Method, apparatus and medium for determining structured query statement

Abstract

Embodiments of the present application disclose a structured query statement determination method and device, equipment and medium, and relate to the technical field of artificial intelligence. The method comprises: performing intent recognition on an initial language text to obtain a target intent; determining a reference business field group according to a candidate business field group in a target business table corresponding to the initial language text and the target intent; when the number of the reference business field group is greater than or equal to a preset number threshold, determining a business field confidence according to the initial language text, the reference business field group and a preset field quality dimension, and determining a target business field group from the reference business field group according to the business field confidence; inputting a target business field name, the initial language text and a target business table name into a text rewriting model to obtain a target language text; and determining a target structured query statement according to the target language text. The accuracy of the structured query statement determined based on the rewritten text is improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of artificial intelligence technology, and in particular to a method, apparatus, device and medium for determining structured query statements. Background Technology

[0002] Natural language text to structured query statements is a type of semantic parsing task that aims to convert user-input natural language text into structured query statements that can be manipulated by databases. Its core objective is to lower the technical threshold for database queries and enable natural interaction between non-technical personnel and databases.

[0003] Current technologies typically involve directly inputting natural language text into a trained neural network model to obtain structured query statements. However, this method suffers from low accuracy in determining the structured query statements because the natural language text input by the user may have irregular formats and unclear structures. Therefore, improving the accuracy of the determined structured query statements is crucial. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for determining structured query statements, so as to rewrite natural language text input by users, making the rewritten text formatted and structured clearly, thereby improving the accuracy of structured query statements determined based on the rewritten text.

[0005] According to one aspect of the present invention, a method for determining a structured query statement is provided, comprising: Obtain initial language text, and perform intent recognition on the initial language text to obtain the target intent; Determine the target business table corresponding to the target intent, and determine the reference business field group based on the initial language text and the candidate business field group in the target business table; When the number of reference business field groups is greater than or equal to a preset number threshold, the business field confidence of the corresponding reference business field group is determined based on the initial language text, the reference business field group, and the preset field quality dimension, and the target business field group is determined from the reference business field group based on the business field confidence. Input the target business field name, the initial language text, and the target business table name from the target business field group into the text rewriting model to obtain the target language text. Based on the target language text, determine the target structured query statement.

[0006] According to another aspect of the present invention, an apparatus for determining a structured query statement is provided, comprising: The target intent determination module is used to acquire initial language text and perform intent recognition on the initial language text to obtain the target intent; The reference business field group determination module is used to determine the target business table corresponding to the target intent, and to determine the reference business field group based on the initial language text and the candidate business field groups in the target business table; The target business field group determination module is used to determine the business field confidence level of the corresponding reference business field group based on the initial language text, the reference business field group and the preset field quality dimension when the number of reference business field groups is greater than or equal to a preset number threshold, and to determine the target business field group from the reference business field group based on the business field confidence level. The target language text determination module is used to input the target business field name in the target business field group, the initial language text, and the target business table name into the text rewriting model to obtain the target language text. The structured query statement determination module is used to determine the target structured query statement based on the target language text.

[0007] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method for determining a structured query statement as described in any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the method for determining a structured query statement as described in any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the method for determining a structured query statement according to any embodiment of the present invention.

[0010] This invention provides a scheme for determining structured query statements. It involves obtaining initial language text and performing intent recognition on the initial language text to obtain the target intent; determining the target business table corresponding to the target intent; and determining a reference business field group based on the initial language text and candidate business field groups in the target business table; when the number of reference business field groups is greater than or equal to a preset threshold, determining the business field confidence of the corresponding reference business field group based on the initial language text, the reference business field group, and a preset field quality dimension; and determining the target business field group from the reference business field group based on the business field confidence; inputting the target business field name, the initial language text, and the target business table name from the target business field group into a text rewriting model to obtain the target language text; and determining the target structured query statement based on the target language text. The above scheme determines a reference business field group based on the candidate business field group in the target business table corresponding to the target intent obtained by intent recognition of the initial language text, and the initial language text. When the number of reference business field groups is greater than or equal to a preset threshold, the target business field group is determined from the reference business field group based on the confidence of the business fields determined by the initial language text, the reference business field group, and the preset field quality dimension. The target business field name, the initial language text, and the target business table name in the target business field group are input into the text rewriting model to obtain the target language text. Based on the target language text, the target structured query statement is determined. This realizes the rewriting of the initial language text to obtain the target language text, improves the format standardization and structural clarity of the target language text, and improves the accuracy of the target structured query statement determined based on the target language text. In other words, it realizes the rewriting of the natural language text input by the user, making the rewritten text formatted and structurally clear, and improving the accuracy of the structured query statement determined based on the rewritten text.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a method for determining a structured query statement provided in Embodiment 1 of the present invention; Figure 2This is a flowchart of a method for determining a structured query statement provided in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of a structured query statement determination device provided in Embodiment 4 of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device that implements a method for determining structured query statements, as provided in Embodiment 5 of the present invention. Detailed Implementation

[0014] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0015] Example 1 Figure 1 This is a flowchart of a method for determining a structured query statement according to Embodiment 1 of the present invention. This embodiment is applicable to the case of rewriting natural language text input by a user and determining a structured query statement based on the rewritten text. The method can be executed by a structured query statement determination device, which can be implemented in hardware and / or software and can be configured in an electronic device that carries the function of determining structured query statements.

[0016] See Figure 1 The method for determining the structured query statement shown includes: S110. Obtain the initial language text and perform intent recognition on the initial language text to obtain the target intent.

[0017] Here, the initial language text refers to the natural language text input by the user. The target intent refers to the intent corresponding to the initial language text.

[0018] In an optional embodiment, the initial language text is subjected to intent recognition to obtain the target intent, including: performing keyword segmentation and matching processing on the initial language text and candidate intents to obtain matching results, and determining the initial recognition state based on the matching results; when the initial recognition state is initial recognition failure, the initial language text is input into a trained intent recognition model to obtain model output results, and the intermediate recognition state corresponding to the initial language text is determined based on the model output results; when the intermediate recognition state is intermediate recognition failure, the intent similarity between the initial language text and the candidate intent description is determined, and the candidate intent corresponding to the largest intent similarity is taken as the target intent corresponding to the initial language text.

[0019] Here, candidate intent refers to pre-defined intents used to determine the structured query statement. Matching result refers to the result of intent recognition of the initial language text based on keyword segmentation matching. For example, the matching result can be the candidate intent corresponding to the initial language text determined by keyword segmentation matching; or, the matching result can be empty, meaning that no candidate intent corresponding to the initial language text was found based on keyword segmentation matching.

[0020] For example, the keyword segmentation matching in this embodiment of the invention does not impose specific limitations, and can be set by technicians based on experience or needs. For instance, keyword segmentation matching can be exact match or reverse longest match, etc.

[0021] The initial recognition status can represent the state of intent recognition of the initial language text based on keyword segmentation and matching. For example, the initial recognition status can be either initial recognition success or initial recognition failure. Initial recognition success indicates that the target intent corresponding to the initial language text has been determined based on keyword segmentation and matching. Initial recognition failure indicates that the target intent corresponding to the initial language text has not been determined based on keyword segmentation and matching.

[0022] For example, if the matching result can be the candidate intent corresponding to the initial language text determined by keyword segmentation matching, then the first recognition status is determined to be the first recognition success. At this time, the candidate intent corresponding to the determined initial language text can be used as the target intent, and there is no need to perform subsequent intent recognition.

[0023] For example, if the matching result is empty, the initial recognition status is determined to be an initial recognition failure, and subsequent intent recognition needs to be performed on the initial language text.

[0024] The intent recognition model can be used to identify the intent of the initial language text and obtain the corresponding recognition result. The model output refers to the result obtained when the initial language text is input into the intent recognition model. For example, the model output can be a candidate intent corresponding to the initial language text, or it can be something else.

[0025] The intermediate recognition state can represent the state of intent recognition of the initial language text based on the intent recognition model. For example, the intermediate recognition state can be intermediate recognition success or intermediate recognition failure. Intermediate recognition success indicates that the target intent corresponding to the initial language text has been determined based on the intent recognition model. Intermediate recognition failure indicates that the target intent corresponding to the initial language text has not been determined based on the intent recognition model.

[0026] For example, if the model output is the candidate intent corresponding to the initial language text, then the intermediate recognition state is determined to be successful. At this time, the candidate intent corresponding to the initial language text can be used as the target intent, and there is no need to perform subsequent intent recognition.

[0027] For example, if the model output is something else, the intermediate recognition state is determined to be an intermediate recognition failure, and subsequent intent recognition needs to be performed on the initial language text.

[0028] Here, candidate intent description refers to the intent description corresponding to a candidate intent; that is, the candidate intent description can be understood as data that interprets the candidate intent. Each candidate intent has a corresponding candidate intent description. Intent similarity can be used to quantify the degree of similarity between the initial text vector and the candidate intent vector.

[0029] For example, the initial language text is vectorized to obtain an initial text vector; the candidate intent description is vectorized to obtain a candidate intent vector; the intent similarity between the initial text vector and the candidate intent vector is determined; the candidate intent description corresponding to the largest intent similarity is taken as the target intent description; and the candidate intent corresponding to the target intent description is taken as the target intent corresponding to the initial language text.

[0030] Here, the initial text vector refers to the vector corresponding to the initial language text. The candidate intent vector refers to the vector corresponding to the candidate intent description. The target intent description refers to the candidate intent description corresponding to the highest intent similarity.

[0031] Understandably, by first determining the target intent based on keyword segmentation and matching, and then using an intent recognition model to determine the target intent in the event of initial recognition failure, and further determining the target intent by determining the intent similarity between the initial language text and candidate intent descriptions in the event of intermediate recognition failure, the system achieves hierarchical intent recognition of the initial language text, thereby improving the accuracy of the determined target intent.

[0032] In this embodiment of the invention, the candidate intent and its description can be vectorized to obtain a candidate intent vector; the intent similarity between the initial text vector and the candidate intent vector can be determined; and the candidate intent with the highest intent similarity can be taken as the target intent. Here, the candidate intent vector is the vector corresponding to the candidate intent and its description.

[0033] S120. Determine the target business table corresponding to the target intent, and determine the reference business field group based on the initial language text and the candidate business field group in the target business table.

[0034] Here, the target business table refers to the table corresponding to the target intent. A candidate business field group refers to the collection of data associated with business fields stored in the target business table. For example, there may be multiple candidate business field groups in the target business table. Each candidate business field group includes candidate business field attributes, candidate business field names, candidate business field descriptions, and candidate synonymous business field names.

[0035] Among them, candidate business field attributes refer to the attributes corresponding to the business field; that is, candidate business field attributes can be understood as the English identifiers of the business field. Candidate business field names refer to the Chinese names of the business fields. Candidate business field descriptions refer to the business descriptions corresponding to the business fields. Candidate synonymous business field names refer to the Chinese field names of the near-synonymous business fields corresponding to the business field.

[0036] The reference business field group refers to the business field group selected from the candidate business field group.

[0037] S130. When the number of reference business field groups is greater than or equal to a preset quantity threshold, determine the business field confidence level of the corresponding reference business field group based on the initial language text, the reference business field group, and the preset field quality dimension, and determine the target business field group from the reference business field group based on the business field confidence level.

[0038] The present invention does not impose any limitation on the size of the preset quantity threshold. It can be set by technicians based on experience or needs, or determined repeatedly through a large number of experiments.

[0039] The preset field quality dimension refers to a pre-set dimension used to evaluate the quality of business fields corresponding to the reference business field names in the reference business field group. This embodiment of the invention does not impose any limitations on the setting of the preset field quality dimension; it can be set by technical personnel based on experience or needs. For example, the preset field quality dimension may include fill rate, anomaly rate, latency, and whether an index exists.

[0040] Among them, the business field confidence score can be used to quantify the reliability of the business fields corresponding to the reference business field names in the corresponding reference business field group. The target business field group refers to the business field group selected from the reference business field group.

[0041] In one optional embodiment, the confidence level of the business field of the corresponding reference business field group is determined based on the initial language text, the reference business field group, and the preset field quality dimension. This includes: extracting business keywords from the initial language text and determining the keyword vectors corresponding to the business keywords; determining the reference field group vector corresponding to the reference business field group, and determining the reference similarity corresponding to the corresponding reference business field group based on the keyword vector and the reference field group vector; wherein, the reference business field group includes reference business field attributes, reference business field names, reference business field descriptions, and reference synonymous business field names; obtaining the field dimension quality value of the reference business field name in the reference business field group under the preset field quality dimension; and determining the confidence level of the business field of the corresponding reference business field group based on the reference similarity and the field dimension quality value.

[0042] Here, business keywords refer to data in the initial text that are related to financial business. Keyword vectors are vectorized versions of business keywords. Reference field group vectors are vectorized versions of data from reference business field groups. Reference similarity can be used to quantify the degree of similarity between keyword vectors and reference field group vectors.

[0043] Among them, "Reference Business Field Attribute" refers to the candidate business field attribute in the reference business field group; "Reference Business Field Name" refers to the candidate business field name in the reference business field group; "Reference Business Field Description" refers to the candidate business field description in the reference business field group; and "Reference Synonymous Business Field Name" refers to the candidate synonymous business field name in the reference business field group.

[0044] The field dimension quality value refers to the specific value of the business field corresponding to the reference business field name under the preset field quality dimension. For example, if the preset field quality dimension includes fill rate, anomaly rate, latency, and whether an index exists, then the field dimension quality value of the reference business field name A could include a fill rate of 99%, an anomaly rate of 0.05%, real-time latency, and the existence of an index. Exemplarily, the specific values ​​of each candidate business field name under the preset field quality dimension can be determined in advance based on historical business data and operating status of the database; this embodiment of the invention does not impose specific limitations on this.

[0045] For example, business keywords are extracted from the initial language text and vectorized to obtain keyword vectors; for any reference business field group, the reference business field attributes, reference business field names, reference business field descriptions, and reference synonym business field names in the reference business field group are vectorized to obtain the reference field group vector for the reference business field group; the reference similarity of the reference business field group is determined based on the keyword vectors and the reference field group vectors; the field dimension quality value of the reference business field name in the reference business field group under the preset field quality dimension is obtained; and the business field confidence corresponding to the reference business field name in the reference business field group is determined based on the reference similarity and the field dimension quality value.

[0046] Understandably, by determining the keyword vector of the business keywords extracted from the initial language text, the reference similarity between it and the reference field group vector of the reference business field group, and the field dimension quality value of the reference business field name in the reference business field group, the confidence of the business field in the corresponding reference business field group is determined. This achieves the determination of the confidence of the business field from multiple dimensions and improves the accuracy of the determined confidence of the business field.

[0047] In an optional embodiment, determining the business field confidence of a reference business field in a corresponding reference business field group based on reference similarity and field dimension quality value includes: for any reference business field group, determining the reference similarity interval in which the reference similarity of the reference business field group falls, and using the reference interval confidence corresponding to the reference similarity interval as the business semantic confidence of the reference business field group; determining the field quality confidence of the reference business field group based on the field dimension quality value of the reference business field group; and determining the business field confidence of the reference business field group based on preset semantic weight, preset quality weight, business semantic confidence, and field quality confidence.

[0048] The reference similarity interval refers to the candidate similarity interval in which the reference similarity lies. The reference interval confidence score refers to the confidence score of the candidate interval corresponding to the reference similarity interval. There is a pre-defined correspondence between candidate similarity intervals and their confidence scores. The candidate similarity interval is a pre-defined similarity interval. The candidate interval confidence score is the pre-defined confidence score corresponding to the candidate similarity interval. The business semantic confidence score refers to the reliability of the business field corresponding to the reference business field name, determined from a semantic relevance perspective.

[0049] Among them, field quality confidence refers to the degree of reliability of the business field corresponding to the reference business field name determined from the perspective of data quality.

[0050] For example, if the preset field quality dimensions include fill rate, anomaly rate, latency, and whether there is an index, then for the fill rate, different fill rate ranges and corresponding fill dimension confidence levels can be preset; for the anomaly rate, different anomaly rate ranges and corresponding anomaly dimension confidence levels can be preset; for the latency, different latency duration ranges and corresponding latency dimension confidence levels can be preset; for whether there is an index, the first index dimension confidence level corresponding to having an index and the second index dimension confidence level corresponding to not having an index can be preset; preset fill weight, preset anomaly weight, preset latency weight, and preset index weight can also be introduced.

[0051] In this embodiment of the invention, no limitations are placed on the setting of the fill rate interval and the corresponding fill dimension confidence, the anomaly rate interval and the corresponding anomaly dimension confidence, the delay duration interval and the corresponding delay dimension confidence, the first index dimension confidence corresponding to the index, and the second index dimension confidence corresponding to the non-index. These can be set by technical personnel based on experience or needs.

[0052] In this embodiment of the invention, the size of the preset fill weight, preset anomaly weight, preset delay weight, and preset index weight are not limited in any way. They can be set by technicians based on experience or needs, or determined repeatedly through a large number of experiments. It is only necessary to ensure that the sum of the preset fill weight, preset anomaly weight, preset delay weight, and preset index weight is 1.

[0053] Continuing the previous example, for the reference business field name A, based on the field dimension quality value, fill rate range, and corresponding fill dimension confidence level of the business field corresponding to reference business field name A under the fill rate dimension, sub-confidence P1 is determined; based on the field dimension quality value, anomaly rate range, and corresponding anomaly dimension confidence level of the business field corresponding to reference business field name A under the anomaly rate dimension, sub-confidence P2 is determined; based on the field dimension quality value, latency duration range, and corresponding latency dimension confidence level of the business field corresponding to reference business field name A under the latency dimension, sub-confidence P2 is determined. Determine the sub-confidence P3 of the reference business field name A based on the confidence level of the delayed dimension; determine the sub-confidence P4 of the reference business field name A based on the field dimension quality value of the business field corresponding to the reference business field name A under the dimension of whether there is an index, the confidence level of the first index dimension corresponding to the index, and the confidence level of the second index dimension corresponding to the absence of an index; and perform a weighted sum of the sub-confidence P1, sub-confidence P2, sub-confidence P3, and sub-confidence P4 based on the preset fill weight, preset anomaly weight, preset delayed weight, and preset index weight to obtain the field quality confidence level corresponding to the reference business field name A.

[0054] In this embodiment of the invention, the magnitude of the preset semantic weight and the preset quality weight are not limited. They can be set by technicians based on experience or needs, or determined repeatedly through a large number of experiments. It is only necessary to ensure that the sum of the preset semantic weight and the preset quality weight is 1.

[0055] For example, based on preset semantic weights and preset quality weights, the business semantic confidence and field quality confidence are weighted and summed to determine the business field confidence of the reference business field name in the reference business field group.

[0056] Understandably, for any reference business field group, the business semantic confidence of the reference business field name in that reference business field group is determined by using the reference interval confidence of the reference similarity interval in which the reference similarity of that reference business field group falls. The field quality confidence of the reference business field name in that reference business field group is determined based on the field dimension quality value of the reference business field name. Finally, the business field confidence of the reference business field name in that reference business field group is determined based on preset semantic weights, preset quality weights, business semantic confidence, and field quality confidence. This achieves the determination of business field confidence from both semantic relevance and data quality dimensions, improving the accuracy of the determined business field confidence.

[0057] For example, the reference business field groups are sorted from largest to smallest according to the confidence level of the business fields, and the top 10 reference business field groups after sorting are selected as the target business field groups.

[0058] S140. Input the target business field name, initial language text, and target business table name from the target business field group into the text rewriting model to obtain the target language text.

[0059] The text rewriting model can be used to rewrite initial language text to obtain target language text. For example, a depth-first search model can be used for text rewriting. The target business table name refers to the name of the target business table. The target language text refers to the structured query statement obtained after rewriting the initial language text.

[0060] For example, the target business field name, initial language text, and target business table name in the target business field group can be input into the trained text rewriting model to obtain the target language text.

[0061] S150. Determine the target structured query statement based on the target language text.

[0062] Among them, the target structured query statement refers to the structured query statement determined based on the target language text.

[0063] For example, target language text can be input into a trained structured query generation model to obtain the target structured query. The structured query generation model can then be used to generate structured query statements.

[0064] This invention provides a scheme for determining structured query statements. It involves obtaining initial language text and performing intent recognition on the initial language text to obtain the target intent; determining the target business table corresponding to the target intent; and determining a reference business field group based on the initial language text and candidate business field groups in the target business table; when the number of reference business field groups is greater than or equal to a preset threshold, determining the business field confidence of the corresponding reference business field group based on the initial language text, the reference business field group, and a preset field quality dimension; and determining the target business field group from the reference business field group based on the business field confidence; inputting the target business field name, the initial language text, and the target business table name from the target business field group into a text rewriting model to obtain the target language text; and determining the target structured query statement based on the target language text. The above scheme determines a reference business field group based on the candidate business field group in the target business table corresponding to the target intent obtained by intent recognition of the initial language text, and the initial language text. When the number of reference business field groups is greater than or equal to a preset threshold, the target business field group is determined from the reference business field group based on the confidence of the business fields determined by the initial language text, the reference business field group, and the preset field quality dimension. The target business field name, the initial language text, and the target business table name in the target business field group are input into the text rewriting model to obtain the target language text. Based on the target language text, the target structured query statement is determined. This realizes the rewriting of the initial language text to obtain the target language text, improves the format standardization and structural clarity of the target language text, and improves the accuracy of the target structured query statement determined based on the target language text. In other words, it realizes the rewriting of the natural language text input by the user, making the rewritten text formatted and structurally clear, and improving the accuracy of the structured query statement determined based on the rewritten text.

[0065] Example 2 Figure 2This is a flowchart of a method for determining a structured query statement according to Embodiment 2 of the present invention. Based on the above embodiments, this embodiment further refines the operation of "determining a reference business field group based on the initial language text and the candidate business field group in the target business table" into "determining the initial text vector corresponding to the initial language text, and determining the candidate field group vector corresponding to the candidate business field group; determining the initial similarity between the initial text vector and each candidate field group vector, and determining the reference business field group based on the initial similarity and a preset similarity threshold," thereby improving the reference business field group determination mechanism. It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments.

[0066] See Figure 2 The method for determining the structured query statement shown includes: S210. Obtain the initial language text and perform intent recognition on the initial language text to obtain the target intent.

[0067] S220. Determine the target business table corresponding to the target intent.

[0068] S230. Determine the initial text vector corresponding to the initial language text, and determine the candidate field group vector corresponding to the candidate business field group.

[0069] Here, the candidate field group vector refers to the vector corresponding to the candidate business field group. For example, the candidate field group vector can be determined as follows: for any candidate business field group, the candidate business field name, candidate business field description, candidate business field attribute, and candidate synonymous business field name in the candidate business field group are vectorized to obtain the candidate field group vector of the candidate business field group; or, the candidate business field name and candidate business field description in the candidate business field group are vectorized to obtain the candidate field group vector of the candidate business field group.

[0070] S240. Determine the initial similarity between the initial text vector and each candidate field group vector, and determine the reference business field group based on the initial similarity and the preset similarity threshold.

[0071] The initial similarity can be used to quantify the degree of similarity between the initial text vector and the candidate field group vector. This embodiment of the invention does not impose any limitation on the size of the preset similarity threshold; it can be set by technicians based on experience or needs, or determined repeatedly through numerous experiments.

[0072] In an optional embodiment, determining a reference business field group based on an initial similarity and a preset similarity threshold includes: for any candidate business field group, if the initial similarity corresponding to the candidate business field group is greater than or equal to the preset similarity threshold, then the candidate business field group is used as a reference business field group; if the initial similarity corresponding to the candidate business field group is less than the preset similarity threshold, then the candidate business field group is prohibited from being used as a reference business field group.

[0073] Understandably, for any candidate business field group, if the initial similarity corresponding to the candidate business field group is greater than or equal to the preset similarity threshold, the candidate business field group is used as the reference business field group; if the initial similarity corresponding to the candidate business field group is less than the preset similarity threshold, the candidate business field group is prohibited from being used as the reference business field group, thereby improving the accuracy of the determined reference business field group.

[0074] S250. When the number of reference business field groups is greater than or equal to a preset number threshold, determine the business field confidence level of the corresponding reference business field group based on the initial language text, the reference business field group, and the preset field quality dimension, and determine the target business field group from the reference business field group based on the business field confidence level.

[0075] S260. Input the target business field name, initial language text, and target business table name from the target business field group into the text rewriting model to obtain the target language text.

[0076] S270. Determine the target structured query statement based on the target language text.

[0077] This invention provides a scheme for determining structured query statements. By refining the operation of determining reference business field groups based on initial language text and candidate business field groups in the target business table, it further refines this process into determining the initial text vector corresponding to the initial language text, and determining the candidate field group vectors corresponding to the candidate business field groups. It then determines the initial similarity between the initial text vector and each candidate field group vector, and determines the reference business field group based on the initial similarity and a preset similarity threshold, thus improving the reference business field group determination mechanism. This scheme improves the accuracy of determining the reference business field group by using the initial similarity between the determined initial text vector and each candidate field group vector, as well as the preset similarity threshold, to determine the reference business field group from the candidate business field groups.

[0078] Example 3 This invention provides an optional example based on the above embodiments. It should be noted that for parts not described in detail in this invention's embodiments, please refer to the descriptions in other embodiments.

[0079] For example, the technology of automatically converting natural language into structured query language aims to lower the technical barrier to database queries, enabling natural interaction between non-technical personnel and databases. Conversational business intelligence is a new data analysis paradigm centered on natural language interaction. It is an intelligent analysis tool based on a large language model, which enables data querying, visualization, and insight generation through natural language interaction.

[0080] Existing technologies typically involve business personnel directly using natural language to describe the customers, organizations, or metrics for which product recommendations are needed. At the same time, they call the software development kit provided by conversational business intelligence to achieve the conversion of natural language into structured query language by automatically converting natural language into structured query language. Then, they use the structured query language to query the database for corresponding metrics and display them.

[0081] Existing technologies suffer from insufficient understanding of intent when querying different tables. Specifically, the most basic implementation of conversational business intelligence distinguishes related table names by having users select corresponding assistants and dimensions. However, users often do not understand the assistants and dimensions. Although simple intent recognition has been added to the tools, insufficient understanding of intent for different tables still occurs, relying on users' insufficient understanding of the descriptions of each assistant.

[0082] Existing technologies for automatically converting natural language into structured query language and rendering charts have lengthy processes, leading to user anxiety due to waiting time. Specifically, the entire process of automatically converting natural language into structured query language is lengthy. Using software development kits (SDKs) results in no intermediate process output. From selecting an assistant to the final chart output, it often involves intent recognition to select related tables, generating semantic structured query statements, converting semantic structured query statements into structured query statements, and performing ultra-wide table queries. The lack of content output during this process causes user anxiety due to waiting time.

[0083] Existing technologies suffer from limitations: business users don't know how to ask questions, and the models can't understand the problems. Specifically, conversational business intelligence models cannot control the format and fields of the questions they receive. Often, the structured queries generated directly by business users are unusable or inaccurate, making it impossible to accurately find the corresponding tables and precise fields.

[0084] The technical solution provided by the embodiments of the present invention overcomes the problems of inaccurate table confirmation, user waiting anxiety, and user not knowing how to ask questions and the model not understanding them in the prior art. It provides a method to clarify the effect of intent recognition and question understanding through large model and intelligent agent technology.

[0085] The technical solution provided by the embodiments of the present invention adds intent recognition, large model, and enhanced retrieval generation technologies to traditional conversational business intelligence to ensure optimization of the problems existing in the invocation of traditional conversational business intelligence.

[0086] For example, to address the limitation of automatic natural language to structured query language technology in recognizing incorrect table names, an intent recognition node is added. This node employs three different intent recognition mechanisms: keyword segmentation and matching, an intent recognition model that understands user questions and automatically categorizes intents, and retrieval enhancement generation technology. The intent recognition node pre-confirms the associated table name (i.e., identifies the target business table) before the automatic natural language to structured query language technology. The range of query table names can be flexibly increased or decreased by adding segmentation rules, adjusting prompts, and enriching the knowledge base used for intent recognition, while simultaneously optimizing the intent recognition effect.

[0087] For example, to address the shortcomings of business personnel not knowing how to ask questions and models not understanding the questions, retrieval enhancement generation technology is used to maintain the synonyms (i.e., candidate synonymous business field names) of each indicator in the target business table into the knowledge base, while adding a description of the indicator (i.e., candidate business field description), which can be manually reviewed by business and technical personnel; target business field names with high similarity to the target business table are found in the initial language text through vector matching.

[0088] For example, the text rewriting model transforms the initial language text from its original arbitrary form into a target language text with a clear structure, accurate indicator names, and accurate query conditions. This makes the final target language text more relevant to the technical problem, making the model that automatically converts natural language into structured query language (structured query statement generation model) easier to understand and easier to generate accurate target structured query statements.

[0089] For example, to address user anxiety caused by the long process of automatically converting natural language to structured query language, querying large tables, and rendering icons, the system provides a streaming output of target business field names, query conditions, and target language text before the automatic conversion. This intermediate output provides users with readable content, reducing their waiting anxiety.

[0090] For example, in addition to rendering all the tables and graphs of traditional business intelligence, an additional intelligent insights function is added after the data is retrieved and rendered. The model base is depth exploration, and the prompt words can be adjusted by technical personnel to provide business personnel with a summary of information that can be obtained based on this business intelligence information and recommendations for relevant opportunities.

[0091] The technical solution provided in this invention associates natural language text with business tables through intent recognition and optimizes prompt words, offering a more convenient tuning method and better optimization results. The combination of a large model and retrieval enhancement generation technology optimizes the accuracy of indicator names (i.e., target business field names) through vector matching and other methods, while uniformly sending them to a text rewriting model to modify them into a text format better understood by the structured query generation model, thus enabling the structured query generation model to better understand natural language text and generate target structured queries. Multiple message nodes and full output and rendering of intermediate steps reduce user waiting anxiety. The large model is used to understand the output indicators, tables, and charts, outputting summaries and recommendation opportunities.

[0092] It should be noted that the information collected in the embodiments of the present invention is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, and necessary confidentiality measures have been taken. It does not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse. If the user chooses to refuse, the process will proceed to the expert decision-making process.

[0093] Example 4 Figure 3 This is a schematic diagram of a structured query statement determination device provided in Embodiment 4 of the present invention. This embodiment is applicable to the case of rewriting natural language text input by a user and determining a structured query statement based on the rewritten text. This method can be executed by a structured query statement determination device, which can be implemented in hardware and / or software and can be configured in an electronic device that carries the function of determining structured query statements.

[0094] like Figure 3 As shown, the device includes: a target intent determination module 310, a reference business field group determination module 320, a target business field group determination module 330, a target language text determination module 340, and a structured query statement determination module 350. Among them, The target intent determination module 310 is used to acquire initial language text and perform intent recognition on the initial language text to obtain the target intent; The reference business field group determination module 320 is used to determine the target business table corresponding to the target intent, and to determine the reference business field group based on the initial language text and the candidate business field groups in the target business table; The target business field group determination module 330 is used to determine the business field confidence level of the corresponding reference business field group based on the initial language text, the reference business field group and the preset field quality dimension when the number of reference business field groups is greater than or equal to a preset number threshold, and to determine the target business field group from the reference business field group based on the business field confidence level. The target language text determination module 340 is used to input the target business field in the target business field group, the initial language text and the target business table name into the text rewriting model to obtain the target language text. The structured query statement determination module 350 is used to determine the target structured query statement based on the target language text.

[0095] This invention provides a scheme for determining structured query statements. It involves obtaining initial language text and performing intent recognition on the initial language text to obtain the target intent; determining the target business table corresponding to the target intent; and determining a reference business field group based on the initial language text and candidate business field groups in the target business table; when the number of reference business field groups is greater than or equal to a preset threshold, determining the business field confidence of the corresponding reference business field group based on the initial language text, the reference business field group, and a preset field quality dimension; and determining the target business field group from the reference business field group based on the business field confidence; inputting the target business field name, the initial language text, and the target business table name from the target business field group into a text rewriting model to obtain the target language text; and determining the target structured query statement based on the target language text. The above scheme determines a reference business field group based on the candidate business field group in the target business table corresponding to the target intent obtained by intent recognition of the initial language text, and the initial language text. When the number of reference business field groups is greater than or equal to a preset threshold, the target business field group is determined from the reference business field group based on the confidence of the business fields determined by the initial language text, the reference business field group, and the preset field quality dimension. The target business field name, the initial language text, and the target business table name in the target business field group are input into the text rewriting model to obtain the target language text. Based on the target language text, the target structured query statement is determined. This realizes the rewriting of the initial language text to obtain the target language text, improves the format standardization and structural clarity of the target language text, and improves the accuracy of the target structured query statement determined based on the target language text. In other words, it realizes the rewriting of the natural language text input by the user, making the rewritten text formatted and structurally clear, and improving the accuracy of the structured query statement determined based on the rewritten text.

[0096] Optionally, the reference business field group determination module 320 includes: The candidate field group vector determination unit is used to determine the initial text vector corresponding to the initial language text and to determine the candidate field group vector corresponding to the candidate business field group. The reference business field group determination unit is used to determine the initial similarity between the initial text vector and each of the candidate field group vectors, and to determine the reference business field group based on the initial similarity and a preset similarity threshold.

[0097] Optionally, the reference business field group determining unit is specifically used for: For any candidate business field group, if the initial similarity corresponding to the candidate business field group is greater than or equal to the preset similarity threshold, then the candidate business field group is used as the reference business field group. If the initial similarity of the candidate business field group is less than the preset similarity threshold, then the candidate business field group shall not be used as the reference business field group.

[0098] Optionally, the target business field group determination module 330 includes: A keyword vector determination unit is used to extract business keywords from the initial language text and determine the keyword vectors corresponding to the business keywords; A reference similarity determination unit is used to determine the reference field group vector corresponding to the reference business field group, and to determine the reference similarity corresponding to the corresponding reference business field group based on the keyword vector and the reference field group vector; wherein, the reference business field group includes reference business field attributes, reference business field names, reference business field descriptions and reference synonymous business field names; The field dimension quality value acquisition unit is used to acquire the field dimension quality value of the reference business field name in the reference business field group under the preset field quality dimension. The business field confidence determination unit is used to determine the business field confidence of the corresponding reference business field group based on the reference similarity and the field dimension quality value.

[0099] Optionally, the business field confidence determination unit is specifically used for: For any reference business field group, determine the reference similarity interval in which the reference similarity of the reference business field group is located, and take the reference interval confidence corresponding to the reference similarity interval as the business semantic confidence of the reference business field group; Based on the field dimension quality values ​​of the reference business field group, determine the field quality confidence level corresponding to the reference business field group; The business field confidence level of the reference business field group is determined based on the preset semantic weight, the preset quality weight, the business semantic confidence level, and the field quality confidence level.

[0100] Optionally, the target intent determination module 310 includes: The initial recognition state determination unit is used to perform keyword segmentation and matching processing on the initial language text and candidate intent, obtain matching results, and determine the initial recognition state based on the matching results; The intermediate recognition state determination unit is used to input the initial language text into the trained intent recognition model when the initial recognition state is "initial recognition failure", obtain the model output result, and determine the intermediate recognition state corresponding to the initial language text based on the model output result. The target intent determination unit is used to determine the intent similarity between the initial language text and the candidate intent description when the intermediate recognition state is intermediate recognition failure, and to take the candidate intent corresponding to the largest intent similarity as the target intent corresponding to the initial language text.

[0101] The structured query statement determination device provided in the embodiments of the present invention can execute the structured query statement determination method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the determination method of each structured query statement.

[0102] The collection, storage, use, processing, transmission, provision, and disclosure of initial language texts and other materials involved in the technical solution of this invention all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0103] Example 5 According to embodiments of the present invention, the present invention also provides an electronic device, a readable storage medium, and a computer program product.

[0104] Figure 4 This is a schematic diagram of an electronic device that implements a method for determining structured query statements, as provided in Embodiment 5 of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0105] like Figure 4As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0106] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0107] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the method for determining structured query statements.

[0108] In some embodiments, the method for determining structured query statements may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for determining structured query statements described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the method for determining structured query statements by any other suitable means (e.g., by means of firmware).

[0109] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0110] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0111] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0112] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0113] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0114] A computing system can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product within the cloud computing service system. This addresses the shortcomings of traditional physical hosts and virtual private servers, such as high management difficulty and weak business scalability.

[0115] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0116] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for determining a structured query statement, characterized in that, include: Obtain initial language text, and perform intent recognition on the initial language text to obtain the target intent; Determine the target business table corresponding to the target intent, and determine the reference business field group based on the initial language text and the candidate business field group in the target business table; When the number of reference business field groups is greater than or equal to a preset number threshold, the business field confidence of the corresponding reference business field group is determined based on the initial language text, the reference business field group, and the preset field quality dimension, and the target business field group is determined from the reference business field group based on the business field confidence. Input the target business field name, the initial language text, and the target business table name from the target business field group into the text rewriting model to obtain the target language text. Based on the target language text, determine the target structured query statement.

2. The method according to claim 1, characterized in that, The step of determining the reference business field group based on the initial language text and the candidate business field group in the target business table includes: Determine the initial text vector corresponding to the initial language text, and determine the candidate field group vector corresponding to the candidate business field group; Determine the initial similarity between the initial text vector and each of the candidate field group vectors, and determine the reference business field group based on the initial similarity and a preset similarity threshold.

3. The method according to claim 2, characterized in that, The step of determining the reference business field group based on the initial similarity and the preset similarity threshold includes: For any candidate business field group, if the initial similarity corresponding to the candidate business field group is greater than or equal to the preset similarity threshold, then the candidate business field group is used as the reference business field group. If the initial similarity of the candidate business field group is less than the preset similarity threshold, then the candidate business field group shall not be used as the reference business field group.

4. The method according to claim 1, characterized in that, The step of determining the business field confidence level of the corresponding reference business field group based on the initial language text, the reference business field group, and the preset field quality dimension includes: Extract business keywords from the initial language text and determine the keyword vectors corresponding to the business keywords; Determine the reference field group vector corresponding to the reference business field group, and determine the reference similarity corresponding to the corresponding reference business field group based on the keyword vector and the reference field group vector; wherein, the reference business field group includes reference business field attributes, reference business field names, reference business field descriptions, and reference synonym business field names; Obtain the field dimension quality value of the reference business field name in the reference business field group under the preset field quality dimension; Based on the reference similarity and the field dimension quality value, the business field confidence level of the corresponding reference business field group is determined.

5. The method according to claim 4, characterized in that, The step of determining the business field confidence level of the corresponding reference business field group based on the reference similarity and the field dimension quality value includes: For any reference business field group, determine the reference similarity interval in which the reference similarity of the reference business field group is located, and take the reference interval confidence corresponding to the reference similarity interval as the business semantic confidence of the reference business field group; Based on the field dimension quality values ​​of the reference business field group, determine the field quality confidence level corresponding to the reference business field group; The business field confidence level of the reference business field group is determined based on the preset semantic weight, the preset quality weight, the business semantic confidence level, and the field quality confidence level.

6. The method according to any one of claims 1-5, characterized in that, The step of performing intent recognition on the initial language text to obtain the target intent includes: The initial language text and candidate intent are subjected to keyword segmentation and matching to obtain matching results, and the initial recognition state is determined based on the matching results. When the initial recognition state is "first recognition failure", the initial language text is input into the trained intent recognition model to obtain the model output result, and the intermediate recognition state corresponding to the initial language text is determined based on the model output result. When the intermediate recognition state is intermediate recognition failure, the intent similarity between the initial language text and the candidate intent description is determined, and the candidate intent corresponding to the largest intent similarity is taken as the target intent corresponding to the initial language text.

7. A device for determining a structured query statement, characterized in that, include: The target intent determination module is used to acquire initial language text and perform intent recognition on the initial language text to obtain the target intent; The reference business field group determination module is used to determine the target business table corresponding to the target intent, and to determine the reference business field group based on the initial language text and the candidate business field groups in the target business table; The target business field group determination module is used to determine the business field confidence level of the corresponding reference business field group based on the initial language text, the reference business field group and the preset field quality dimension when the number of reference business field groups is greater than or equal to a preset number threshold, and to determine the target business field group from the reference business field group based on the business field confidence level. The target language text determination module is used to input the target business field name in the target business field group, the initial language text, and the target business table name into the text rewriting model to obtain the target language text. The structured query statement determination module is used to determine the target structured query statement based on the target language text.

8. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a method for determining a structured query statement as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements a method for determining a structured query statement as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method for determining a structured query statement according to any one of claims 1-6.

Images