Analysis method, system and equipment based on AI model and storage medium

Abstract

The invention relates to an analysis method, system and device based on an AI model and a storage medium, and belongs to the technical field of service decision making through AI.The method comprises the steps that an analysis instruction is obtained, and initial content contained in the analysis instruction is input into a pre-constructed analysis model; based on a pre-constructed decision tree, decision nodes contained in the decision tree and a plurality of preset basic questions corresponding to the decision nodes, through the analysis model, the initial content is disassembled and then analyzed to determine answer content associated with the basic questions; and integrating the answer content to generate an analysis result, and outputting the analysis result. The method has the effect of improving the analysis efficiency and analysis accuracy of the AI model on general and complicated initial contents.

Description

Technical Field

[0001] The present application relates to the technical field of using AI to make business decisions, and in particular to an analysis method, system, device and storage medium based on an AI model. Background Art

[0002] With the continuous development of deep learning technology, the current analysis method of using AI models to analyze a problem has gradually replaced the manual analysis method to improve the efficiency of analysis. Among them, for example, the above-mentioned problem can be specifically a text material, such as a report material, and the report material can specifically include a patient's medical examination report, a project construction report, etc.; such as by inputting the patient's medical examination report into the AI ​​model, using the AI ​​model to extract the specified type of data in the patient's medical examination report for analysis, and finally analyzing the patient's condition; or by inputting the project construction report into the AI ​​model, using the AI ​​model to analyze the specified type of data in the project construction report, and finally analyzing the construction quality.

[0003] However, when the AI ​​model is needed to analyze a complex problem, such as inputting a report material containing complex data from multiple sources into the AI ​​model at one time, it may easily lead to the AI ​​model being unable to efficiently and accurately extract the required data for analysis, so there is room for improvement. Summary of the invention

[0004] In order to improve the efficiency and accuracy of AI models in analyzing abstract and complex content, the present application provides an analysis method, system, device and storage medium based on an AI model.

[0005] In the first aspect, the present application provides an analysis method based on an AI model, which adopts the following technical solutions: Obtaining an analysis instruction, and inputting initial content contained in the analysis instruction into a pre-built analysis model; Based on a pre-constructed decision tree, decision nodes contained in the decision tree, and several preset basic questions corresponding to the decision nodes, the initial content is disassembled and analyzed through the analysis model to determine the answer content associated with the basic questions; The answer contents are integrated to generate analysis results, and the analysis results are output.

[0006] By adopting the above technical solution, when analyzing complex and general initial content, the present application proposes to input it into a pre-built analysis model, and then based on pre-defined basic questions and a decision tree for defining the query order and query logic of the basic questions, the initial content is targetedly decomposed into a number of detailed contents (wherein each detailed content is related to any basic question), and then each detailed content after decomposition is analyzed separately. The analysis method is to extract the answer to the corresponding basic question (i.e., the answer content), and finally the integrated result obtained by integrating all the answer contents is the analysis result of the initial content, and finally the analysis and judgment of the aforementioned complex and general initial content is realized, thereby improving the accuracy, efficiency and pertinence of the analysis of complex and general current content.

[0007] Optionally, the analyzing model is used to analyze and determine the answer content associated with the basic question by disassembling the initial content, including: Preliminarily determine a set of questions, and disassemble the answer content associated with the basic questions included in the set of questions from the initial content through the analysis model; wherein the set of questions includes several basic questions; Whenever it is determined that an answer content associated with a target basic question is obtained, it is determined whether the first answer content satisfies a preset stop analysis condition; wherein the target basic question refers to any basic question in a question set, and the first answer content refers to an answer content associated with the target basic question; the stop analysis condition at least includes: a decision node corresponding to the first answer content does not contain a lower-level node in the decision tree; If the first answer content does not meet the preset stop analysis condition, then based on the first answer content, the basic questions included in the question set are adjusted so that the decision nodes corresponding to the basic questions included in the adjusted question set are all decision nodes related to the first answer content until the stop analysis condition is met; The step of integrating the answer contents to generate analysis results includes: When the first answer content meets the preset analysis stop condition, the answer content is integrated to generate an analysis result.

[0008] By adopting the above technical solution, whether the analysis operation is completed is determined according to the answer content corresponding to each basic question, and the basis for the determination is whether the answer content obtained from each analysis meets the conditions for stopping the analysis, that is, the decision node corresponding to the first answer content does not contain lower-level nodes in the decision tree. In other words, it means that the first answer content is clear and unique, and there are no other questions that need to be asked subsequently. In summary, the basic questions that need to be asked can be dynamically adjusted through the answer content, so as to achieve targeted analysis and improve analysis efficiency.

[0009] Optionally, adjusting the basic questions included in the question set based on the first answer content includes: Based on the first answer content, taking the decision nodes of all lower nodes in the decision tree that are the decision nodes corresponding to the first answer content as target decision nodes, and adding the basic questions corresponding to the target decision nodes to the question set; If there is a first target decision node in the target decision node, and the first target decision node satisfies: the first target decision node corresponds to a sub-decision tree, then the guiding question corresponding to the sub-decision tree is added to the question set; Whenever a second answer content associated with a target guiding question is determined, it is determined whether the second answer content satisfies a preset guidance condition. If so, the basic questions of all decision nodes contained in the sub-decision tree corresponding to the target guiding question are added to the question set.

[0010] By adopting the above-mentioned technical solution, the present application stipulates that a single decision node can further contain an independent sub-decision tree. When the guiding question corresponding to the sub-decision tree is inquired and the second answer content is determined, if the second answer content meets the preset guidance conditions, the basic questions of all decision nodes contained in the sub-decision tree are added to the question set to achieve further refined analysis of the initial content.

[0011] Optionally, the extracting, from the initial content, the answer content associated with the basic question pointed to by the preset query engine through the analysis model, further includes: In the initial content, a label is added to each answer content; wherein the label is used to represent the association relationship between the answer content and the corresponding basic question.

[0012] By adopting the above technical solution, after extracting the answer content related to the basic question from the initial content, the location of the answer content corresponding to the basic question in the initial content is marked by adding tags (i.e. marking the source of the answer to the basic question), so that experts can know the source of the answer content when reviewing.

[0013] Optionally, the inputting the initial content included in the analysis instruction into a pre-built analysis model includes: According to the report type of the initial content in the analysis instruction, a report template is matched, and based on the report template obtained by matching, the basic data in the initial content in the analysis instruction is classified and integrated, and the initial content is input into a pre-built classification model.

[0014] By adopting the above technical solution, before using the analysis model to analyze the initial content, a report template can be matched for the initial content, and the basic data in the initial content can be classified and integrated according to the preset arrangement of the report template. The pre-classification and integration operation can help the subsequent analysis model to more efficiently realize the association between basic questions and answer content.

[0015] Optionally, the method further includes: receiving a manual editing instruction, adding a decision node in the decision tree based on the manual editing instruction, and setting a trigger condition for the currently added decision node according to the node attribute contained in the manual editing instruction; Whenever an analysis instruction is received, a preliminary analysis is performed on the initial content through the analysis model to determine all decision nodes that meet the triggering conditions, and a decision tree is constructed using the decision nodes that meet the triggering conditions.

[0016] By adopting the above technical solution, editing instructions can be triggered during expert review, such as adding decision nodes and defining applicable scenarios (i.e., trigger conditions) for decision nodes. Whenever an analysis instruction is received, it can be determined whether all decision nodes meet the trigger conditions based on the specific circumstances of the analysis report in the analysis instruction, and only all decision nodes that meet the trigger conditions can be used to construct a decision tree, thereby achieving adaptability and dynamic adjustment of the decision tree.

[0017] Optionally, the initial content also includes basic information of the analyzed object and the time when the report was generated; The method further comprises: Whenever an analysis instruction is received, based on the analysis result corresponding to the initial content, basic attributes of the analyzed object are determined, wherein the basic attributes at least include a time transition type; For the initial content whose basic attribute is time-shifting, the decision node corresponding to when the analysis model ends the analysis operation on the initial content is used as the end node, and according to the answer content corresponding to the end node, the change trend of the basic data in the initial content corresponding to the analyzed object over time is predicted, and an estimated index is generated; wherein the estimated index contains a number of basic questions, and the basic questions at least include the basic questions corresponding to the end node; Generate an analysis record based on the analysis result corresponding to the initial content, the analysis result and the corresponding estimated index, and store the analysis record; The extracting, from the initial content, the answer content associated with the basic question pointed to by the preset query engine by means of the analysis model includes: Determine whether there is an analysis record for the analyzed object corresponding to the initial content in the historical period. If so, retrieve the most recent analysis record before the current time, and based on the analysis record, use a preset query engine to point to basic questions included in the estimated index in the analysis record one by one; The analysis model is used to extract answer content associated with the basic question pointed out by the preset query engine from the initial content.

[0018] By adopting the above technical solution, if the analyzed object corresponding to the initial content changes with time, its corresponding initial content (i.e., the basic data in the initial content) will change. Then, in order to avoid repeatedly asking basic questions that have been asked in the past, the present application proposes to predict the changing trend of the initial content of the analyzed object, and reorganize a number of basic questions according to the prediction results to generate an estimated index. The basic questions included in the estimated index can be understood as the basic questions that need to be asked when analyzing the initial content generated by the analyzed object next time. The basic questions can cover or partially cover the basic questions included in the current analysis process, that is, a group of basic questions are comprehensively obtained based on the results of this analysis and combined with the prediction of future change trends. By making full use of historical analysis results, unnecessary questions in the next analysis process can be reduced as much as possible, thereby improving analysis efficiency.

[0019] In a second aspect, the present application provides an analysis system based on an AI model, comprising: An analysis instruction acquisition module, used to acquire an analysis instruction and input the initial content contained in the analysis instruction into a pre-built analysis model; An initial content analysis module is used to analyze and determine the answer content associated with the basic questions by disassembling the initial content through the analysis model based on a pre-constructed decision tree, decision nodes contained in the decision tree, and several preset basic questions corresponding to the decision nodes; The analysis result review module is used to integrate the answer content to generate analysis results and output the analysis results.

[0020] In a third aspect, the present application provides an analysis device based on an AI model, comprising a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute any one of the methods described in the first aspect.

[0021] In a fourth aspect, the present application provides a computer-readable storage medium, characterized in that it stores a computer program that can be loaded by a processor and execute any one of the methods described in the first aspect.

[0022] In summary, this application includes the following beneficial technical effects: In the present application, when analyzing complex and abstract initial content, the present application proposes to decompose a general and complicated initial content into several basic and simple detailed contents, and finally realize the analysis and judgment of the aforementioned complex and abstract initial content by analyzing the detailed contents one by one, thereby improving the accuracy and efficiency of the analysis of the complex and abstract initial content. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 It is a flowchart of an analysis method based on an AI model disclosed in an embodiment of the present application.

[0025] Figure 2 It is a schematic diagram of a decision tree structure in a medical scenario in an embodiment of the present application.

[0026] Figure 3 It is a structural block diagram of an analysis system based on an AI model disclosed in an embodiment of the present application.

[0027] Description of the accompanying drawings: 201, analysis instruction acquisition module; 202, initial content analysis module; 202, analysis result review module. DETAILED DESCRIPTION

[0028] The following is combined with Figure 1-3 This application is described in further detail.

[0029] The present application embodiment discloses an analysis method based on an AI model. Figure 1 The analysis method based on the AI ​​model is used to analyze the general and complex initial content, disassemble it and analyze each disassembled content, and finally integrate the analysis results to form the analysis results of the aforementioned initial content, so as to achieve efficient and accurate analysis of the initial content (especially the general and complex initial content). Among them, the execution subject of the analysis method based on the AI ​​model is an analysis system based on the AI ​​model (hereinafter referred to as the analysis system). Figure 1-2 The process steps of the analysis method based on the AI ​​model are specifically explained.

[0030] S101, obtaining an analysis instruction, and inputting the initial content included in the analysis instruction into a pre-built analysis model.

[0031] In implementation, users can access the analysis system through a web page or APP and trigger analysis instructions; the initial content can specifically be text content, such as an electronic data report, and the initial content at least includes the report generation time, basic information of the analyzed object, etc. The analyzed object here can be considered as the descriptive subject of all basic data in the initial content; exemplarily, in a medical scenario, the analyzed object can specifically be a patient, and the corresponding basic information can be the patient's identity information, and the corresponding initial content can specifically be the patient's pathological data report, and the basic data contained therein can be the patient's various examination data, surgical conditions, etc.; in a construction project scenario, the analyzed object can be the main building being constructed, and the corresponding initial content can specifically be a construction monitoring report, and the corresponding basic data contained therein can be various measurement data during the construction process.

[0032] After receiving the initial content, the analysis system inputs the initial content into a pre-built analysis model, where the analysis model is specifically an AI big model, which is used to extract relevant features from the initial content for analysis. The specific analysis steps are as follows: S102, based on a pre-built decision tree, decision nodes contained in the decision tree, and several preset basic questions corresponding to the decision nodes, the initial content is disassembled and analyzed through an analysis model to determine the answer content associated with the basic questions; Among them, "by analyzing the model, disassembling the initial content and analyzing and determining the answer content related to the basic question" in S102 specifically includes the following steps: S1021, preliminarily determining a question set, and disassembling the answer content associated with the basic questions included in the question set from the initial content through the analysis model; wherein the question set includes several basic questions; S1022, whenever it is determined that an answer content associated with the target basic question is obtained, determine whether the first answer content meets a preset stop analysis condition; wherein the target basic question refers to any basic question in the question set, and the first answer content refers to an answer content associated with the target basic question; the stop analysis condition at least includes: the decision node corresponding to the first answer content does not contain a lower-level node in the decision tree; S1023, if the first answer content does not meet the preset stop analysis condition, then based on the first answer content, adjust the basic questions included in the question set so that the decision nodes corresponding to the basic questions included in the adjusted question set are all decision nodes related to the first answer content, until the stop analysis condition is met; The “adjusting the basic questions included in the question set based on the first answer content” in S1023 further includes: Based on the first answer content, the decision nodes of all lower nodes in the decision tree that are the decision nodes corresponding to the first answer content are taken as target decision nodes, and the basic questions corresponding to the target decision nodes are added to the question set; If there is a first target decision node in the target decision node, and the first target decision node satisfies: the first target decision node corresponds to a sub-decision tree, then the guiding question corresponding to the sub-decision tree is added to the question set; Whenever a second answer content associated with a target guiding question is determined, it is determined whether the second answer content satisfies a preset guiding condition. If so, the basic questions of all decision nodes contained in the sub-decision tree corresponding to the target guiding question are added to the question set.

[0033] In the implementation, the analysis system is preset with a decision tree, which is a classification model with a tree structure. Figure 2 The schematic diagram of the decision tree structure in the medical scenario shown in FIG. 1 includes a plurality of decision nodes, which are connected to each other by wires, and the decision nodes connected by the wires form a relationship between upper and lower nodes. The lowest node of each decision node is connected to a stop node, which refers to a node that does not have a directly connected lower node, such as Figure 2 The node indicated by arrow F.

[0034] The analysis system pre-stores basic questions corresponding to each decision node and a benchmark answer corresponding to each basic question. The basic questions and benchmark answers may be pre-defined manually. For example, the basic question corresponding to the decision node as shown by arrow A may be "whether to cough", and its corresponding benchmark answer may be a set, and the set content may include "cough" and "no cough".

[0035] Furthermore, the analysis system also pre-establishes and stores the association relationship between the benchmark answer corresponding to each decision node and its lower-level nodes; for example, in the benchmark answer corresponding to arrow A, the benchmark answer with the content "cough" is associated with the nodes indicated by arrows B and arrow C, and the benchmark answer with the content "no cough" is associated with the decision node indicated by arrow D.

[0036] Correspondingly, the analysis model is an AI big model, which is used to find the answer content for answering basic questions from the initial content. This operation is the disassembly of the initial content and the analysis operation after the disassembly; the answer content can be considered to be consistent with any set of content of the benchmark answers corresponding to the corresponding basic questions, and it should be noted here that the basic questions corresponding to all decision nodes in the default decision tree and their corresponding benchmark answers cover the initial content, that is, the basic questions can be used to realize the disassembly of the initial content. In addition, the benchmark answers corresponding to the basic questions mentioned in this application have unique answer content corresponding to the initial content, and the answer content is all included in the initial content.

[0037] Regarding the selection and order of basic questions, the embodiment of the present application proposes: first, a question set is constructed, and some basic questions are preliminarily selected to be added to the question set. The selected basic questions can be considered as the basic questions corresponding to the topmost decision nodes in the decision tree (i.e., the root node of the decision tree, such as the decision node directly connected to the start node). Figure 2 The decision node is shown by the arrow E.

[0038] Furthermore, the analysis system is preset with a query engine, which is used to point to the basic questions in the question set in sequence according to the upper and lower level relationships of the decision nodes corresponding to the basic contents in the question set in the decision tree, in the order of upper priority and lower level. The analysis model is used to analyze the basic questions pointed by the query engine, and whenever the analysis results are obtained (that is, the answer content related to the corresponding basic question is obtained), the corresponding basic question is deleted from the question set.

[0039] And whenever the analysis model analyzes and obtains the answer content, the analysis system will determine whether it is necessary to end the analysis operation of the analysis model based on the answer content, that is, whether the analysis answer content will meet the preset stop analysis condition. The stop analysis condition means that in the decision tree, the lower-level node of the decision node corresponding to the current answer content (hereinafter referred to as the first node) is the stop node; the lower-level node here refers to the decision node directly connected to the first node.

[0040] If the analysis stop condition is not met, the basic questions included in the question set will be adjusted according to the current answer content (i.e., the first answer content mentioned above). The specific adjustment rules can be: Whenever the analysis model obtains the answer content, the set of benchmark answers corresponding to the decision node corresponding to the answer content and the set content consistent with the answer content are determined, and combined with the association relationship described above, the basic question corresponding to the decision node associated with the set content (hereinafter referred to as the second node) is added to the question set; then, based on the added second node, it is determined that the second node corresponds to a sub-decision tree. If so, the second node is the first target decision node described above (such as Figure 2 The decision node indicated by the arrow A in the middle), at this time, the preset guiding question corresponding to the second node is added to the question set. The guiding question can be considered as the basis for determining whether it is necessary to further inquire about the basic question in the sub-decision tree. If the second answer content associated with the guiding question determined by the analysis model meets the preset guiding condition, it is assumed here that the analysis system pre-stores a benchmark answer for each guiding question. If the second answer content is consistent with the benchmark answer, it is considered that the preset guiding condition is met. Then, if the guiding condition is met, the basic questions corresponding to all decision nodes contained in the sub-decision tree corresponding to the second node are added to the question set; finally, the basic content in the query set is pointed to in sequence by the pointing order of the relevant query engine described above until the analysis stop condition is met.

[0041] In another embodiment, all decision nodes directly or indirectly connected to the first node may be used as second nodes, and the basic questions corresponding to the second nodes may be added to the question set, covering all basic questions in the original question set (in other embodiments, all basic questions in the original question set may be retained).

[0042] S103, when the first answer content meets the preset analysis stop condition, the answer content is integrated to generate an analysis result, and the analysis result is output.

[0043] In implementation, when the analysis model completes the analysis operation, the analysis system is used to merge all the answer contents obtained by the analysis model to generate an analysis result. In other embodiments, the analysis system is also used to send the analysis result to a review terminal, so that an expert can review the analysis result through the review terminal, and then send the analysis result to the review terminal for a person (such as an expert) to access the analysis system through the review terminal and view the analysis result. During the review, the analysis result can be modified manually, such as manually triggering a modification instruction to the analysis system through the review terminal and editing the analysis result. The modified analysis result obtained after the final editing is completed is the analysis result, and the analysis system user output displays the final analysis result.

[0044] Optionally, after “extracting, from the initial content, by analyzing the model, the answer content associated with the basic question pointed to by the preset query engine” in S1021, the following may also be included: In the initial content, a label is added to each answer content; wherein the label is used to represent the relationship between the answer content and the corresponding basic question.

[0045] During implementation, every time an answer content is output through the analysis model, the analysis system is used to mark the answer content in the initial content (i.e., add a label), such as underlining the answer content, adding a label in the form of an annotation, and adding corresponding basic questions in the annotation, so as to know the source of the answer content during manual review.

[0046] Optionally, the analysis method based on the AI ​​model also includes the following steps: The step of “inputting the initial content included in the analysis instruction into the pre-built analysis model” in S101 specifically includes the following content: According to the report type of the initial content in the analysis instruction, the report template is matched, and according to the report template obtained by matching, the basic data in the initial content in the analysis instruction is classified and integrated, and the initial content is input into the pre-built classification model.

[0047] In implementation, the initial content can be pre-classified according to different application scenarios to form corresponding report types (such as patient medical reports). In medical scenarios, each basic data in the default initial content corresponds to a data generation time, and the order of the data generation time can be used to reflect the order of patient medical treatment (such as the order of registration, examination, prescription / surgery). Then, when the analysis system receives the analysis instruction, it can adjust the distribution position of the basic data in the initial content according to the order of the data generation time of the basic data, ensuring that the earlier the data generation time, the higher the position in the initial content, and use this data generation order as the basis for generating the report template.

[0048] In other embodiments, in order to facilitate the analysis model to quickly find the answer content related to the basic question from the initial content, a report template can be generated in advance. When an analysis instruction is received, the report template is sent to the user who triggered the analysis instruction. For example, a report template for the initial content is in the form of a table, and input instructions are displayed in the cell corresponding to row E and column E in the cell at a specified position (such as row E and column E) to prompt the user of the basic data that should be entered in the corresponding cell, so as to limit the position of the basic data in the initial content and realize the classification and integration processing of the basic data, thereby improving the efficiency of associating the answer content.

[0049] Optionally, the analysis method based on the AI ​​model also includes the following steps: S104, receiving a manual editing instruction, adding a decision node in the decision tree based on the manual editing instruction, and setting a trigger condition for the currently added decision node according to the node attribute contained in the manual editing instruction; S105, whenever an analysis instruction is received, a preliminary analysis is performed on the initial content through the analysis model to determine all decision nodes that meet the triggering conditions, and a decision tree is constructed using the decision nodes that meet the triggering conditions.

[0050] In implementation, experts can trigger manual editing instructions through the review terminal. The triggering time is not limited to the review period, that is, experts can trigger manual editing instructions at any time to edit the nodes in the decision tree. The editing content can be specifically to add decision nodes or answer nodes, and define the trigger conditions of any decision nodes or answer nodes. The trigger condition refers to the judgment condition of whether the corresponding decision node needs to appear in the decision tree each time the decision tree is used for the preset query engine to point to and is analyzed through the analysis model. That is, every time the analysis system receives an analysis instruction and calls the decision tree, it is necessary to determine whether all decision nodes meet the trigger condition, and only select the decision nodes with the trigger condition to form the decision tree. For example, in a medical scenario, a complete decision tree can contain decision nodes corresponding to all medical subjects (such as internal medicine, surgery, otolaryngology, gynecology, etc.). The purpose of this solution is to realize the screening of the decision tree structure through the determination of the trigger condition. For example, when the analyzed object or the initial content only involves otolaryngology, then only the decision nodes related to otolaryngology are used to form the decision tree, which helps to screen out unnecessary decision nodes and improve the efficiency of inquiry analysis. It should be noted that the aforementioned “related to ENT” refers to other non-ENT medical subjects that are directly related to ENT or may cause complications and induce ENT-related diseases.

[0051] Among them, the trigger condition can be customized. For example, for medical scenarios, the trigger condition can be the applicable department of the node to which it belongs, such as otolaryngology. Otolaryngology can also be further refined to thyroid, etc., and the aforementioned content can be used as the specific content of the trigger condition. For medical scenarios, the default initial content includes content related to the trigger condition. In other embodiments, whenever an analysis instruction is received, the analysis model can be used to interact with the user who triggered the analysis instruction, that is, AI is used to output preset questions to the user, and the preset questions correspond to the trigger conditions and are pre-stored in the analysis system. The analysis model receives feedback content related to the preset questions from the user, and compares the feedback content with the trigger conditions to determine whether the corresponding trigger conditions are met. In other embodiments, when the user triggers the analysis model, the user can also select the analysis scope of the current initial content by himself or herself, or an expert can help the user set the analysis scope in advance before review. The analysis scope here can be considered as the content corresponding to the trigger condition and used to determine whether the trigger condition is met. For example, in a medical scenario, the analysis scope can be the patient's historical disease type and / or the department visited (such as otolaryngology, internal medicine), and the trigger condition can be considered as a subset of the analysis scope.

[0052] Optionally, the analysis method based on the AI ​​model also includes the following steps: Whenever an analysis instruction is received, based on the analysis result corresponding to the initial content, basic attributes of the analyzed object are determined, the basic attributes at least including a time transition type; For the initial content whose basic attribute is time-shifting, the decision node corresponding to the end of the analysis operation of the analysis model on the initial content is taken as the end node, and the change trend of the basic data in the initial content corresponding to the analyzed object over time is predicted according to the answer content corresponding to the end node, and an estimated index is generated; wherein the estimated index contains a number of basic questions, and the basic questions at least include the basic questions corresponding to the end node; Generate an analysis record based on the analysis result corresponding to the initial content, the analysis result and the corresponding estimated index, and store the analysis record; The step of “extracting the answer content associated with the basic question pointed to by the preset query engine from the initial content by analyzing the model” in S105 includes the following steps: Determine whether there is an analysis record for the analyzed object corresponding to the initial content in the historical period. If so, retrieve the most recent analysis record before the current time, and based on the analysis record, use the preset query engine to point to the basic questions contained in the estimated index in the analysis record one by one; The analysis model is used to extract the answer content associated with the basic question pointed out by the preset query engine from the initial content.

[0053] In implementation, time-shifting initial content means that the specific content of the basic data in the initial content will change with time, that is, after a certain period of time, when the initial content of the analyzed object is obtained again, the type and / or specific content of the basic data in the initial content has changed compared with the previous initial content.

[0054] For time-shifting initial content, after each analysis operation on the initial content is completed, the change trend of the initial content corresponding to the analyzed object over time is predicted according to the end node and its corresponding answer content, and an estimated index is generated; exemplarily, the analysis system may pre-store each decision node as an end node, and its corresponding estimated index, and the method of generating the estimated index may vary according to the actual application scenario of the corresponding initial content. For example, in a medical scenario, the basic data (such as vital signs data) in the initial content of the patient may change over time, such as the condition improves or worsens and relapses, etc. Therefore, the estimated index will be a set of basic questions generated under different changing conditions. The set may be determined by an expert during review, or it may be automatically generated by the analysis system. Exemplarily, the analysis system may pre-store each decision node as an end node, and its corresponding estimated index. all estimated indexes; thereafter, when the initial content of the same analyzed object is received again, the analysis system can retrieve the analysis record stored corresponding to the analyzed object in the historical period, and according to the estimated index contained in the most recent analysis record, when analyzing through the analysis model, use the preset query engine to preferentially point to the basic questions contained in the estimated index, and after completing the pointing of all basic questions in the estimated index according to the pointing order (the pointing order is the order from top to bottom of the basic questions contained in the estimated index according to the upper and lower layer relationships of the corresponding decision nodes in the decision tree), if the corresponding answer content does not meet the preset stop condition, then in accordance with the method described above, the basic question corresponding to the lower node (decision node) of the decision node corresponding to the basic question pointed to last in the estimated index is used as the basic question pointed to by the preset query engine, until the preset stop condition is met, the analysis operation is completed.

[0055] Optionally, the analysis method based on the AI ​​model also includes the following steps: Whenever an analysis instruction is received, the basic questions pointed out by the query engine are determined in real time, and when the analysis model completes the analysis operation, a query index is generated. The query index refers to the collection of all basic questions pointed out by the query engine in chronological order; The step of “extracting the answer content associated with the basic question pointed to by the preset query engine from the initial content by analyzing the model” in S105 includes the following steps: S1051, whenever the query engine updates the basic question pointed to, for the basic data corresponding to the answer content associated with the query content pointed to by the current query engine, analyze whether there is basic data that has a correlation relationship with the basic data and is not included in any answer content, wherein the correlation relationship at least includes a temporal relationship and a causal relationship; S1052, if it exists, then generate a logical index, and match the logical index with the query index stored in the historical period, determine the first query index containing the logical index, select the target basic question from the basic questions contained in the first query index through the preset skip query engine and point to the target basic question, so that the analysis model extracts the answer content associated with the target basic question from the initial content; wherein, the target basic question is a basic question that has not been pointed to by the query engine before the current time, the pointing priority of the skip query engine is higher than the pointing priority of the query engine, and the analysis model preferentially extracts the answer content associated with the basic question with a higher pointing priority from the initial content; The step of “selecting a target basic question from the basic questions included in the first query index and pointing to the target basic question” in S1052 specifically includes the following steps: When there is a confluence node between the first query index and other query indexes, or the first query index contains a confluence node, and the basic question corresponding to the confluence node has not been pointed to by the query engine before the current time, then any basic question is selected from the query content corresponding to all decision nodes in the first query index and located at and after the confluence node as the target basic question and pointed to the target basic question; wherein, the confluence node refers to a decision node that takes several upper-level decision nodes as input.

[0056] In implementation, the analysis system is used to determine the decision node (hereinafter referred to as the fourth node, and the fourth node is the lower node of the third node) that has an association relationship with the third node according to the basic question pointed to by the current preset query engine and its corresponding decision node (hereinafter referred to as the third node), and use the third node and the fourth node and their connection as logical indexes. For example, if the basic question corresponding to the decision node pointed to by the arrow J is the basic question pointed to by the current preset query engine, that is, the decision node pointed to by the arrow J is the third node, and its corresponding lower-level decision node is two (that is, there are two fourth nodes), which are the decision nodes pointed to by the arrow K respectively. The node and the decision node pointed to by the arrow L, then correspondingly, the decision node pointed to by the arrow J and the decision node pointed to by the arrow K and their connection constitute a logical index (hereinafter referred to as logical index one), the decision node pointed to by the arrow J and the decision node pointed to by the arrow L and their connection constitute a logical index (hereinafter referred to as logical index two), the analysis system is used to determine the corresponding first query index for each logical index, and the query index is composed of decision nodes and the connections between decision nodes; the first query index is a query index with the decision node corresponding to the logical index as the top-level node and containing all decision nodes in the corresponding logical index.

[0057] After determining the query index, select the target decision node from the decision nodes included in the query index, and the basic question corresponding to the target decision node is the target basic question. The target decision node is selected by determining whether there is a confluence node between the first query index and other first query indexes, or whether there is a confluence node in the decision nodes included in the first query index. If so, the confluence node is used as the target decision node, that is, the number of target decision nodes depends on the number of confluence nodes, and the confluence node meets the following conditions: 1. The confluence node is a decision node; 2. The number of upper-level decision nodes directly connected to the confluence node is at least 2.

[0058] After determining the confluence node, one of the decision nodes is selected from the first query index to which the confluence node belongs, from the confluence node and its lower-level nodes as the target decision node, and the skip query engine is used to point to the target basic question corresponding to the target decision node; at this time, the skip query engine and the preset query engine each point to a basic question, then according to the pointing priority, the analysis model will give priority to the basic question pointed to by the skip query engine, and extract the corresponding answer content from the initial content.

[0059] Furthermore, for any target decision node corresponding to the target basic question, once the answer content corresponding to the target basic question is found in the initial content through the analysis model, the first query index of the target decision node corresponding to the target basic question is used as the second query index, and the basic question pointed to by the preset query engine is changed to: the basic question corresponding to the lower-level decision node connected to the target decision node.

[0060] The present application also discloses an analysis system based on an AI model. Figure 3 ,include: The analysis instruction acquisition module 201 is used to acquire the analysis instruction and input the initial content contained in the analysis instruction into the pre-built analysis model; The initial content analysis module 202 is used to analyze and determine the answer content associated with the basic questions by disassembling the initial content through an analysis model based on a pre-constructed decision tree, decision nodes contained in the decision tree, and several preset basic questions corresponding to the decision nodes; The analysis result review module 203 is used to integrate the answer content to generate analysis results and output the analysis results.

[0061] Optionally, the initial content analysis module 202 is further used to preliminarily determine a question set, and to extract, from the initial content, the answer content associated with the basic questions included in the question set by analyzing the model; wherein the question set includes several basic questions; Whenever it is determined that the answer content associated with the target basic question is obtained, it is determined whether the first answer content meets the preset stop analysis condition; wherein, the target basic question refers to any basic question in the question set, and the first answer content refers to the answer content associated with the target basic question; the stop analysis condition at least includes: the decision node corresponding to the first answer content does not contain a lower-level node in the decision tree; if the first answer content does not meet the preset stop analysis condition, then based on the first answer content, the basic questions included in the question set are adjusted so that the decision nodes corresponding to the basic questions included in the adjusted question set are all decision nodes related to the first answer content, until the stop analysis condition is met.

[0062] The analysis result review module 203 is used to integrate the answer content to generate the analysis result when the first answer content meets the preset analysis stop condition.

[0063] Optionally, the analysis result review module 203 is also used to, based on the first answer content, take the decision nodes of all lower-level nodes in the decision tree that are decision nodes corresponding to the first answer content as target decision nodes, and add the basic questions corresponding to the target decision nodes to the question set; if there is a first target decision node in the target decision node, and the first target decision node satisfies: the first target decision node corresponds to a sub-decision tree, then the guiding question corresponding to the sub-decision tree is added to the question set; and is also used to determine whether the second answer content satisfies a preset guiding condition whenever a second answer content associated with the target guiding question is determined. If so, the basic questions of all decision nodes contained in the sub-decision tree corresponding to the target guiding question are added to the question set.

[0064] Optionally, the initial content analysis module 202 is further configured to add a label to each answer content in the initial content, so that a human can manually learn the association between the answer content and the basic question based on the label.

[0065] Optionally, a report template creation module is further included, which is used to determine whether there is an association relationship between the decision nodes corresponding to the decision tree, and if so, generate a report template based on the association relationship; The analysis instruction acquisition module 201 is also used to classify and integrate the basic data in the initial content in the analysis instruction when a report template exists, and then input the initial content into a pre-built classification model.

[0066] Optionally, a decision tree adjustment module is used to receive manual editing instructions, add decision nodes to the decision tree based on the manual editing instructions, and set the trigger conditions of the currently added decision nodes according to the node attributes contained in the manual editing instructions; it is also used to update the decision tree using all decision nodes that meet the trigger conditions before extracting the answer content associated with the basic question pointed to by the preset query engine from the initial content through the analysis model based on the pre-built decision tree and the basic question corresponding to each decision node in the decision tree, so that the updated decision tree only contains decision nodes that meet the corresponding trigger conditions.

[0067] Optionally, it also includes an analysis record storage module, which is used to determine the basic attributes of the analyzed object based on the analysis results corresponding to the initial content every time an analysis instruction is received, and the basic attributes at least include time transition type; it is also used to, for the initial content whose basic attributes are time transition type, use the decision node corresponding to the end of the analysis operation of the analysis model on the initial content as the end node, and predict the change trend of the initial content corresponding to the analyzed object over time according to the answer content corresponding to the end node, and generate an estimated index; wherein the estimated index contains a number of basic questions, and the basic questions at least include the basic questions corresponding to the end node; it is also used to generate an analysis record from the analysis results corresponding to the initial content, the analysis results and the corresponding estimated index, and store the analysis record; The decision tree adjustment module is also used to determine whether there is an analysis record for the analyzed object corresponding to the initial content in the historical period. If so, the most recent analysis record before the current time is retrieved, and based on the analysis record, the preset query engine is used to point to the basic questions contained in the estimated index in the analysis record one by one; the answer content associated with the basic question pointed to by the preset query engine is extracted from the initial content through the analysis model. An embodiment of the present application also discloses an analysis device based on an AI model. The analysis device based on the AI ​​model includes a memory and a processor. The memory stores a computer program that can be loaded by the processor and executes the above-mentioned AI model-based analysis method.

[0068] An embodiment of the present application also discloses a computer-readable storage medium, which stores a computer program that can be loaded by a processor and executes the above-mentioned AI model-based analysis method. The computer-readable storage medium includes, for example: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0069] It should be noted that, in this document, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0070] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit the protection scope of the application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

Claims

1. An analysis method based on an AI model, characterized in that: include: Obtaining an analysis instruction, and inputting initial content contained in the analysis instruction into a pre-built analysis model; Based on a pre-constructed decision tree, decision nodes contained in the decision tree, and several preset basic questions corresponding to the decision nodes, the initial content is disassembled and analyzed through the analysis model to determine the answer content associated with the basic questions; The answer contents are integrated to generate analysis results, and the analysis results are output.

2. The analysis method based on the AI ​​model according to claim 1, characterized in that: The step of analyzing the initial content by the analysis model to determine the answer content associated with the basic question includes: Preliminarily determine a set of questions, and disassemble the answer content associated with the basic questions included in the set of questions from the initial content through the analysis model; wherein the set of questions includes several basic questions; Whenever it is determined that an answer content associated with a target basic question is obtained, it is determined whether the first answer content satisfies a preset stop analysis condition; wherein the target basic question refers to any basic question in a question set, and the first answer content refers to an answer content associated with the target basic question; the stop analysis condition at least includes: a decision node corresponding to the first answer content does not contain a lower-level node in the decision tree; If the first answer content does not meet the preset stop analysis condition, then based on the first answer content, the basic questions included in the question set are adjusted so that the decision nodes corresponding to the basic questions included in the adjusted question set are all decision nodes related to the first answer content until the stop analysis condition is met; The step of integrating the answer contents to generate analysis results includes: When the first answer content meets the preset analysis stop condition, the answer content is integrated to generate an analysis result.

3. The analysis method based on the AI ​​model according to claim 2, characterized in that: The step of adjusting the basic questions included in the question set based on the first answer content includes: Based on the first answer content, taking the decision nodes of all lower nodes in the decision tree that are the decision nodes corresponding to the first answer content as target decision nodes, and adding the basic questions corresponding to the target decision nodes to the question set; If there is a first target decision node in the target decision node, and the first target decision node satisfies: the first target decision node corresponds to a sub-decision tree, then the guiding question corresponding to the sub-decision tree is added to the question set; Whenever a second answer content associated with a target guiding question is determined, it is determined whether the second answer content satisfies a preset guidance condition. If so, the basic questions of all decision nodes contained in the sub-decision tree corresponding to the target guiding question are added to the question set.

4. The analysis method based on the AI ​​model according to claim 2, characterized in that: The method further comprises: extracting, from the initial content, the answer content associated with the basic question pointed to by the preset query engine through the analysis model, and then: In the initial content, a label is added to each answer content; wherein the label is used to represent the association relationship between the answer content and the corresponding basic question.

5. The analysis method based on the AI ​​model according to claim 4, characterized in that: The step of inputting the initial content contained in the analysis instruction into a pre-built analysis model comprises: According to the report type of the initial content in the analysis instruction, a report template is matched, and based on the report template obtained by matching, the basic data in the initial content in the analysis instruction is classified and integrated, and the initial content is input into a pre-built classification model.

6. The analysis method based on the AI ​​model according to claim 2, characterized in that: The method further comprises: receiving a manual editing instruction, adding a decision node in the decision tree based on the manual editing instruction, and setting a trigger condition for the currently added decision node according to the node attribute contained in the manual editing instruction; Whenever an analysis instruction is received, a preliminary analysis is performed on the initial content through the analysis model to determine all decision nodes that meet the triggering conditions, and a decision tree is constructed using the decision nodes that meet the triggering conditions.

7. The analysis method based on the AI ​​model according to claim 2, characterized in that: The initial content also includes basic information of the analyzed object and the time when the report was generated; The method further comprises: Whenever an analysis instruction is received, based on the analysis result corresponding to the initial content, basic attributes of the analyzed object are determined, wherein the basic attributes at least include a time transition type; For the initial content whose basic attribute is time-shifting, the decision node corresponding to when the analysis model ends the analysis operation on the initial content is used as the end node, and according to the answer content corresponding to the end node, the change trend of the basic data in the initial content corresponding to the analyzed object over time is predicted, and an estimated index is generated; wherein the estimated index contains a number of basic questions, and the basic questions at least include the basic questions corresponding to the end node; Generate an analysis record based on the analysis result corresponding to the initial content, the analysis result and the corresponding estimated index, and store the analysis record; The extracting, from the initial content, the answer content associated with the basic question pointed to by the preset query engine by means of the analysis model includes: Determine whether there is an analysis record for the analyzed object corresponding to the initial content in the historical period. If so, retrieve the most recent analysis record before the current time, and based on the analysis record, use a preset query engine to point to basic questions included in the estimated index in the analysis record one by one; The analysis model is used to extract answer content associated with the basic question pointed out by the preset query engine from the initial content.

8. An analysis system based on an AI model, characterized in that: include, An analysis instruction acquisition module (201) is used to acquire an analysis instruction and input the initial content contained in the analysis instruction into a pre-built analysis model; An initial content analysis module (202) is used to analyze and determine the answer content associated with the basic questions by disassembling the initial content through the analysis model based on a pre-constructed decision tree, decision nodes contained in the decision tree, and a number of preset basic questions corresponding to the decision nodes; The analysis result review module (203) is used to integrate the answer content to generate an analysis result and output the analysis result.

9. An analysis device based on an AI model, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute any one of the methods according to claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute any one of the methods as claimed in claims 1 to 7.

Images