Business query response method and device, equipment and storage medium

Abstract

The invention relates to the technical field of computers, in particular to a service query response method and device, equipment and a storage medium. The business query response method comprises the following steps: analyzing business query information by adopting a hierarchical intention recognition model, extracting key index data, and mapping report menu links of one or more visual indexes related to the business query information; obtaining identity information of a user, judging user permission according to the identity information, and filtering report menu links of the visual indexes according to the user permission to obtain permissible links; a link recommendation reinforcement learning model is adopted, high-priority links are screened out from the multiple permissible links, and the high-priority links are labeled; and selecting and rendering a matched UI component in interface display by adopting a structured data template. The method has high accuracy in business index identification during query, and can intelligently recommend links and intelligently present an efficient visual report entry.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a business query response method and device, equipment and storage medium. BACKGROUND

[0002] At present, intelligent dialogue systems are widely deployed in enterprises for business data query, and the existing technical solutions usually have the following inherent defects: 1. Data and view are separated, the system returns cold business values, but lacks an entry directly leading to the data visualization report, the user needs to manually navigate, the operation path is long, and the efficiency is low; 2. The granularity of intent recognition is rough, and a general intent recognition model is difficult to accurately distinguish hundreds of highly similar business indicators in an enterprise, resulting in misidentification or a flood of returned results; 3. Static coupling of permissions and links: the existing solution feeds the menu link as static knowledge to the model, which cannot real-time perceive the change of user permissions, resulting in returning no permission links, and interrupting the user experience; 4. Response structure is rigid: the returned values and links are simply text splicing, and cannot be dynamically and structurally optimally presented according to the query intent, user role, and data sensitivity. It can be seen that the existing technology still needs to be improved and improved. SUMMARY

[0003] In view of the above shortcomings of the prior art, the purpose of the present application is to provide a business query response method, device, equipment and storage medium, aiming at solving the technical problems of data and view separation, low data accuracy and link without permission in the prior art.

[0004] In order to achieve the above purpose, the present application adopts the following technical solutions: The first aspect of the present application provides a business query response method, comprising the following steps: obtaining business query information of a user, analyzing the business query information by using a hierarchical intent recognition model, extracting key indicator data, and mapping one or more report menu links of visual indicators related to the business query information; obtaining identity information of the user, judging the user's permission according to the identity information, and filtering the report menu links of visual indicators according to the user's permission to obtain the links with permission; constructing a link recommendation reinforcement learning model based on user behavior feedback, using the link recommendation reinforcement learning model to select high-priority links from the multiple links with permission, and labeling the high-priority links; using a structured data template, selecting and rendering matching UI components in the interface display according to the key indicator data and the type of the link with permission.

[0005] Optionally, in the first implementation manner of the first aspect of the present application, the business query information of the user is acquired, the hierarchical intention recognition model is used to analyze the business query information, key index data is extracted, and one or more report menu links of visual indicators related to the business query information are mapped, specifically including: a hierarchical intention recognition model is constructed, a BERT-like encoder is set to extract general semantic features in the business query information; a fully connected layer and a Bi-LSTM-CRF layer are connected in parallel on the basis of the BERT-like encoder, respectively used to output the probability distribution of the macro business domain and identify the indicators in the business query information, a dynamic configuration system is set to map the identified indicators and the corresponding report menu links of visual indicators, and the hierarchical intention recognition model is obtained; the business query information of the user is acquired, the hierarchical intention recognition model is used to analyze the business query information, key index data is extracted, and one or more report menu links of visual indicators related to the business query information are mapped.

[0006] Optionally, in the second implementation manner of the first aspect of the present application, the identity information of the user is acquired, the user permission is determined according to the identity information, the report menu links of visual indicators are filtered according to the user permission, and the permission link is obtained, specifically including: the identity information of the user is acquired, the identity information including the user ID, the department to which the user belongs, the role permission level, etc.; the user permission is determined according to the preset permission rule and the identity information of the user, and the permission result is output; the report menu links of visual indicators are filtered according to the permission result, and the permission link is obtained.

[0007] Optionally, in the third implementation manner of the first aspect of the present application, the report menu links of visual indicators are filtered according to the permission result, and the permission link is obtained, specifically including: the permission result is acquired, the permission result including the viewing permission and the export permission; the report menu links of visual indicators are filtered according to the permission result, the permission link is obtained, the permission link is classified according to types, and each category is labeled; the links without permission are summarized, and the report related information of the links is labeled respectively, and the permission application portal of the links without permission is generated.

[0008] Optionally, in the fourth implementation form of the first aspect of the present application, the link recommendation reinforcement learning model based on user behavior feedback is constructed, the high-priority links are filtered from the multiple authorized links by using the link recommendation reinforcement learning model, and the high-priority links are marked, and the method specifically comprises: constructing the link recommendation reinforcement learning model based on user behavior feedback, the link recommendation reinforcement learning model continuously optimizes the recommendation strategy according to the user behavior feedback data; filtering the high-priority links from the multiple authorized links according to the user's business query information, identity information and historical behavior data by using the link recommendation reinforcement learning model; marking the high-priority links and folding the remaining links.

[0009] Optionally, in the fifth implementation form of the first aspect of the present application, the link recommendation reinforcement learning model based on user behavior feedback is constructed, and the link recommendation reinforcement learning model continuously optimizes the recommendation strategy according to the user behavior feedback data, and the method specifically comprises: obtaining the identity information and historical query data of multiple users, pre-processing the historical query data, and collecting the pre-processed data and the identity information to obtain a training set; taking the Thompson Sampling or UCB algorithm as a basic model, training the basic model by using the training set to obtain the link recommendation reinforcement learning model; and obtaining the user behavior feedback data to iteratively optimize the link recommendation reinforcement learning model.

[0010] Optionally, in the sixth implementation form of the first aspect of the present application, the structured data template is used to select and render the matched UI components in the interface display according to the key indicator data and the types of the authorized links, and the method specifically comprises: classifying the data and the links, setting the display rules and the interaction modes, constructing the structured data template according to the types of the data and the links, the display rules and the interaction modes; using the structured data template to select and render the matched UI components in the interface display according to the key indicator data and the types of the authorized links; and highlighting the key indicator data according to the display rules, and embedding a micro trend chart in the interface for the indicators that need to show the change trend.

[0011] The second aspect of the present application provides a service query response device, comprising: a mapping module configured to obtain service query information of a user, analyze the service query information using a hierarchical intention recognition model, extract key indicator data, and map one or more report menu links of visual indicators related to the service query information; a permission module configured to obtain identity information of the user, determine user permissions according to the identity information, and filter the report menu links of visual indicators according to the user permissions to obtain links with permissions; a recommendation module configured to construct a link recommendation reinforcement learning model based on user behavior feedback, select high-priority links from the links with permissions using the link recommendation reinforcement learning model, and label the high-priority links; and a display module configured to select and render matching UI components in interface display according to the key indicator data and the types of the links with permissions using a structured data template.

[0012] Optionally, in the first implementation manner of the second aspect of the present application, the mapping module comprises: a first setting unit configured to construct the hierarchical intention recognition model, set a BERT-like encoder to extract general semantic features in the service query information; a second setting unit configured to connect a fully connected layer and a Bi-LSTM-CRF layer in parallel on the basis of the BERT-like encoder to respectively output probability distribution of a macro business domain and identify indicators in the service query information, and set a dynamic configuration system to map the identified indicators to corresponding report menu links of visual indicators to obtain the hierarchical intention recognition model; and an analysis unit configured to obtain the service query information of the user, analyze the service query information using the hierarchical intention recognition model, extract the key indicator data, and map one or more report menu links of visual indicators related to the service query information.

[0013] Optionally, in the second implementation manner of the second aspect of the present application, the permission module comprises: an information acquisition unit configured to obtain identity information of the user, the identity information comprising a user ID, a department to which the user belongs, a role permission level, and the like; a judgment unit configured to determine user permissions according to a preset permission rule and the identity information of the user, and output a permission result; and a filtering unit configured to filter the report menu links of visual indicators according to the permission result to obtain links with permissions.

[0014] Optionally, in a third implementation form of the second aspect of the present application, the filtering unit comprises: an obtaining subunit, configured to obtain a permission result, the permission result comprising a viewing permission and an exporting permission; a processing subunit, configured to filter report menu links of the visualized index according to the permission result, to obtain links with permission, to classify the links with permission by type, and to label each category; and a generating subunit, configured to summarize links without permission, to label report related information of the links respectively, and to generate a permission application portal for the links without permission.

[0015] Optionally, in a fourth implementation form of the second aspect of the present application, the recommendation module comprises: a model constructing unit, configured to construct a link recommendation reinforcement learning model based on user behavior feedback, the link recommendation reinforcement learning model being capable of continuously optimizing a recommendation strategy according to user behavior feedback data; a screening unit, configured to screen high-priority links from a plurality of links with permission according to user business query information, identity information and historical behavior data by using the link recommendation reinforcement learning model; and a labeling unit, configured to label the high-priority links and to fold display the remaining links.

[0016] Optionally, in a fifth implementation form of the second aspect of the present application, the model constructing unit comprises: a preprocessing subunit, configured to obtain identity information and historical query data of a plurality of users, to pre-process the historical query data, to summarize the pre-processed data and the identity information, and to obtain a training set; a training subunit, configured to take a Thompson Sampling or UCB algorithm as a base model, to train the base model by using the training set, and to obtain the link recommendation reinforcement learning model; and an optimization subunit, configured to obtain user behavior feedback data, and to iteratively optimize the link recommendation reinforcement learning model.

[0017] Optionally, in a sixth implementation form of the second aspect of the present application, the display module comprises: a template constructing unit, configured to classify data and links, to set display rules and interaction modes, and to construct a structured data template according to types of the data and the links, the display rules and the interaction modes; a rendering unit, configured to select and render matching UI components in an interface display according to key index data and types of links with permission by using the structured data template; and a display unit, configured to highlight the key index data when displaying according to the display rules, and to embed a micro trend chart in the interface for an index that needs to display a trend.

[0018] The third aspect of the present application provides a business query response device, comprising a memory and at least one processor, the memory storing computer readable instructions; the at least one processor invokes the computer readable instructions in the memory to perform each step of the business query response method as described above.

[0019] The fourth aspect of the present application provides a computer readable storage medium, which stores computer readable instructions, and the computer readable instructions are executed by a processor to implement each step of the business query response method.

[0020] Beneficial effects: The present application provides a business query response method, which first analyzes the business query information by using a hierarchical intention recognition model, extracts key indicator data, and maps one or more report menu links of visual indicators related to the business query information, improves the accuracy and granularity of business indicator recognition, effectively distinguishes highly similar queries, and ensures the accuracy of response content from the source; then, according to the user's identity information, the user's authority is judged, and the report menu link of the visual indicator is filtered according to the user's authority to obtain the authorized link, avoiding returning the unauthorized link; then, a link recommendation reinforcement learning model based on user behavior feedback is constructed, and then the link recommendation reinforcement learning model is used to filter out high-priority links from multiple authorized links, and the user role, historical behavior is used to optimize the push strategy in real time; finally, by using a structured data template, according to the key indicator data and the type of the authorized link, the matching UI component is selected and rendered in the interface display, and the most efficient visual report entry is intelligently presented, which completely liberates the user from manual searching. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The first flowchart of the business query response method provided by the embodiment of the present application; Figure 2 The second flowchart of the business query response method provided by the embodiment of the present application; Figure 3 The third flowchart of the business query response method provided by the embodiment of the present application; Figure 4 The fourth flowchart of the business query response method provided by the embodiment of the present application; Figure 5 The fifth flowchart of the business query response method provided by the embodiment of the present application; Figure 6 The sixth flowchart of the business query response method provided by the embodiment of the present application; Figure 7 The seventh flowchart of the business query response method provided by the embodiment of the present application; Figure 8 A structural schematic diagram of the business query response device provided by the embodiment of the present application; Figure 9 Another structural schematic diagram of the business query response device provided by the embodiment of the present application; Figure 10 A structural schematic diagram of a service query response device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0022] The present application provides a service query response method, device, equipment and storage medium, the present application first analyzes and processes the service query information through the hierarchical intention recognition model, greatly improves the accuracy and granularity of the service index recognition, effectively distinguishes highly similar queries, and guarantees the accuracy of the response content from the source; then the user's authority is judged according to the identity information, and the report menu link of the visual index is filtered according to the user's authority to obtain the link with authority; then the link recommendation reinforcement learning model based on user behavior feedback is adopted, a dynamic link recommendation mechanism based on reinforcement learning is introduced, the system can optimize the push strategy according to the user role and historical behavior in real time, and finally the most efficient visual report portal is intelligently presented by adopting a structured data template, and the user is completely liberated from manual searching, and the decision efficiency is doubled.

[0023] The terms "first", "second", "third", "fourth" and the like in the description, claims, and drawings of the present application, and those above and herein (if any) are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of such terms, if any, are not construed to limit the scope of the application, unless specified otherwise. The terminology of "comprising", "having", "including", and "containing" refers to the open-ended inclusion of the stated elements, and does not exclude the presence of additional elements, or the possibility of additional elements, unless otherwise specified. The terminology of "consisting of" refers to the closed set of elements, and excludes the presence of additional elements, unless otherwise specified.

[0024] For the sake of understanding, the specific flow of the embodiment of the present application is described below, please refer to Figure 1 The first embodiment of the service query response method in the embodiment of the present application comprises: S101. Obtain the service query information of the user, analyze the service query information by using the hierarchical intention recognition model, extract the key index data, and map one or more report menu links of the visual index related to the service query information; S102. Obtain the identity information of the user, judge the user's authority according to the identity information, and filter the report menu link of the visual index according to the user's authority to obtain the link with authority; S103. Construct a link recommendation reinforcement learning model based on user behavior feedback, and select a high-priority link from the multiple links with authority by using the link recommendation reinforcement learning model, and mark the high-priority link; S104. Adopting a structured data template, according to key indicator data and types of authorized links, selecting and rendering matching UI components in the interface display.

[0025] In this embodiment, a hierarchical intent recognition model based on multi-task learning is first constructed, and the hierarchical intent recognition model has a multi-level structure, wherein the first layer is used for identifying a user macro business domain; the second layer accurately identifies a specific business indicator under the business domain; and the third layer finally maps to one or more report menu links for visualizing the indicator through a dynamic configuration system according to the identified indicator, thereby solving the problem of coarse granularity of intent recognition, effectively distinguishing highly similar queries, and ensuring the accuracy of response content from the source.

[0026] To solve the problem that the existing scheme often returns no permission links, the identity information of the user is obtained in real time in this embodiment, and the user's permission is judged according to the latest identity information, so that it is determined which report menu links the user has the right to view, thereby avoiding invalid output. Specifically, the user's permission can be obtained by asynchronously calling the permission center interface, and it is determined whether the user's permission is applicable to each report menu link. This asynchronous processing method can avoid blocking the main process and ensure the smoothness of the user's operation experience.

[0027] After ensuring data security and permission compliance, a link recommendation reinforcement learning model based on user behavior feedback is established in this embodiment, a link optimization strategy is introduced, the most likely high-priority link that meets the user's intent is intelligently recommended according to the context of this query, and the link is highlighted, and the remaining links are folded and displayed. The link recommendation reinforcement learning model based on user behavior feedback can learn the long-term revenue maximization strategy of link recommendation in different scenarios through continuous interaction with real users. For example, it will learn to prioritize the provincial general driving cockpit link when querying the signing rate for the provincial general role, and recommend the branch detail analysis link for the branch manager.

[0028] Finally, in order to optimize the front-end display effect and information transmission efficiency, the system of this embodiment adopts a structured data template design, which supports the front-end client to adaptively render the optimal UI component according to the data type. The structured data template will standardize the data returned to the front-end, ensuring that the front-end can quickly identify and accurately process. This adaptive rendering method not only ensures the consistency of display on different terminals (such as PC, tablet), but also maximizes the information transmission efficiency according to the data characteristics, allowing users to quickly obtain key content in a simple interface, significantly improving the user experience.

[0029] Please refer to Figure 2 The second embodiment of the business query response method in the embodiment of the present application comprises: S201. Construct a hierarchical intent recognition model, set a BERT-like encoder to extract general semantic features in business query information; S202. On the basis of the BERT-like encoder, connect a fully connected layer and a Bi-LSTM-CRF layer in parallel to respectively output the probability distribution of the macro business domain and identify the indicators in the business query information, set a dynamic configuration system to map the identified indicators and the report menu links of the corresponding visual indicators, and obtain a hierarchical intent recognition model; S203. Obtain the business query information of the user, analyze the business query information by using the hierarchical intent recognition model, extract the key indicator data, and map one or more report menu links of the visual indicators related to the business query information.

[0030] In this embodiment, the hierarchical intent recognition model based on multi-task learning uses a shared BERT-like encoder to extract general semantic features of the query, and then connects two specific task output heads in parallel: the two tasks share the underlying features and promote each other, which not only ensures the accuracy of the business domain classification, but also finely identifies the key indicator entities, solving the problem of similar indicators being easily confused. Task one is to output the probability distribution of the macro business domain: a fully connected layer is used to output the probability distribution of the macro business domain (such as cockpit, service quality, cost) using Softmax. Task two is to identify the indicators in the business query information: a Bi-LSTM-CRF layer is used to identify the specific business indicator keywords in the query (such as signing rate, 1D2359 is regarded as an entity for extraction).

[0031] Specifically, the hierarchical intent recognition model based on multi-task learning constructed by the system can deeply extract general semantic features of the query sentence by using a shared BERT-like encoder, avoiding repeated feature calculation; meanwhile, two special task output heads are connected in parallel to realize feature sharing and mutual promotion between tasks: task one (business domain classification) outputs the probability distribution of each macro business domain through a fully connected layer and a Softmax function, ensuring the accuracy of business domain judgment; task two (indicator sequence labeling) uses a Bi-LSTM-CRF layer to accurately identify the indicator keywords in the query, even for similar indicators such as 1D2359 and 2D signing rate, efficient entity extraction can be achieved, and the industry pain point of similar indicators being difficult to distinguish is completely solved.

[0032] The hierarchical intention recognition model realizes intention penetration through macro-meso-micro three-layer progressive recognition logic. The first layer focuses on the macro business domain positioning of the user query, for example, judging whether the user demand belongs to the cockpit data viewing, service quality analysis or cost control category, to determine the scope for subsequent recognition; the second layer further accurately extracts specific business indicators within the determined business domain, such as locking key indicators such as direct connection rate and complaint resolution rate in the service quality domain; the third layer establishes a mapping between the recognized indicators and the corresponding visual report menu links through dynamic configuration of the system, realizing the direct association of intention to resources.

[0033] Referring to Figure 3 The third embodiment of the business query response method in the embodiment of the application comprises: S301. Obtain the identity information of the user, including user ID, department affiliation, role permission level, etc. S302. Determine the user's permissions according to the pre-set permission rules and the user's identity information, and output the permission result. S303. Filter the report menu links of the visual indicators according to the permission result to obtain the links with permission.

[0034] In this embodiment, to ensure data security and compliance with permissions, the system adds a dynamic permission verification link before returning the report link to the user. After the system matches the corresponding report link through the hierarchical intention recognition model, it does not directly return the result, but first obtains the core identity information of the user (including user ID, department affiliation, role permission level, etc.) and interfaces with the permission center. The permission center interface will verify in real time whether the user currently has the permission to access the target report system according to the pre-set permission rules (such as department data isolation and role permission range limitation); after the verification is completed, the permission center will feed back the result to the system, and the system will only filter out the links that pass the verification and return them to the user. For links without permission to access, the system will directly filter them or only provide a list of links without permission to access, effectively preventing unauthorized access to sensitive data and complying with the requirements of enterprise data security management.

[0035] Referring to Figure 4 The fourth embodiment of the business query response method in the embodiment of the application comprises: S401. Obtain the permission result, which includes viewing permissions and export permissions. S402. Filter the report menu links of the visual indicators according to the permission result to obtain the links with permission, classify the links with permission by type, and label each category. S403. Aggregate the links without permission and label the report-related information of the links, and generate a permission application portal for the links without permission.

[0036] In this embodiment, the authority center will give different report links different explicit permissions according to the regulations. For example, some report links only have the permission of viewing, and some report links have the permission of viewing and exporting. After filtering according to the permission results, this embodiment further classifies and labels the filtered links to facilitate users to find the target content. For example, when the user's business query information matches a large number of links, the user will have difficulty in viewing each link in an orderly manner when facing the link list. By classifying a large number of links, such as classifying according to business types, classifying according to business levels, and classifying according to business regions, when the user wants to view the report of each province, the user can quickly find each link according to the category. In some other working scenarios, the user may also need to access some report data without permission, which involves the process of permission application. In order to improve the convenience of application, this embodiment simply identifies the report menu links without permission access, and provides a permission application entrance. After the user clicks into it, the corresponding process can be initiated, and the relevant report data content is automatically specified in the process, without the need for the user to fill in and apply separately through other systems.

[0037] Please refer to Figure 5 The fifth embodiment of the business query response method in the embodiment of the application includes the following steps: S501. A link recommendation reinforcement learning model based on user behavior feedback is constructed, and the link recommendation reinforcement learning model continuously optimizes the recommendation strategy according to the user's behavior feedback data; S502. The link recommendation reinforcement learning model is used to filter high-priority links from a plurality of links with permission according to the user's business query information, identity information and historical behavior data; S503. The high-priority links are labeled, and the remaining links are folded and displayed.

[0038] In this embodiment, to further improve the efficiency of users obtaining target reports, the system establishes a link recommendation reinforcement learning model based on user behavior feedback, and achieves personalized link ranking through an intelligent optimization strategy. The core logic of this model is to continuously interact with real users, collect user behavior feedback data (such as the number of link clicks, dwell time, whether the user accesses the report through the link, etc.), continuously optimize the recommendation strategy, and ultimately maximize long-term benefits, making the recommended links more in line with the user's real needs. In specific recommendations, the model will comprehensively consider multi-dimensional query context information: for example: 1. Considering role differences, for provincial managers who are more concerned with the overall data of the province, the provincial manager dashboard link is recommended first; for branch managers, who need to focus on the details of a single branch, the branch detailed analysis link is recommended; 2. Considering query keywords, if the keywords contain detailed analysis, the detailed report link is recommended; if the keywords contain overall overview, the summary report link is recommended; 3. Considering the historical click data of the indicator, links that users have clicked frequently in the past will receive higher weight. Based on this information, the model will select 1-2 high-priority links that are most likely to match the user's intent and highlight them with eye-catching labels (such as special color buttons or recommendation tags). The remaining matching links will be collapsed, which simplifies the interface display and reduces the user's search cost.

[0039] Please see Figure 6 The sixth embodiment of the business query response method in this invention includes: S601. Obtain the identity information and historical query data of multiple users, preprocess the historical query data, and summarize the preprocessed data and identity information to obtain the training set; S602. Using Thompson Sampling or UCB algorithm as the base model, train the base model with the training set to obtain the link recommendation reinforcement learning model; S603. Obtain user behavior feedback data and iteratively optimize the link recommendation reinforcement learning model.

[0040] In this embodiment, to avoid falling into local optima and continuously improve recommendation performance, the link recommendation reinforcement learning model adopts Thompson Sampling or UCB algorithm to achieve a dynamic balance between exploration (trying to recommend new potential matching links and uncovering unmet user needs) and utilization (prioritizing links with good historical click performance to ensure current recommendation accuracy).

[0041] Specifically, before constructing the training set, user identity information needs to be collected: for example, user A (a sales manager, with permission to include regional sales data), user B (a financial clerk, with permission to include cost accounting data), and user C (a human resources director, with permission to include employee turnover rate data); historical query data also needs to be collected: for example, user A has queried "Q3 sales in the East China region" and clicked on the "East China sales report" link; user B has queried "raw material cost changes" and clicked on the "cost trend report" link. Then the historical data is preprocessed, and the preprocessing operations include: removing invalid queries (such as repeated meaningless statements), labeling "click / non-click behavior" (such as "click = 1" and "non-click = 0"), and then aggregating with user identity information (department, permission) to form the training set.

[0042] By training the model using the above constructed training set, the model can autonomously learn the recommendation rules based on historical data without manually defining complex rules, and learn the association between "user identity, query intent" and "link click effect".

[0043] To keep the recommended content of the model up-to-date, the embodiment also dynamically optimizes the model in real time based on user behavior, so that the recommendation effect improves over time and the model is not fixed. For example: when user D (a sales clerk) queries "Q4 sales in the South China region", the model first recommends the "South China sales report" that is often clicked, and also tries to recommend a newly online "South China sales comparison report" (to explore potential demand). If user D clicks on the new report, the model will record this feedback, and the weight of recommending the new report to this user will be increased in the future; if not, the weight will be reduced. By continuously collecting user click and non-click behavior, the model continuously optimizes the recommendation strategy.

[0044] Referring to Figure 7 The seventh embodiment of the business query response method in the embodiment of the application comprises: S701. Classify data and links, and set display rules and interaction modes, and construct a structured data template according to the types of data and links, display rules and interaction modes; S702. Using the structured data template, select and render matching UI components in the interface display according to the types of key indicator data and links with permission; S703. According to the display rules, highlight the key indicator data when displaying, and for indicators that need to show the trend of change, embed a micro trend chart in the interface.

[0045] In this embodiment, in order to optimize the front-end display effect and information transmission efficiency, the system adopts a structured data template design, and supports the front-end client to adaptively render the optimal UI component according to the data type. The structured data template defines the data returned to the front-end in a standardized manner, clearly defines the type of data (such as numerical type, link type, trend type, etc.), display rules (such as numerical precision, color identification logic, etc.), and interaction mode, to ensure that the front-end can quickly identify and accurately process.

[0046] In actual rendering, the front-end will automatically select the adaptive UI component according to the template definition, such as: for key numerical data (such as direct connection rate 98.5%), highlight color, bold font, etc. Style highlights allow users to quickly capture core information; for report link data, a beautiful button group is used to present, and the priority is distinguished by the difference in button style (such as using a primary color button for a primary recommended link, and using a secondary color button for a normal link), to improve the convenience of operation; for indicators that need to show the trend of change (such as the signing rate in the past 7 days), a micro trend chart (such as a line chart thumbnail, a column chart thumbnail) is directly embedded in the interface, so that users can intuitively understand the fluctuation of the indicator without jumping to the report page. This adaptive rendering method not only ensures the consistency of display on different terminals (such as PC and tablet terminals), but also maximizes the information transmission efficiency according to the data characteristics, allowing users to quickly obtain key content in a simple interface, significantly improving the user experience.

[0047] The above describes the business query response method in the embodiment of the application, and the following describes the business query response device in the embodiment of the application, please refer to Figure 8 An embodiment of the business query response device in the embodiment of the application includes: The mapping module 10 is configured to obtain the business query information of the user, analyze the business query information by using a hierarchical intention recognition model, extract key indicator data, and map one or more report menu links of visual indicators related to the business query information; The permission module 20 is configured to obtain the identity information of the user, determine the user permission according to the identity information, filter the report menu links of the visual indicators according to the user permission, and obtain the links with permission; The recommendation module 30 is configured to construct a link recommendation reinforcement learning model based on user behavior feedback, filter high-priority links from the multiple links with permission by using the link recommendation reinforcement learning model, and mark the high-priority links; The display module 40 is configured to use a structured data template, select and render matching UI components in the interface display according to the types of the key indicator data and the links with permission.

[0048] Please refer to Figure 9The business query response device in the embodiment of the application comprises: The mapping module 10 is configured to obtain the business query information of the user, analyze the business query information by using a hierarchical intention recognition model, extract key indicator data, and map one or more report menu links of visual indicators related to the business query information; The permission module 20 is configured to obtain the identity information of the user, determine the user permission according to the identity information, and filter the report menu links of the visual indicators according to the user permission to obtain the links with permission; The recommendation module 30 is configured to construct a link recommendation reinforcement learning model based on user behavior feedback, filter high-priority links from the links with permission by using the link recommendation reinforcement learning model, and label the high-priority links; The display module 40 is configured to select and render a matching UI component in the interface display according to the key indicator data and the type of the links with permission by using a structured data template; In the embodiment, the mapping module 10 comprises: The first setting unit 11 is configured to construct a hierarchical intention recognition model and set a BERT-like encoder to extract general semantic features in the business query information; The second setting unit 12 is configured to connect a fully connected layer and a Bi-LSTM-CRF layer in parallel on the basis of the BERT-like encoder, so as to respectively output the probability distribution of the macro business domain and identify the indicators in the business query information, and set a dynamic configuration system to map the identified indicators to the report menu links of the corresponding visual indicators to obtain the hierarchical intention recognition model; The analysis unit 13 is configured to obtain the business query information of the user, analyze the business query information by using the hierarchical intention recognition model, extract key indicator data, and map one or more report menu links of visual indicators related to the business query information; In the embodiment, the permission module 20 comprises: The information acquisition unit 21 is configured to obtain the identity information of the user, wherein the identity information comprises a user ID, a department to which the user belongs, and a role permission level; The judgment unit 22 is configured to determine the user permission according to a preset permission rule and the identity information of the user, and output a permission result; The filtering unit 23 is configured to filter the report menu links of the visual indicators according to the permission result to obtain the links with permission; In the embodiment, the filtering unit 23 comprises: The acquisition subunit 231 is configured to obtain the permission result, wherein the permission result comprises a viewing permission and an export permission; The processing subunit 232 is configured to filter the report menu links of the visualization indicators according to the permission result, obtain the links with permission, classify the links with permission by type, and label each category; The generating subunit 233 is configured to summarize the links without permission, label the report related information of the links respectively, and generate a permission application portal for the links without permission; In this embodiment, the recommendation module 30 comprises: The model construction unit 31 is configured to construct a link recommendation reinforcement learning model based on user behavior feedback, and the link recommendation reinforcement learning model continuously optimizes the recommendation strategy according to user behavior feedback data; The screening unit 32 is configured to use the link recommendation reinforcement learning model to screen high-priority links from multiple links with permission according to user business query information, identity information, and historical behavior data; The labeling unit 33 is configured to label the high-priority links and fold the remaining links for display; In this embodiment, the model construction unit 31 comprises: The preprocessing subunit 311 is configured to obtain identity information and historical query data of multiple users, pre-process the historical query data, and aggregate the pre-processed data and the identity information to obtain a training set; The training subunit 312 is configured to use the training set to train a base model based on a Thompson Sampling or UCB algorithm to obtain the link recommendation reinforcement learning model; The optimization subunit 313 is configured to obtain user behavior feedback data and iteratively optimize the link recommendation reinforcement learning model; In this embodiment, the display module 40 comprises: The template construction unit 41 is configured to classify data and links, set display rules and interaction modes, and construct a structured data template according to the types of data and links, the display rules, and the interaction modes; The rendering unit 42 is configured to use the structured data template to select and render matching UI components in the interface display according to key indicator data and the types of links with permission; The display unit 43 is configured to highlight the key indicator data according to the display rules, and embed a micro trend chart in the interface for indicators that need to display trend changes.

[0049] The business query response device of the present application first analyzes the business query information through a hierarchical intention recognition model, extracts key indicator data, and maps one or more visual indicator report menu links related to the query, thereby improving the accuracy and precision of business indicator identification, accurately distinguishing highly similar queries, and ensuring the accuracy of the response content from the source; then, according to the user's identity information, the permissions are judged, and the visual indicator report menu links are filtered according to the permissions, and the links that the user has access to are selected to avoid returning links without permissions; then, a link recommendation reinforcement learning model based on user behavior feedback is constructed, and the model is used to further filter high-priority links from multiple authorized links, and the user role and historical behavior are used to optimize the push strategy in real time; finally, a structured data template is used, and according to the key indicator data and the type of authorized links, the matching UI component is selected and rendered when the interface is displayed, and the most efficient visual report entry is intelligently presented, thereby completely liberating the user from the tedious operation of manual searching. This process realizes accurate matching through hierarchical identification, ensures access effectiveness through permission filtering, dynamically optimizes recommendation priority through reinforcement learning, and intelligently adapts presentation form through structured templates, which comprehensively improves the accuracy, security and user experience of business query response, and significantly shortens the path for users to obtain target information.

[0050] The above is a detailed description of the business query response device in the embodiment of the present application from the perspective of modular functional entities. The business query response device in the embodiment of the present application is described in detail from the perspective of hardware processing.

[0051] Figure 10 A structural schematic diagram of a business query response device provided by the embodiment of the present application is shown. The business query response device 900 can have great differences due to different configurations or performances, and can include one or more central processing units (CPUs) 910 (for example, one or more processors) and a memory 920, and one or more storage media 930 (for example, one or more mass storage devices) storing application programs 933 or data 932. The memory 920 and the storage medium 930 can be temporary storage or persistent storage. The programs stored in the storage medium 930 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations in the business query response device 900. Further, the processor 910 can be configured to communicate with the storage medium 930, execute a series of instruction operations in the storage medium 930 on the business query response device 900, to realize the steps of the business query response method provided by each method embodiment described above.

[0052] The service query responding device 900 can also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that, Figure 10 The illustrated service query responding device structure does not constitute a limitation on the service query responding device, and can include more or fewer components than illustrated, or combine certain components, or arrange the components differently.

[0053] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium, or a volatile computer readable storage medium. The computer readable storage medium stores instructions, which, when executed on a computer, cause the computer to perform the steps of the service query responding method.

[0054] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device or apparatus can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0055] The integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0056] It can be understood that, for those skilled in the art, the technical solutions of the present application and the inventive concept can be replaced or changed equivalently, and all these changes or replacements shall belong to the protection scope of the claims appended to the present application.

Claims

1. A business query response method, characterized in that, Includes the following steps: Obtain user business query information, analyze the business query information using a hierarchical intent recognition model, extract key indicator data, and map one or more report menu links related to the business query information. Obtain the user's identity information, determine the user's permissions based on the identity information, and filter the report menu links of the visualization indicators according to the user's permissions to obtain the links that the user is authorized to access. A link recommendation reinforcement learning model based on user behavior feedback is constructed. The link recommendation reinforcement learning model is used to filter out high-priority links from multiple authorized links and to label the high-priority links. Using a structured data template, the interface displays and renders matching UI components based on key metric data and the type of authorized links.

2. The business query response method according to claim 1, characterized in that, The process of obtaining user business query information involves analyzing the information using a hierarchical intent recognition model, extracting key indicator data, and mapping one or more report menu links for visual indicators related to the business query information. Specifically, this includes: Construct a hierarchical intent recognition model and set up a BERT-like encoder to extract common semantic features from business query information; A fully connected layer and a Bi-LSTM-CRF layer are connected in parallel on the basis of a BERT-like encoder to output the probability distribution of the macro business domain and identify the indicators in the business query information, respectively. A dynamic configuration system is set up to link the identified indicators with the corresponding visualization indicator report menu to establish a mapping and obtain a hierarchical intent recognition model. The system acquires users' business query information, analyzes the information using a hierarchical intent recognition model, extracts key indicator data, and maps one or more report menu links to visual indicators related to the business query information.

3. The business query response method according to claim 1, characterized in that, The process of obtaining user identity information, determining user permissions based on the identity information, and filtering the report menu links of the visualization metrics according to user permissions to obtain authorized links specifically includes: Obtain the user's identity information, which includes user ID, department, and role / permission level; Based on the preset permission rules and the user's identity information, determine the user's permissions and output the permission results; The report menu links for visualization metrics are filtered based on the permission results to obtain the links that the user has permission to access.

4. The business query response method according to claim 3, characterized in that, The step of filtering the report menu links of the visualization indicators based on the permission results to obtain the links with the appropriate permissions specifically includes: Obtain permission results, which include view permissions and export permissions; Based on the permission results, the report menu links of the visualization indicators are filtered to obtain the authorized links. The authorized links are then categorized by type, and each category is labeled. Summarize the links that do not have permission, and annotate the report information for each link. Generate permission request entry points for the links that do not have permission.

5. The business query response method according to claim 1, characterized in that, The construction of a link recommendation reinforcement learning model based on user behavior feedback involves using this model to filter out high-priority links from multiple authorized links and then labeling these high-priority links. Specifically, this includes: A link recommendation reinforcement learning model based on user behavior feedback is constructed, and the link recommendation reinforcement learning model continuously optimizes the recommendation strategy based on user behavior feedback data; A link recommendation reinforcement learning model is adopted to filter high-priority links from multiple authorized links based on users' business query information, identity information, and historical behavior data. High-priority links are highlighted, and the remaining links are collapsed.

6. The business query response method according to claim 5, characterized in that, The construction of a link recommendation reinforcement learning model based on user behavior feedback, wherein the link recommendation reinforcement learning model continuously optimizes the recommendation strategy based on user behavior feedback data, specifically includes: Obtain the identity information and historical query data of multiple users, preprocess the historical query data, and summarize the preprocessed data and identity information to obtain the training set; Using Thompson Sampling or UCB algorithm as the base model, the base model is trained with a training set to obtain a link recommendation reinforcement learning model; Obtain user behavior feedback data and iteratively optimize the link recommendation reinforcement learning model.

7. The business query response method according to claim 1, characterized in that, The structured data template is used to select and render matching UI components in the interface display based on key indicator data and the type of authorized links. Specifically, this includes: Classify the data and links, set display rules and interaction methods, and build structured data templates based on the data and link types, display rules and interaction methods; Using structured data templates, the interface displays and renders matching UI components based on key indicator data and the type of authorized links. According to the display rules, key indicator data should be highlighted during the display, and for indicators that need to show the trend of change, a mini trend chart should be embedded in the interface.

8. A business query response device, characterized in that, include: The mapping module is used to obtain users' business query information, analyze the business query information using a hierarchical intent recognition model, extract key indicator data, and map one or more report menu links of visual indicators related to the business query information. The permissions module is used to obtain the user's identity information, determine the user's permissions based on the identity information, and filter the report menu links of the visualization indicators according to the user's permissions to obtain the links that the user is authorized to access. The recommendation module is used to build a link recommendation reinforcement learning model based on user behavior feedback. The link recommendation reinforcement learning model is used to filter out high-priority links from multiple authorized links and to label the high-priority links. The display module is used to select and render matching UI components in the interface display based on structured data templates, key indicator data, and the type of authorized links.

9. A business query response device, characterized in that, It includes a memory and at least one processor, wherein the memory stores computer-readable instructions; The at least one processor invokes the computer-readable instructions in the memory to perform the steps of the business query response method as described in any one of claims 1-7.

10. A computer-readable storage medium storing computer-readable instructions thereon, characterized in that, When the computer-readable instructions are executed by a processor, they implement the steps of the business query response method as described in any one of claims 1-7.

Images