of index tags.

[0179] S605, based on the first hit count and the second hit count, determine whether the recall rate and precision rate meet the set requirements. If not, proceed to step S606; if yes, proceed to step S607.

[0180] The recall rate is the ratio of the number of first hits of historical task commands in the historical command cluster that match the corresponding candidate index tag to the total number of historical task commands in the historical command cluster.

[0181] Precision is the ratio of the first hit count to the sum of the first and second hit counts.

[0182] For example, recall rate This can be expressed as Formula 1 below:

[0183] (Formula 1):

[0184] accuracy This can be expressed as Formula Two:

[0185] (Formula 2);

[0186] TP represents the first number of true positives. A true positive refers to a historical task command that belongs to the historical command cluster C_target and whose corresponding candidate index tag successfully hits the historical task command of the index tag set R_best. Therefore, the first number of true positives is also the first number of hits.

[0187] FN represents the second number of false negatives, where a false negative refers to a historical task command belonging to C_target whose corresponding candidate index label failed to match R_best. The sum of the first and second numbers represents the total number of historical task commands in that historical command cluster.

[0188] FP is the third number of false positives. A false positive refers to a command set that belongs to the intent category of the historical command cluster C_target but does not belong to the historical command cluster C_target. The corresponding candidate index label is incorrectly matched with the reference candidate task command of the index label set R_best. It can be seen that this third number is the second number of hits mentioned above.

[0189] The required thresholds for recall and precision can be set according to actual needs. For example, recall may be greater than a first threshold and precision may be greater than a second threshold; or both recall and precision may be greater than the set thresholds and converge.

[0190] In one possible implementation, in order to more reliably analyze recall and precision and improve the effectiveness of each index tag in the optimized index tag set, in this application, the recall and precision can meet the set requirements as follows: the growth value of the harmonic mean determined based on recall and precision converges.

[0191] The harmonic mean, also known as the F1 score or simply F1, comprehensively considers both the accuracy and completeness of the match when indexing based on the set of index tags. For example, F1 can be expressed as Formula 3 below:

[0192] (Formula 3);

[0193] As mentioned above, For recall rate, For accuracy.

[0194] The convergence of the growth value of the harmonic mean (F1) means that the value of F1 no longer increases and tends to stabilize.

[0195] S606, based on the candidate index tags in the reference history command cluster that match the reference history task command of the index tag set, optimize the index tags in the index tag set, and return to step S603 based on the optimized index tag set.

[0196] Understandably, the candidate index tags in the reference historical command cluster that match the reference historical task commands in that index tag set are the aforementioned false positives. The existence of such false positives indicates that the coverage of the index tags in that index tag set is too broad. This allows other historical task commands outside the historical command cluster and dissimilar to those in the cluster to also match the index tag set and be considered applicable to the corresponding boot information, thus causing errors. Therefore, it is necessary to optimize the index tags in the index tag set based on the false positives, so that these reference historical task commands can no longer match the index tag set.

[0197] There are various possible implementations for optimizing the index tags in the index tag set, and this application does not impose any restrictions on them.

[0198] In one possible implementation, the index label to be corrected can be determined from the set of index labels, matching the candidate index label corresponding to the referenced historical command cluster, and then corrected. For example, the correction of the index label to be corrected can be performed using either of the following two methods:

[0199] One correction method involves using a correction model to analyze the differences between the index label to be corrected and the correct index labels in the index label set, and then adjusting the index label to be corrected based on these differences. Here, the correct index labels are those in the index label set that do not belong to the index label to be corrected. The correction model can be a large language model or other trained models; there are no restrictions on which one is chosen.

[0200] Another correction method is to remove the index label to be corrected from the index label set.

[0201] Furthermore, considering that after correcting the index tags in the index tag set, the candidate indexes corresponding to historical task commands in the historical command cluster may not be able to match the corrected index tag set, resulting in missed detections, the optimization of index tags in step S606 can also be performed as follows:

[0202] First, based on the reference historical task commands that match the candidate index tags in the reference historical command cluster, the index tags in the index tag set are corrected to obtain the corrected index tag set.

[0203] The correction of index tags in the index tag set can be done as described above, such as determining the index tag to be corrected that matches the candidate index tag corresponding to the referenced historical command cluster from the index tag set, and then correcting the index tag to be corrected. The correction method can be any of the two methods described above.

[0204] Secondly, based on the revised set of index tags, the missed historical task commands in the historical command cluster are identified. Missed historical task commands are those whose candidate index tags cannot be matched within the revised set of index tags.

[0205] Finally, the candidate index tags corresponding to the missed historical task commands are updated, and the updated candidate index tags are added to the corrected index tag set.

[0206] There are several ways to update the candidate index tags corresponding to the missed historical task commands. For example, you can regenerate the candidate index tags for the missed historical task commands; or you can add content to the candidate index tags of the missed historical task commands based on the missed historical task commands, so that the information of the updated candidate index tags can more accurately express the summary information of the missed historical task commands.

[0207] Furthermore, to avoid a decrease in the F1 score after modifying the index label set, in this application, after modifying the index labels in the index label set based on the candidate index labels corresponding to the reference historical command cluster that match the reference historical task command of the index label set, the F1 score can be recalculated based on the modified index label set (e.g., by executing steps S603 and S604). If the F1 score increases, then based on the modified index label set, the missed historical task commands are identified, and the candidate index labels corresponding to the missed historical task commands are updated. Conversely, if the F1 score decreases, the index label set is restored to the set before the modification, and then the modification method is selected again to modify the index label set.

[0208] Similarly, after updating the candidate index label corresponding to a missed historical task command and adding it to the corrected index label set, F1 needs to be recalculated. If the F1 value increases, the process can return to step S603 based on the optimized index label set to determine whether the increase in F1 has converged. If the F1 value does not increase, the candidate index label corresponding to the missed historical task command can be updated again, or the updated candidate index label can be left unadded to the corrected index label set.

[0209] S607, the currently obtained set of index tags is determined as the set of index tags corresponding to the boot information of the historical command cluster.

[0210] It is understood that, after determining at least one historical command cluster corresponding to each of the different intent categories, the guidance information corresponding to the historical command cluster, and the set of index tags corresponding to the guidance information through the above embodiments of this application, after obtaining the task command, the target guidance information can be matched from at least one set of guidance information based on the target index tag of the task command. For details, please refer to the previous... Figure 1 and Figure 2 The relevant descriptions of the embodiments will not be repeated here.

[0211] In particular, in order to continuously update and optimize the guidance information, if the target guidance information cannot be matched based on the target index label of the task command, this application can also add the task command to the dataset. When the number of historical task commands not classified into the historical command cluster in the dataset reaches a set threshold, or when the guidance information update conditions are met, a historical command cluster is constructed based on the unclassified historical task commands, or the unclassified historical task commands are added to the existing historical command cluster, and guidance information is generated for the newly generated historical command cluster.

[0212] Furthermore, if the target intent category corresponding to the command intent of a task command is determined, but there is no set of guidance information corresponding to that target intent category, then the task command can also be added to the dataset so that when there are enough historical task commands belonging to that target intent category, a historical command cluster can be built for that target intent category, and guidance information can be generated for the historical command cluster. For details, please refer to the previous related introduction.

[0213] Furthermore, this application also provides an electronic device in its embodiments. For example... Figure 7 As shown, it illustrates a schematic diagram of the composition structure of the electronic device, which includes at least a processor 701 and an input unit 702;

[0214] The input unit 702 is used to obtain task commands. For example, the input unit can be a touch screen or an input device, etc., without any specific limitations.

[0215] The processor 701 is configured to, in response to receiving a task command, determine the target task characteristics of the target task indicated by the task command; based on the target task characteristics, determine target guidance information from at least one guidance information, the guidance information being generated based on a historical command cluster, the historical command cluster including at least one historical task command belonging to the same intent category; the historical command cluster corresponding to the target guidance information includes: historical task commands whose corresponding task characteristics match the target task characteristics; and process the target task indicated by the task command based on the task command and the target guidance information.

[0216] The operations performed by the input unit and the processor can be found in the relevant descriptions of the preceding method embodiments, and will not be repeated here.

[0217] Furthermore, the electronic device may also include a memory 703 for storing programs required by the processor to perform operations.

[0218] Of course, the electronic device may also include components such as a display unit 704, without any specific restrictions.

[0219] Understandably, this electronic device can also have more than Figure 7 There are no restrictions on the number of components, whether more or fewer.

[0220] This application also provides a computer program product, including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the task processing methods provided in this application.

[0221] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the task processing methods provided in this application.

[0222] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0223] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0224] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0225] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A task processing method, comprising: In response to receiving a task command, determine the target task characteristics of the target task indicated by the task command; Based on the target task characteristics, target guidance information is determined from at least one guidance information, wherein the guidance information is generated based on a historical command cluster, and the historical command cluster includes at least one historical task command belonging to the same intent category; The historical command cluster corresponding to the target guidance information includes: historical task commands whose corresponding task features match the target task features; Based on the task command and the target guidance information, process the target task indicated by the task command.

2. The task processing method according to claim 1, wherein determining the target task characteristics of the target task indicated by the task command includes: Determine the target index tag corresponding to the task command, wherein the target index tag is used to characterize the summary information of the target task indicated by the task command; The step of determining target guidance information from at least one guidance information based on the target task characteristics includes: Based on the index tag set of each of the at least one type of guidance information and the target index tag, a target guidance information is determined from the at least one type of guidance information, wherein the index tag set includes at least one index tag, and the index tag set of the target guidance information includes index tags that match the target index tag.

3. The task processing method according to claim 2 further includes: Determine the target intent category corresponding to the command intent of the task command; Determine the guidance information set corresponding to the target intent category, the guidance information set including: at least one guidance information and a set of index tags corresponding to the guidance information.

4. The task processing method according to any one of claims 1 to 3, wherein the guidance information is strategy information for avoiding task execution errors; The historical command cluster includes at least one first historical task command that was executed incorrectly, and at least one second historical task command that is semantically similar to the first historical task command and was executed correctly. The historical command cluster is associated with error description information, which is generated based on at least a portion of the first historical task commands in the historical command cluster and is used to describe the error cause information of the first historical task commands in the historical command cluster. The guidance information is generated based on the error description information corresponding to the historical command cluster, as well as each historical task command in the historical command cluster, the actual processing result of the historical task command, and the correct processing result of the historical task command marked.

5. The task processing method according to claim 4, wherein the guidance information is obtained in the following manner: Based on the error description information corresponding to the historical command cluster, as well as at least some historical task commands in the historical command cluster, the actual processing results of the at least some historical task commands, and the labeled correct processing results, candidate guidance information is determined using a model. The candidate guidance information is strategy information for avoiding the execution errors of the historical task commands. Based on the candidate guidance information, each historical task command in the historical command cluster is processed to obtain the test processing result of the historical task command; Based on the test processing results and the marked correct processing results of the historical task commands, the candidate guidance information is optimized, and the optimized candidate guidance information is determined as the guidance information corresponding to the historical command cluster.

6. The task processing method according to claim 5, wherein optimizing the candidate guidance information based on the test processing results and labeled correct processing results of the historical task commands, and determining the optimized candidate guidance information as the guidance information corresponding to the historical command cluster, includes: Based on the test processing results and the marked correct processing results of the historical task commands, determine the actual proportion of historical task commands that are currently executing correctly in the historical command cluster, as well as at least one candidate historical task command that is currently executing incorrectly. If the actual proportion is less than a set threshold or the number of iterations is less than a set number of iterations, based on the at least one candidate historical task command and the test processing results and labeled correct processing results of the candidate historical task command, the candidate guidance information is optimized using the model, and the number of iterations is incremented by one. Based on the optimized candidate guidance information, the operation of processing each historical task command in the historical command cluster using the model is returned until the actual proportion is not less than the set threshold or the number of iterations is not less than the set number.

7. The task processing method according to claim 2, wherein the set of index tags for the guidance information includes: From the candidate index tags of each historical task command in the historical command cluster, select at least one index tag that meets the requirements of recall and precision. The candidate index tags of the historical task command are used to characterize the summary information corresponding to the historical task indicated by the historical task command; The recall rate is related to the first number of times that the candidate index tags of each historical task command in the historical command cluster hit the set of index tags. The accuracy is related to the first hit count and the second hit count. The second hit count is the number of candidate index tags of each reference historical task command in at least one reference historical command cluster that hit the set of index tags. The reference historical command cluster and the historical command cluster correspond to the same intent category.

8. The task processing method according to claim 7, wherein the set of index tags for the guidance information is obtained in the following manner: Determine the candidate index tags of each historical task command in the historical command cluster and the candidate index tags of each reference historical task command in the reference historical command cluster. Each candidate index label corresponding to the historical command cluster is determined as an index label in the index label set; Based on the candidate index tags of each historical task command in the historical command cluster and the set of index tags, determine the first number of times the candidate index tags of each historical task command in the historical command cluster hit the set of index tags. Based on the candidate index tags of each reference task command in the reference history command cluster and the set of index tags, determine the second number of times the candidate index tags of each reference history task command in the reference history command cluster hit the set of index tags. If the recall and precision rates determined based on the first and second hit counts do not meet the set requirements, the index tags in the index tag set are optimized based on the candidate index tags in the reference historical command cluster that hit the reference historical task command of the index tag set until the recall and precision rates meet the set requirements.

9. The task processing method according to claim 8, wherein optimizing the index tags in the index tag set based on the reference historical task command that matches the index tag set in the corresponding candidate index tag in the reference historical command cluster includes: Based on the reference historical task command that matches the index tag set in the reference historical command cluster with the corresponding candidate index tag, the index tags in the index tag set are corrected to obtain the corrected index tag set. Based on the corrected set of index tags, the missed historical task commands in the historical command cluster are determined. The missed historical task commands are historical task commands whose candidate index tags cannot match the corrected set of index tags. The candidate index tags corresponding to the missed detection historical task commands are updated, and the updated candidate index tags are added to the corrected index tag set.

10. An electronic device, comprising: Input unit and processor; The input unit is used to obtain task commands; The processor is configured to, in response to receiving a task command, determine the target task characteristics of the target task indicated by the task command; Based on the target task characteristics, target guidance information is determined from at least one guidance information, wherein the guidance information is generated based on a historical command cluster, and the historical command cluster includes at least one historical task command belonging to the same intent category; The historical command cluster corresponding to the target guidance information includes: historical task commands whose corresponding task features match the target task features; Based on the task command and the target guidance information, process the target task indicated by the task command.

Images