Query task-oriented problem analysis method and apparatus, and computing device

Abstract

The embodiment of the invention provides a problem analysis method and device for a query task and computing device.The method comprises the steps that an intention analysis vector and a reference vector obtained by executing the query task are mapped into different mapping points in a first display area, the intention analysis vector is determined by performing intention analysis on query information corresponding to the query task, and the reference vector is determined by querying in alternative data based on the intention analysis vector; in response to an execution parameter adjustment instruction for the query task, adjusting an execution parameter; on the basis of the updated intention analysis vector and the reference vector, the mapping points in the first display area are updated, and the updated intention analysis vector and the reference vector are obtained by executing the query task on the basis of the adjusted execution parameters; and determining problem information existing in the execution process of the query task based on the updated distribution information of the mapping points in the first display area. According to the method, the problem analysis effect for the query task can be improved.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a problem analysis method for query tasks, as well as a problem analysis apparatus for query tasks, a computing device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] With the development of computer technology and artificial intelligence technology, large models are widely used, and many tasks can be completed with the help of large models. The requirements for task execution performance are also getting higher and higher.

[0003] Large models typically utilize Retrieval Augmented Generation (RAG) techniques to generate execution results for user-specified tasks, ensuring the accuracy of the generated results. For example, reference data can be obtained by pre-executing a query against a database based on user input, allowing the large model to generate the execution results for the user-specified task based on this data. The accuracy of determining this reference data will impact the performance of the large model in executing the task.

[0004] When the performance of a large model fails to meet requirements, it is necessary to analyze the problems leading to this result and then resolve them to optimize the performance. This problem analysis includes analyzing the process of determining the reference data. However, the effectiveness of current problem analysis methods for determining this reference data still needs improvement. Summary of the Invention

[0005] This specification provides a problem analysis method for query tasks, which can improve the effectiveness of problem analysis for query tasks. One or more embodiments of this specification also relate to a problem analysis apparatus for query tasks, a computing device, a computer-readable storage medium, and a computer program product.

[0006] According to one aspect of the embodiments of this specification, a problem analysis method for query tasks is provided, including: The intent analysis vector and reference vector obtained from executing the query task are respectively mapped to different mapping points in the first display area. The intent analysis vector is determined by performing intent analysis on the query information corresponding to the query task, and the reference vector is determined by querying the candidate data based on the intent analysis vector. In response to an execution parameter adjustment instruction for the query task, the execution parameters are adjusted; Based on the updated intent analysis vector and reference vector, the mapping points in the first display area are updated, wherein the updated intent analysis vector and reference vector are obtained by executing the query task based on the adjusted execution parameters; Based on the distribution information of the mapping points in the updated first display area, problem information in the execution of the query task is determined.

[0007] According to another aspect of the embodiments of this specification, a problem analysis apparatus for query tasks is provided, comprising: The first mapping module is used to map the intent analysis vector and reference vector obtained from executing the query task to different mapping points in the first display area, wherein the intent analysis vector is determined by performing intent analysis on the query information corresponding to the query task, and the reference vector is determined by querying the candidate data based on the intent analysis vector. An adjustment module is configured to adjust the execution parameters in response to an execution parameter adjustment instruction for the query task; An update module is used to update the mapping points in the first display area based on the updated intent analysis vector and reference vector, wherein the updated intent analysis vector and reference vector are obtained by performing the query task based on the adjusted execution parameters; The first determining module is used to determine the problem information that exists in the execution of the query task based on the distribution information of the mapping points in the updated first display area.

[0008] According to another aspect of the embodiments of this specification, a computing device is provided, including: a memory and a processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the above method.

[0009] According to another aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0010] According to another aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed in a processor, implement the steps of the above-described method.

[0011] In one embodiment of this specification, the intent analysis vector and reference vector obtained from executing a query task can be mapped to different mapping points in a first display area for display. This allows for a visual understanding of the relationship between the reference vector and the intent analysis vector through the distribution of these mapping points. Furthermore, by adjusting the execution parameters of the query task, the obtained intent analysis vector and reference vector can be updated, and the mapping points can be updated accordingly, enabling observation of changes in mapping points corresponding to different execution parameters. Since different problems can lead to different changes in the relationship between vectors, the distribution changes of these execution parameter-related mapping points can facilitate a convenient and intuitive identification of problems related to those execution parameters, improving the ease of problem analysis for query tasks and enhancing the effectiveness of problem analysis. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating a problem analysis method for query tasks provided in one embodiment of this specification; Figure 2 This is a schematic diagram of a display page provided in one embodiment of this specification; Figure 3 This is a schematic diagram of the structure of a problem analysis device for query tasks provided in one embodiment of this specification; Figure 4 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0013] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0014] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items. The term “at least one” as used in one or more embodiments of this specification means “one or more,” and “a plurality of” means “two or more.” The term “comprising” is an open-ended description and should be understood as “including but not limiting,” and may include other content in addition to what has been described.

[0015] It should be understood that although the terms “first,” “second,” etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, “first” may also be referred to as “second” without departing from the scope of one or more embodiments of this specification, and similarly, “second” may also be referred to as “first.” Depending on the context, the word “if” as used herein may be interpreted as “when,” “in response to a determination,” or “when…”.

[0016] Furthermore, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. The collection, use and processing of related data must comply with relevant standards and requirements, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0017] In one or more embodiments of this specification, the model referred to may be a large model. A large model is a deep learning model with a large number of model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even tens of trillions of model parameters. A large model can also be called a foundation model. It is pre-trained on a large-scale unlabeled corpus to produce a pre-trained model with hundreds of millions of parameters. Such a model can adapt to a wide range of downstream tasks and has good generalization ability. Examples include large-scale language models (LLMs) and multi-modal pre-training models. In practical applications, large models require only a small number of samples to fine-tune the pre-trained model before being applied to different tasks. Large models can be widely used in fields such as Natural Language Processing (NLP) and Computer Vision. Specifically, they can be applied to computer vision tasks such as Visual Question Answering (VQA), Image Captioning (IC), and Image Generation, as well as NLP tasks such as text-based sentiment classification, text summarization, and machine translation. Major application scenarios for large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design. The models mentioned in the embodiments of this specification include, but are not limited to, the large models described above, and can be any type of machine learning model, such as convolutional neural network (CNN) models, Transformer models, etc.

[0018] Currently, large models are being used more and more widely. These models can interact with users through dialogue, execute tasks corresponding to user input, and output responses containing those results. However, large models suffer from the problem of "illusion" (i.e., inaccurate or irrelevant content), and Retrieval-Enhanced Generation (RAG) technology is an effective means of reducing this illusion. RAG is an AI framework that combines retrieval and generation techniques. When using RAG, large models first retrieve task-related reference data from external data sources such as databases, documents, or web pages. This reference data is then used to guide and enhance the generated responses. This approach ensures the accuracy and relevance of the responses, reduces the possibility of generating illusory content, and enriches the context with up-to-date, specific information, ensuring that the large model outputs more evidence-based and detailed responses.

[0019] The determination of the reference data has a significant impact on the performance of large-scale models. If the reference data is reasonably determined and does not contain a large amount of useless data, it will greatly improve the efficiency and accuracy of the large-scale model's task execution. The process of determining reference data based on user input information (referred to as query information) typically includes steps such as query intent analysis, vector computation to recall candidate results, and scoring and ranking the candidate results. This process is usually implemented using a black-box model, hiding the complex relationship between the user's query input and the query results within the black-box model.

[0020] This black-box model can include a query input module, an intent analysis module, a vector recall module, a ranking module, and a result output module, connected in sequence. The intent analysis module receives the user's query input and parses the corresponding query intent using natural language processing techniques. The vector recall module searches for similar vectors in a pre-constructed high-dimensional vector space based on the intent analysis results, recalling candidate results that match the intent analysis results. The ranking module scores the recalled candidate results based on similarity and outputs the top N results as reference data for the query.

[0021] When the performance of a large-scale model fails to meet user expectations, users can provide feedback. In such cases, staff need to conduct a problem analysis of the entire task execution process to identify the root cause of the unsatisfactory performance. Solving this problem allows for optimization of the task execution process, improving the subsequent performance of the large-scale model. Problem analysis is crucial for optimizing large-scale models.

[0022] The task execution process includes querying reference data before generating execution results. The effectiveness of determining this reference data significantly impacts the overall task execution performance of the large model. Therefore, it is necessary to analyze whether there are problems at each node during the reference data query process. Since this reference data is obtained through a black-box model executing the query task, the process lacks transparency, does not provide visualization of intermediate processes, and makes it difficult to diagnose the reasons for poor reference data determination. Furthermore, when the reference data determination does not meet expectations, it is impossible to pinpoint whether the problem lies in the intent analysis, recall, or ranking phases, and it is impossible to intervene in the intermediate processes, perform segmented debugging, and optimization. Therefore, the current analysis of problems in the reference data determination process is quite challenging, and optimizing the reference data determination process for executing the query task is also very difficult.

[0023] This specification provides a problem analysis method for query tasks, which simplifies the problem analysis process and improves the effectiveness of problem analysis. This specification also relates to a problem analysis apparatus for query tasks, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments. This problem analysis method for query tasks can be executed by the problem analysis apparatus or the computing device, such as a server device or a terminal device such as a desktop computer, laptop computer, or smartphone.

[0024] Figure 1 This is a flowchart illustrating a problem analysis method for query tasks provided in one embodiment of this specification. This method can be executed when the task execution performance of a large model is determined to be unsatisfactory, and primarily focuses on problem analysis of the process by which the large model uses retrieval enhancement generation techniques to retrieve task-related reference data. The following example illustrates the application of this method to a problem analysis device for query tasks, which can be a device with display functionality. Figure 1 As shown, the problem analysis method for query tasks includes the following steps 102 to 108.

[0025] Step 102: Map the intent analysis vector and reference vector obtained from executing the query task to different mapping points in the first display area. The intent analysis vector is determined by performing intent analysis on the query information corresponding to the query task, and the reference vector is determined by querying the candidate data based on the intent analysis vector.

[0026] The large model can utilize retrieval-enhanced generation techniques to generate response information corresponding to user input information. This response information is also the execution result of the task corresponding to the user input information. The process of the large model generating this execution result includes sequentially executing a query task and a generation task. The large model can first execute a query task based on the user input information, query reference data related to the user input information from the candidate data, and then execute a generation task based on the reference data to generate the response information corresponding to the user input information. The candidate data may include data from data sources such as databases, documents, or web pages pre-specified by the staff. The problem analysis method provided in the embodiments of this specification is used to perform problem analysis on the process of the large model executing the query task based on user input information.

[0027] During problem analysis, a large model can execute the query task, and the output results of each node during execution can be analyzed to identify problems in the query task's execution. In this process, the query task can be executed based on query information, which can be analogous to user input information provided by the large model when actually serving users. This query information can include text, and in some implementations, audio, images, or video. The query information can be determined based on the user input information used to report problems; for example, the user input information can be directly used as the query information, or multiple user inputs can be integrated into a single query. In some implementations, the query information can also be constructed by staff based on needs, such as constructing different query information for different domains.

[0028] In the embodiments described in this specification, the large model can be deployed on a problem analysis device, allowing the device to directly use the large model to execute query tasks and obtain the output results of each node during the execution of the query task. Alternatively, the large model can be deployed on other devices, allowing the problem analysis device to call the large model through an interface to execute query tasks and receive the output results of each node during the execution of the query task from the other device.

[0029] In some implementations, prior to step 102, the problem analysis method provided in this specification further includes: using a large model to perform a query task based on query information to obtain an intent analysis vector and a reference vector. During the execution of the query task, intent analysis processing can be performed based on the query information to obtain the intent analysis vector. The intent analysis vector is a vector that matches the query intent corresponding to the query information and is used to represent the semantic features of the query target.

[0030] Next, based on the intent analysis vector, a query is performed in the candidate data to determine reference vectors whose similarity to the intent analysis vector is greater than a similarity threshold. These reference vectors can belong to the execution result of the query task. The data corresponding to these reference vectors is the aforementioned reference data. The similarity threshold can vary depending on the query information (e.g., query information from different domains). For example, if the candidate data is data from a database storing a vector set, querying reference vectors involves retrieving vectors from this database. The problem analysis device can calculate the similarity between the intent analysis vector and each vector in this vector set (e.g., calculate cosine similarity) to determine reference vectors that meet the similarity threshold. The number of reference vectors retrieved can be one or more, or no reference vectors may be found. Each reference vector is a candidate that may match the query information.

[0031] In some implementations, both the intent analysis vector and the reference vector can be high-dimensional vectors, such as vectors with up to thousands of dimensions. For example, if the query information includes "movie recommendations," the query analysis device can perform intent analysis on the query information, identify that the query information belongs to the recommendation category, and determine intent features such as "recommended content is movies" and "sorted by popularity." Based on these intent features, the query information is converted into a numerical vector to obtain the intent analysis vector. For example, the intent analysis vector could be [0.45, -0.12, 0.32, ...]. Based on this intent analysis vector, hundreds of related movie vectors can be recalled as candidate results. For example, the reference vector could be a high-dimensional vector representation of movie 1, movie 2, etc., and a certain reference vector could be represented as [0.51, -0.19, 0.25, ...].

[0032] The problem analysis device can acquire intent analysis vectors and reference vectors obtained from executing query tasks, and visualize these vectors by mapping each vector to a display area. For example, each vector can be mapped to a point in the display area, allowing for the display of each vector using a scatter plot. To distinguish it from other content display areas, the display area containing the mapping points of each vector is referred to as the first display area.

[0033] In some implementations, the problem analysis device can perform dimensionality reduction on the high-dimensional vectors (such as intent analysis vectors and reference vectors) obtained from performing the query task, and determine the corresponding mapping points based on the vector values ​​obtained after dimensionality reduction, thereby displaying the mapping points corresponding to each vector. This allows for convenient determination of the corresponding mapping points based on the vector values, enabling each vector to represent the degree of correlation through relative distance, facilitating the visualization of each vector.

[0034] Taking the dimensionality reduction of a vector to obtain a two-dimensional vector as an example, step 102 may include: reducing the dimensionality of the intent analysis vector and reference vector obtained from executing the query task to two-dimensional vectors; displaying the mapping points corresponding to each two-dimensional vector obtained from the dimensionality reduction in a two-dimensional coordinate system in the first display area; wherein, the two coordinate axes of the two-dimensional coordinate system correspond to the two vector values ​​in the two-dimensional vector, and the display methods of the mapping points corresponding to the intent analysis vector and the reference vector in the first display area are different. After reducing the dimensionality of the intent analysis vector and the reference vector to two-dimensional vectors, since the two-dimensional vector is represented by only two vector values, these two vector values ​​can be directly used as two coordinate values ​​to determine a point in the two-dimensional coordinate system, which is the mapping point of the two-dimensional vector.

[0035] Dimensionality reduction maps high-dimensional data to a low-dimensional space while preserving the local and global structure of the original data. Problem analysis tools can utilize dimensionality reduction algorithms such as t-SNE, Principal Component Analysis (PCA), UMAP (Uniform Manifold Approximation and Projection), and Sammon mapping to reduce the dimensionality of the intent analysis vector and reference vector. For example, using the t-SNE algorithm, a loss function is constructed, and gradient descent is used to map high-dimensional data to a low-dimensional space. First, the similarity between sample points in the high-dimensional data is calculated, and then the similarity between sample points in the low-dimensional space is calculated. The mapping relationship is optimized by minimizing the difference between the two similarities. t-SNE uses the t-distribution to define the similarity between samples in high-dimensional data, hence the name t-SNE.

[0036] In the embodiments described in this specification, the display methods for the mapping points corresponding to the intended analysis vector and the reference vector are different. For example, they can be displayed using different colors or represented by patterns of different shapes. Optionally, the problem analysis device can use data visualization tools (such as Matplotlib, Plotly, D3.js, Tableau) to draw two-dimensional dot plots or scatter plots based on the vector values ​​of each vector after dimensionality reduction, and distinguish the mapping points of different types of vectors by color and markers.

[0037] Figure 2 This is a schematic diagram of a display page provided in one embodiment of this specification. For example... Figure 2 As shown, a two-dimensional coordinate system can be displayed in the first display area Q1 of the display page of the problem analysis device. The two coordinate axes of the two-dimensional coordinate system correspond to the two vector values ​​of the two-dimensional vector, respectively. Figure 2 In the first display area, each small circle represents a vector. The black-filled small circle Y1 represents the intent analysis vector, and the hollow small circle Y2 represents the reference vector. An example diagram uses eight reference vectors. For instance, after dimensionality reduction to a two-dimensional space, the intent analysis vector might be mapped to a point with coordinates [0.25, -0.35], while the mapped points of the reference vectors are distributed near the point corresponding to the intent analysis vector.

[0038] In some implementations, the first display area can be a three-dimensional display area, or the vector obtained by performing the query task can be dimensionality reduced to obtain a three-dimensional vector, so as to display the mapping points of each three-dimensional vector in the three-dimensional display area.

[0039] Step 104: In response to the execution parameter adjustment instruction for the query task, adjust the execution parameter.

[0040] In the embodiments of this specification, corresponding execution parameters can be pre-set for query tasks performed on large models, so that the large model performs the query task based on these execution parameters. These execution parameters may include a similarity threshold on which the query reference vector is based, multiple data attribute weights, and possibly other execution parameters, which are not limited here. The candidate data may contain data with multiple data attributes, and each data attribute weight can correspond to a data attribute. Each data attribute weight can characterize the degree of influence of the corresponding data attribute on the queried reference vector. These data attribute weights can also be called vector field weights.

[0041] When searching for a reference vector in the candidate data, relevant content may be found in data with multiple data attributes. However, the relevant content for different data attributes needs to be combined with the corresponding data attribute weights to determine the final similarity between the relevant content and the intent analysis vector. The reference vector is then determined based on this final similarity. For example, if content with the same degree of similarity to the intent analysis vector is found in both data attribute a and data attribute b, but the weight of data attribute a is greater than the weight of data attribute b, then the final similarity of the relevant content for data attribute a can be greater than the final similarity of the relevant content for data attribute b.

[0042] For example, these multiple data attributes may include title attributes, body attributes, and category information attributes. The weight of the data attribute corresponding to the title attribute can be greater than the weight of the data attribute corresponding to the body attribute. Optionally, the weights of each data attribute can be different for different query information.

[0043] In this embodiment of the specification, the intent analysis vector and reference vector in step 102 above can be obtained by the large model performing a query task based on pre-set execution parameters. In step 104, the user of the problem analysis device (i.e., the staff performing the problem analysis) can adjust the execution parameters so that the problem analysis device receives an execution parameter adjustment instruction for the query task and adjusts the corresponding execution parameters based on the instruction. The execution parameter adjustment instruction can indicate the adjustment of one or more execution parameters, and can also indicate the parameter value to which the execution parameter is to be adjusted.

[0044] In some implementations, the problem analysis device can display adjustment controls for each execution parameter, allowing users to interactively adjust the parameters. This parameter adjustment method is relatively simple and quick. Accordingly, in step 104, in response to an execution parameter adjustment instruction for the query task, before adjusting the execution parameters, the problem analysis method in this embodiment further includes: displaying adjustment controls for the execution parameters of the query task in a second display area; and receiving an execution parameter adjustment instruction for the query task when the adjustment controls are triggered. These adjustment controls may include tools such as sliders, checkboxes, and input boxes.

[0045] For example, please continue to refer to Figure 2 The second display area Q2 can display a first adjustment control K1 with a similarity threshold. This first adjustment control K1 may include an increase button and a decrease button, as well as a parameter value display box. Users can adjust the similarity threshold by clicking the increase or decrease button, and the parameter value display box can show the current similarity threshold (e.g., 0.8). In some embodiments, users can also adjust the similarity threshold through other input methods. For example, users can place the mouse cursor on the location of the first adjustment control K1 and slide the mouse ball to adjust the similarity threshold. The problem analysis device can be triggered to receive a parameter adjustment instruction when the value in the parameter value display box changes.

[0046] The second display area Q2 may also display a second adjustment control K2 containing data attribute weights. This second adjustment control K2 may include an input box where the user can enter adjustment information for each data attribute weight. The problem analysis device can automatically identify this adjustment information, determine the data attribute weights to be adjusted and the weight values ​​of the values ​​to be adjusted, receive the corresponding execution parameter adjustment instructions, and adjust the data attribute weights based on these instructions.

[0047] In this specification, the examples provided only illustrate the two implementation methods described above for adjusting the execution parameters. Each execution parameter adjustment control can be adjusted according to user needs, and no limitation is made here. In some implementations, a confirmation control can also be displayed in the second display area Q2 for each execution parameter. After the user inputs the parameter value to be adjusted, they can click the confirmation control, triggering the problem analysis device to receive the execution parameter adjustment instruction carrying that parameter value.

[0048] Step 106: Based on the updated intent analysis vector and reference vector, update the mapping points in the first display area, wherein the updated intent analysis vector and reference vector are obtained by performing a query task based on the adjusted execution parameters.

[0049] After adjusting the execution parameters on which the large model executes the query task (before step 106), the problem analysis device can cause the large model to re-execute the query task based on the adjusted execution parameters, obtaining updated intent analysis vectors and reference vectors. The problem analysis device can then map these updated intent analysis vectors and reference vectors to mapping points in the first display area. For example, the problem analysis device can clear the previously displayed mapping points in the first display area and re-display the mapping points based on the updated vectors, thus updating the mapping points in the first display area. For details regarding the execution of the query task and the mapping method, please refer to the relevant description in step 102 above; further details will not be provided here.

[0050] In some implementations, by adjusting execution parameters (such as the aforementioned similarity threshold and data attribute weights), only the reference vector obtained from executing the query task can be changed, while the intent analysis vector may remain unchanged. In some implementations, the execution parameters may also include parameters related to intent analysis processing. In this case, adjusting the execution parameters will cause a change in the intent analysis vector, which will correspondingly lead to a change in the reference vector.

[0051] In the embodiments described in this specification, the user can adjust the execution parameters multiple times. For each adjustment, the problem analysis device will acquire the updated intent analysis vector and reference vector in real time, and update the mapping points in the first display area accordingly.

[0052] Step 108: Based on the distribution information of the mapped points in the updated first display area, determine the problem information that exists in the execution of the query task.

[0053] After updating the mapping points in the first display area by adjusting the execution parameters, the problem analysis device can determine the problems existing in the execution of the query task based on the updated mapping points (including the distribution information of the updated mapping points). The problem analysis device can locate the problems existing in the execution of the query task by comparing the changes in the distribution information of the mapping points before and after the update, such as based on the changes in the distribution information in multiple updates.

[0054] In some implementations, the problem information determined by the problem analysis device may only indicate nodes where problems exist during the execution of the query task, such as a problem with a schematic diagram analysis node or a problem with a vector recall node. In other implementations, the problem information may also indicate specific problems, such as improper setting of the similarity threshold or improper setting of the weights of certain data attributes.

[0055] In the embodiments described in this specification, the problem information can be automatically analyzed by a problem analysis device based on a variety of pre-set problem determination conditions. For example, problem determination conditions can be pre-set for various problems, and these conditions are related to the distribution of mapping points. The problem analysis device can determine whether the distribution of mapping points resulting from parameter adjustments matches a certain problem determination condition, thereby determining the existence of a problem corresponding to that condition. In some implementations, the problem information can also be manually analyzed by a user observing the distribution of mapping points.

[0056] In the embodiments of this specification, if the execution process of the query task is problem-free and the execution effect of the query task is good, then the reference vector obtained by the query should have a high similarity to the intent analysis vector, and the mapping points should be distributed according to the target distribution. This target distribution is that the mapping points of each reference vector are densely distributed around the mapping points of the intent analysis vector. Users can adjust the execution parameters to observe whether the mapping points can be distributed according to the target distribution to determine problem information. If adjusting a certain execution parameter can change the mapping points from other situations to the target distribution, then it can be determined that the problem information is due to an improper setting of that execution parameter. If adjusting the execution parameters cannot make the mapping points distribute according to the target distribution, then it can be determined that the problem does not occur in the setting of that execution parameter, but in other aspects, and further problem information can be determined from other aspects.

[0057] In some implementations, the execution parameters include a similarity threshold, and the reference vector is a vector whose similarity to the intent analysis vector is greater than the similarity threshold. Step 108, based on the distribution information of the mapping points in the updated first display area, determines the problem information existing in the execution of the query task, including: if, by adjusting the similarity threshold, the mapping points in the updated first display area are clustered within the target range, determining that the problem information existing in the execution of the query task includes an incorrect similarity threshold setting on which the query reference vector is based; wherein, the target range includes the location of the mapping point corresponding to the intent analysis vector. This target range can be a certain size range surrounding the mapping point corresponding to the intent analysis vector. The clustering of mapping points within the target range can conform to the above-described target distribution. For example, the target range can be preset, and the target distribution can be set such that there are at least a target number of mapping points within the target range. The problem analysis device can automatically analyze and determine whether there is relevant problem information based on this preset target distribution and the actual distribution of the mapping points.

[0058] By adjusting the similarity threshold, if the distribution of mapping points in the first display area changes from not conforming to the target distribution to conforming to the target distribution, then the problem lies with the vector recall node, specifically an improperly set similarity threshold. For example, if adjusting the similarity threshold changes the mapping points corresponding to reference vectors in the first display area from being scattered over a large area to being clustered within the target area, then the original similarity threshold was set too low and needs to be increased. If, for instance, 10 reference vectors are found, and the mapping points of 8 of them are located around the mapping point of the intent analysis vector, while the other two are far away, the user can increase the similarity threshold to filter out those two mapping points, causing the mapping points of the reference vectors to cluster within the target area. Conversely, if adjusting the similarity threshold changes the number of mapping points corresponding to reference vectors from being too few or even nonexistent to being more clustered within the target area, then the original similarity threshold was set too high and needs to be decreased.

[0059] In some implementations, the execution parameters include multiple data attribute weights, which characterize the degree of influence of the corresponding data attribute on the query reference vector. Step 108, based on the distribution information of the mapping points in the updated first display area, determines the problem information existing in the execution of the query task, including: if, by adjusting the data attribute weights to make the mapping points in the updated first display area cluster within the target range, determining that the problem information existing in the execution of the query task includes an incorrect setting of the data attribute weights on which the query reference vector is based; wherein, the target range includes the location of the mapping point corresponding to the intent analysis vector. This implementation can refer to the aforementioned method of adjusting the similarity threshold to determine problem information, and will not be elaborated further here.

[0060] For example, when adjusting the weights of data attributes, users can arbitrarily adjust the weights of various data attributes. They can increase or decrease the weights of multiple data attributes simultaneously, or adjust the weight of a single data attribute. The sum of all data attribute weights is 1, and the problem analysis device can automatically and adaptively change the values ​​of other data attribute weights based on the user's adjustments to some data attribute weights.

[0061] If adjusting the data attribute weights changes the distribution of the mapped points corresponding to the reference vector from not conforming to the target distribution to conforming to the target distribution, then the problem lies with the vector recall node, specifically with improperly set data attribute weights. In some implementations, the similarity threshold can be adjusted first. If adjusting the similarity threshold makes the mapped points conform to the target distribution, the problem can be directly identified without further analysis. Only when adjusting the similarity threshold fails to make the mapped points conform to the target distribution should the data attribute weights be adjusted. Since adjusting the similarity threshold only requires adjusting one parameter, this method is relatively simple, allowing for the prior identification or elimination of this type of problem, ensuring high efficiency in problem identification.

[0062] The process of executing a query task based on query information may also include a node that converts the query information into a query vector. This node may be located before the intent analysis node. By comparing the query vector and the intent analysis vector, problems in the execution of the query task can also be located. Since both the query vector and the intent analysis vector represent query information, their similarity should be high during normal task execution. If the query vector and the intent analysis vector differ significantly, the execution effect of the query task will be poor, and the reference vector obtained from the query will not match the query information well. In the embodiments of this specification, the query vector may also be mapped to a mapping point in the first display area to determine whether the query vector and the intent analysis vector match through the mapping point in the first display area.

[0063] Accordingly, the problem analysis method provided in the embodiments of this specification further includes: mapping a query vector obtained by vector transformation of query information to a mapping point in a first display area; when the target distance in the first display area is greater than a distance threshold, determining the problem information existing in the execution of the query task includes: the query information is ambiguous, or there is a problem in the intent analysis processing; wherein, the target distance is the distance between the mapping point corresponding to the query vector and the mapping point corresponding to the intent analysis vector. The query vector can also be a high-dimensional vector, and the problem analysis device can perform dimensionality reduction processing on the query vector to perform mapping based on the vector obtained after dimensionality reduction. For example, please continue to refer to... Figure 2 The small circle Y3 enclosed by a dashed line represents the mapping point corresponding to the query vector. For details, please refer to the previous introduction regarding intent analysis vectors and reference vectors; it will not be repeated here. In this case, the problem information can be automatically determined by the problem analysis device or manually determined by the user.

[0064] When the similarity between the query vector and the intent analysis vector is high, it can be assumed that both the definite node of the query vector and the definite node of the intent analysis vector (i.e., the intent analysis node) are problem-free. This similarity can be represented by the target distance between relative mapping points in the first display area. If this target distance is less than or equal to a distance threshold, the similarity between the two vectors meets the requirements; otherwise, the similarity between the two vectors is considered unsatisfactory. If the similarity does not meet the requirements, it may be due to a problem with either the definite node of the query vector or the intent analysis node. It would be beneficial to include information about both nodes in the problem information, allowing staff to conduct further analysis based on this information.

[0065] Problems with determining the nodes of the query vector may stem from ambiguity in the query information itself. For example, if the query information includes a term that refers to different things in different contexts, the generated query vector and the generated intent analysis vector may correspond to two different things, resulting in vectors with unsatisfactory similarity.

[0066] In the embodiments of this specification, after the problem analysis device determines the problem information that exists in the execution process of the query task, it can send the problem information to the problem solving device or to the staff who solve the problem, so that the execution process of the query task can be adjusted and optimized based on the problem information.

[0067] In some implementations, the problem analysis device can also determine suggested solutions or optimization directions. Accordingly, the problem analysis method provided in this specification further includes: determining solution information for the problem information based on the distribution information of different execution parameters and corresponding mapping points. The problem analysis device can analyze the distribution information of mapping points corresponding to the adjustment of execution parameters, and determine solution information for the problem information based on the execution parameters corresponding to a state that is relatively close to the target distribution. This solution information may include specific solutions, such as suggested execution parameters and the values ​​to which the execution parameters should be adjusted. The solution information may also only include a general direction of solution, such as increasing or decreasing a certain execution parameter, or optimizing a certain node. In this way, the corresponding problem can be solved quickly and accurately based on the solution information, ensuring high optimization efficiency for the task processing of large models.

[0068] This specification provides an interactive problem analysis method. The problem analysis device can map the high-dimensional vectors obtained from each node during the execution of a query task to a two-dimensional space, so as to visualize and analyze the output results of each node and realize problem diagnosis. The problem analysis process can include the following steps: query information input, intent analysis, vector retrieval, high-dimensional vector mapping, two-dimensional space visualization, interactive analysis, problem localization and debugging.

[0069] In the query information input step, the query information can be the query text entered by the user. The question analysis device can convert the query information into a query vector, which is a high-dimensional semantic representation of the query text. This query vector can characterize the user's query intent. For example, if the user enters the query information "movie recommendations," the question analysis device can parse that the user's intent is to obtain movie recommendations that match their preferences, and accordingly generate a semantic vector of the query information. The vector dimension can be as high as hundreds of dimensions.

[0070] In the intent analysis step, the question analysis device can utilize a large language model to perform intent analysis on the query information, supplementing it with more features to obtain an intent analysis vector that matches the query intent. This intent analysis vector is also a high-dimensional vector used to represent the semantic features of the retrieval target. In the vector recall step, the question analysis device can, based on the intent analysis vector, recall multiple reference vectors from the candidate data to form a high-dimensional vector set, representing the set of content that matches the user's query information. Each reference vector corresponds to a candidate that may match the intent analysis vector.

[0071] In the high-dimensional vector mapping step, the problem analysis device uses a dimensionality reduction algorithm to reduce the dimensionality of the recalled high-dimensional vector set, along with the previously determined query vector and intent analysis vector, to a two-dimensional space, resulting in a two-dimensional mapping point set. In the two-dimensional space visualization step, visualization tools are used to display the obtained two-dimensional mapping point set, generating a two-dimensional distribution map of each mapping point, thus visualizing the vectors obtained in each step. For example, the two-dimensional mapping point set can be displayed as a scatter plot, with different colors used to mark mapping points of different categories (query vector, intent analysis vector, reference vector), showing the relative position of each vector in space, making it easy for users to observe the distribution and correlation of the mapping points.

[0072] In the interactive analysis step, users can use the displayed adjustment controls to dynamically adjust the execution parameters, triggering updates to the vectors obtained in each step and correspondingly updating the two-dimensional distribution map. This allows for real-time observation of changes in the distribution and ranking of the mapping points. For example, users can dynamically modify vector recall conditions (such as similarity thresholds) or data attribute weights using the adjustment controls, adjusting the real-time recalled reference vectors and the generated two-dimensional distribution map. For instance, if a user adjusts the similarity threshold from 0.7 to 0.8, the problem analysis device can recalculate the reference vectors that meet the new similarity threshold and update the distribution of mapping points in the two-dimensional distribution map in real time, allowing users to observe the impact of the newly recalled reference vectors on the final distribution.

[0073] In the problem localization and debugging process, the problem analysis device can locate problem information and determine parameter improvement suggestions based on the adjusted two-dimensional distribution map. Users can analyze the two-dimensional distribution map to determine whether the mapping points corresponding to the queried reference vectors are clustered within the expected range, and further adjust the execution parameters based on this analysis. For example, as the similarity threshold increases, the mapping points of the reference vectors in the two-dimensional distribution map can become more concentrated and closer to the mapping points of the intent analysis vector. Users can adjust the execution parameters multiple times, observing how each change affects the two-dimensional distribution map, and ultimately find the optimal similarity threshold range, achieving debugging and optimization of the similarity threshold. This determined similarity threshold range can serve as parameter improvement suggestions.

[0074] For example, assuming a user inputs the query "Recommended popular science fiction movies," the query input step allows the user to enter the text "Recommended popular science fiction movies," resulting in a query vector. In the intent analysis step, semantic information such as "Content Category = Movie," "Category Tag = Science Fiction," and "Preference Type = Popular" can be parsed and an intent analysis vector generated. In the vector recall step, high-dimensional vectors of multiple movies in the database similar to the "popular science fiction movies" intent can be recalled, forming a high-dimensional vector set. In the high-dimensional vector mapping step, t-SNE can be used to map the query vector, intent analysis vector, and recalled reference vector to a two-dimensional space. In the two-dimensional space visualization step, the mapped points are displayed in different colors, forming a scatter plot to visually demonstrate the relationship between the query information and the recalled reference vector. In the interactive analysis step, the user adjusts the similarity threshold and observes the real-time changes in the two-dimensional distribution plot. In the problem localization and debugging step, based on the observation results, it is determined whether further adjustments to retrieval parameters (such as data attribute weights) are needed to optimize retrieval performance, and problem information and suggested improvements to parameters are identified.

[0075] In the embodiments of this specification, the high-dimensional vector set of the recall is the semantic vector of each candidate option for the query information in the candidate data. Each vector contains hundreds of dimensions and rich content features. After dimensionality reduction to two-dimensional space using t-SNE, each candidate option becomes a point on a two-dimensional scatter plot. After dimensionality reduction, the spatial relationship between the query, intent analysis, and candidate options can be displayed in the two-dimensional plane, facilitating intuitive observation of the similarity and correlation between intent and recall results, and enabling visualized problem diagnosis. In the embodiments of this specification, this dimensionality reduction technique is applied to the field of semantic retrieval, especially in the context of intent analysis and retrieval effect diagnosis. It can preserve the relative distance and structure between the original data, thereby allowing users to intuitively analyze and understand the retrieval process in two-dimensional space.

[0076] In the embodiments described in this specification, visualization technology allows users to directly observe the intermediate processes of intent analysis, reference vector recall, and ranking, improving the transparency of the query task execution process. Interactive analysis tools allow users to dynamically adjust various execution parameters, pinpoint the specific stage of the problem, and perform segmented debugging and optimization of the task execution process, facilitating rapid problem-solving and improving the overall performance of the retrieval system.

[0077] In summary, the problem analysis method for query tasks provided in this specification can map the intent analysis vector and reference vector obtained from executing the query task to different mapping points in a first display area for display. This allows for a direct and intuitive determination of the relationship between the reference vector and the intent analysis vector through the distribution of these mapping points. Furthermore, by adjusting the execution parameters of the query task, the obtained intent analysis vector and reference vector can be updated, and the mapping points can be updated accordingly, enabling observation of changes in mapping points corresponding to different execution parameters. Since different problems lead to different changes in the relationship between vectors, the distribution changes of these execution parameter-related mapping points facilitate intuitive and easy location of problems related to those execution parameters, improving the simplicity and effectiveness of problem analysis for query tasks.

[0078] Corresponding to the above method embodiments, this specification also provides embodiments of a problem analysis device for query tasks. Figure 3 This is a schematic diagram of the structure of a problem analysis device for query tasks provided in one embodiment of this specification. Figure 3 As shown, the problem analysis device for query tasks includes: The first mapping module 302 is used to map the intent analysis vector and reference vector obtained from executing the query task to different mapping points in the first display area, wherein the intent analysis vector is determined by performing intent analysis on the query information corresponding to the query task, and the reference vector is determined by querying the candidate data based on the intent analysis vector. The adjustment module 304 is used to adjust the execution parameters in response to the execution parameter adjustment command for the query task; The update module 306 is used to update the mapping points in the first display area based on the updated intent analysis vector and reference vector, wherein the updated intent analysis vector and reference vector are obtained by performing a query task based on the adjusted execution parameters; The first determining module 308 is used to determine the problem information that exists in the execution of the query task based on the distribution information of the mapped points in the updated first display area.

[0079] Optionally, the execution parameters include a similarity threshold, and the reference vector is a vector whose similarity to the intent analysis vector is greater than the similarity threshold; the first determining module 308 is used for: By adjusting the similarity threshold to ensure that the mapped points in the updated first display area cluster within the target range, information indicating problems in the execution of the query task includes an incorrect similarity threshold setting on which the query reference vector is based; the target range includes the location of the mapped points corresponding to the intent analysis vector.

[0080] Optionally, the execution parameters include various data attribute weights, whereby the data attribute weights characterize the degree of influence of the corresponding data attribute on the reference vector of the query; the first determining module 308 is used for: By adjusting the data attribute weights to ensure that the mapped points in the updated first display area are clustered within the target range, the problem information identified during the execution of the query task includes incorrect data attribute weight settings on which the query reference vector is based; the target range includes the location of the mapped points corresponding to the intent analysis vector.

[0081] Optionally, the problem analysis device further includes: The second mapping module is used to map the query vector obtained by vector transformation of the query information to mapping points in the first display area. The second determining module is used to determine, when the target distance in the first display area is greater than a distance threshold, the problem information existing in the execution of the query task, including: the query information is ambiguous, or there is a problem in the intent analysis processing; wherein, the target distance is the distance between the mapping point corresponding to the query vector and the mapping point corresponding to the intent analysis vector.

[0082] Optionally, the problem analysis device further includes: The display module is used to respond to the execution parameter adjustment command for the query task and, before adjusting the execution parameters, display the adjustment control for the execution parameters of the query task in the second display area; The receiving module is used to receive the execution parameter adjustment instructions for the query task when the adjustment control is triggered.

[0083] Optionally, the first mapping module 302 is used for: The intent analysis vector and reference vector obtained from the query task are reduced in dimensionality to two-dimensional vectors. In the two-dimensional coordinate system of the first display area, the mapping points corresponding to each two-dimensional vector obtained by dimensionality reduction are displayed; wherein, the two coordinate axes of the two-dimensional coordinate system correspond to the two vector values ​​in the two-dimensional vector respectively, and the display methods of the mapping points corresponding to the intended analysis vector and the reference vector in the first display area are different.

[0084] Optionally, the problem analysis device further includes: The third determination module is used to determine the solution information for the problem information based on different execution parameters and the distribution information of the corresponding mapping points.

[0085] In summary, in the problem analysis device for query tasks provided in this embodiment, the first mapping module can map the intent analysis vector and reference vector obtained from executing the query task to different mapping points in the first display area for display. This allows for a direct and intuitive determination of the relationship between the reference vector and the intent analysis vector through the distribution of the mapping points. Furthermore, the adjustment module can update the obtained intent analysis vector and reference vector by adjusting the execution parameters of the query task, and the update module updates the mapping points accordingly to observe the changes in mapping points corresponding to different execution parameters. Since different problems can lead to different changes in the relationship between vectors, the first determining module can easily and intuitively locate problems related to the execution parameters by observing the distribution changes of the mapping points related to the execution parameters, thus improving the ease of problem analysis for query tasks and enhancing the effectiveness of problem analysis.

[0086] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. For the problem analysis apparatus for query tasks, since it is basically similar to the problem analysis method embodiments for query tasks, the description is relatively simple; relevant parts can be referred to in the description of the problem analysis method embodiments for query tasks.

[0087] Figure 4 This is a structural block diagram of a computing device according to an embodiment of this specification. The components of the computing device 400 include, but are not limited to, a memory 410 and a processor 420. The processor 420 is connected to the memory 410 via a bus 430, and a database 450 is used to store data.

[0088] The computing device 400 also includes an access device 440, which enables the computing device 400 to communicate via one or more networks 460. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 440 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0089] In one embodiment of this specification, the aforementioned components of the computing device 400 and Figure 4 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 4 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0090] The computing device 400 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 400 can also be a mobile or stationary server.

[0091] The processor 420 is used to execute computer programs / instructions, which, when executed by the processor, implement the above-mentioned... Figure 1 The method shown.

[0092] As the computing device embodiment is basically similar to the problem analysis method embodiment for query tasks, the description is relatively simple. For relevant details, please refer to the description of the problem analysis method embodiment for query tasks.

[0093] One embodiment of this specification also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of the problem analysis method for the query task described above. The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium may include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable storage medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium may not include electrical carrier signals and telecommunication signals.

[0094] One embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed in a processor, implement the steps of the problem analysis method for query tasks described above.

[0095] The computer-readable storage medium embodiment and the computer program product embodiment are described simply because they are basically similar to the problem analysis method embodiment for query tasks. For relevant details, please refer to the description of the problem analysis method embodiment for query tasks.

[0096] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0097] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0098] It should be noted that the above description describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of this specification.

[0099] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0100] The preferred embodiments disclosed above are merely illustrative of this specification. Optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described in this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification.

Claims

1. A problem analysis method for a query task, characterized in that: include: Mapping the intention analysis vector and the reference vector obtained by executing the query task to different mapping points in the first display area, respectively, wherein the intention analysis vector is determined by performing intention analysis on the query information corresponding to the query task, and the reference vector is determined by querying in the candidate data based on the intention analysis vector; In response to an execution parameter adjustment instruction for the query task, adjusting the execution parameter; Based on the updated intention analysis vector and the reference vector, updating the mapping points in the first display area, wherein the updated intention analysis vector and the reference vector are obtained by executing the query task based on the adjusted execution parameters; Based on the updated distribution information of the mapping points in the first display area, problem information existing in the execution process of the query task is determined.

2. The method according to claim 1, characterized in that The execution parameter includes a similarity threshold, and the reference vector is a vector whose similarity with the intention analysis vector is greater than the similarity threshold; The determining, based on the updated distribution information of the mapping points in the first display area, problem information existing in the execution process of the query task includes: When the mapping points in the first display area are concentrated in a target range after update by adjusting the similarity threshold, it is determined that the problem information existing in the execution process of the query task includes that the similarity threshold based on the query reference vector is set incorrectly; wherein, the target range includes the location of the mapping point corresponding to the intention analysis vector.

3. The method according to claim 1, characterized in that The execution parameters include a plurality of data attribute weights, wherein the data attribute weights represent the influence of the corresponding data attribute on the reference vector of the query; The determining, based on the updated distribution information of the mapping points in the first display area, problem information existing in the execution process of the query task includes: When the mapping points in the first display area are concentrated in a target range after updating by adjusting the data attribute weights, it is determined that the problem information existing in the execution process of the query task includes that the data attribute weights based on the query reference vector are incorrectly set; wherein, the target range includes the location of the mapping points corresponding to the intention analysis vector.

4. The method according to any one of claims 1 to 3, characterized in that: Also includes: Mapping a query vector obtained by performing vector conversion on the query information to a mapping point in the first display area; When the target distance in the first display area is greater than the distance threshold, the problem information existing in the execution process of the query task is determined, including: there is ambiguity in the query information, or there is a problem in the intent analysis processing process; wherein the target distance is the distance between the mapping point corresponding to the query vector and the mapping point corresponding to the intent analysis vector.

5. The method according to any one of claims 1 to 3, characterized in that: Before adjusting the execution parameter in response to the execution parameter adjustment instruction for the query task, the method further includes: Displaying in the second display area an adjustment control for the execution parameter of the query task; When the adjustment control is triggered, an execution parameter adjustment instruction for the query task is received.

6. The method according to any one of claims 1 to 3, characterized in that: The intention analysis vector and the reference vector obtained by executing the query task are mapped to different mapping points in the first display area, respectively, including: The intent analysis vector and the reference vector obtained by executing the query task are reduced in dimension into a two-dimensional vector; In the two-dimensional coordinate system in the first display area, the mapping points corresponding to each two-dimensional vector obtained by the dimensionality reduction processing are displayed; wherein, the two coordinate axes of the two-dimensional coordinate system respectively correspond to the two vector values ​​in the two-dimensional vector, and the mapping points corresponding to the intention analysis vector and the reference vector in the first display area are displayed differently.

7. The method according to any one of claims 1 to 3, characterized in that: Also includes: Based on different execution parameters and the corresponding distribution information of the mapping points, solution information for the problem information is determined.

8. A problem analysis device for a query task, characterized in that: include: A first mapping module, used to map the intention analysis vector and the reference vector obtained by executing the query task to different mapping points in the first display area, respectively, wherein the intention analysis vector is determined by performing intention analysis on the query information corresponding to the query task, and the reference vector is determined by querying in the candidate data based on the intention analysis vector; An adjustment module, configured to adjust the execution parameter in response to an execution parameter adjustment instruction for the query task; An updating module, configured to update mapping points in the first display area based on an updated intention analysis vector and a reference vector, wherein the updated intention analysis vector and the reference vector are obtained by executing the query task based on the adjusted execution parameters; A first determining module is configured to determine problem information existing in the execution process of the query task based on the updated distribution information of the mapping points in the first display area.

9. A computing device, characterized in that include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program / instruction is stored, and when the computer program / instruction is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

11. A computer program product, characterized in that The method comprises a computer program / instruction, which implements the method according to any one of claims 1 to 7 when the computer program / instruction is executed in a processor.

Images