Report generation method and device based on large model, medium and equipment

Abstract

The invention relates to the technical field of report generation, in particular to a report generation method and device based on a large model, a medium and equipment, and aims to judge the problem depth evolution direction by using the time sequence semantic reasoning capability of a preset large model on the basis of collecting interaction data of a full interaction round. Therefore, the interaction report can reveal whether the user demand is deepened or generalized from a dynamic perspective, and the round order is converted into the information adoption degree according to the problem depth trend, so that the information value evaluation accords with the real demand evolution logic of the user, the core information is prevented from being submerged due to the traditional average weight distribution, and the user experience is improved. And finally, deeply fusing interaction data, information adoption degree and a preset report template, and carrying out content generation and structure arrangement by utilizing a preset large model, so that the interaction report not only can reserve original details of user interaction, but also can highlight the key information value, and meanwhile, format standardization and insight depth are realized, and the user experience is improved. And a high-quality interaction report with high pertinence and high speciality is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of report generation, in particular to a report generation method and device based on a large model, a medium and equipment. BACKGROUND

[0002] With the maturity of large model technology in the field of natural language generation, the report generation method based on a large model has been widely used in customer service dialogue summary, medical inquiry record, enterprise meeting minutes and other scenes.

[0003] The existing report generation method based on a large model usually adopts a single input generation mode. For example, in the patent with the publication number CN119203966A and the patent name "A large model intelligent report generation method, device, electronic equipment and medium", based on the natural language dialogue of the target object, a structured query statement is generated by combining meta information through a large language model, an intelligent report request is generated based on the structured query statement, and a report text is generated based on a preset template through a template engine. Since only single-round queries or isolated texts are processed, the semantic evolution logic in multi-round interactions cannot be captured, and the content structure cannot be adjusted according to the depth trend of user interaction, the information value in the interaction process is not quantitatively analyzed, resulting in insufficient professional and targeted reports.

[0004] Therefore, when generating a report based on a large model, how to improve the professional and targeted summary of user interaction in the report becomes a problem to be solved. SUMMARY

[0005] To solve the above technical problems, the technical solution adopted by the present application is a report generation method based on a large model, which includes the following steps: S100, obtaining interaction data between a target client and a query platform for a plurality of interaction rounds, wherein the interaction data of each interaction round includes a target query statement input by the target client into a trained preset large model of the query platform, and a target analysis result text of the preset large model for the target query statement.

[0006] S200, obtaining a problem depth trend corresponding to the target client according to all target query statements corresponding to all interaction rounds, wherein the problem depth trend is positive or negative.

[0007] S300, mapping an information adoption degree corresponding to each interaction round according to the problem depth trend corresponding to the target client.

[0008] S400, inputting the interaction data, the information adoption degree corresponding to each interaction round and a preset report template into the preset large model to obtain an interaction report between the target client and the query platform.

[0009] The application further provides a large model-based report generation device, which comprises: A data interaction module is configured to acquire interaction data between a target client and a query platform for a plurality of interaction rounds, wherein the interaction data of each interaction round comprises a target query statement input by the target client into a trained preset large model of the query platform, and a target analysis result text of the preset large model for the target query statement.

[0010] A trend analysis module is configured to acquire a problem depth trend of the target client according to all target query statements corresponding to all interaction rounds, wherein the problem depth trend is positive or negative.

[0011] A degree mapping module is configured to map an information adoption degree corresponding to each interaction round according to the problem depth trend of the target client.

[0012] A report generation module is configured to input the interaction data, the information adoption degree corresponding to each interaction round and a preset report template into the preset large model, and acquire an interaction report between the target client and the query platform.

[0013] The application further provides a non-transitory computer-readable storage medium, which stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the large model-based report generation method of any one of the above.

[0014] The application further provides an electronic device comprising a processor and the above non-transitory computer-readable storage medium.

[0015] The application has at least the following beneficial effects: by collecting user query and platform response data of all interaction rounds, the interaction data not only retains the consistency of user identity, but also dynamically adapts the changes in the problem field, thereby providing high-quality interaction data for analyzing the dynamic changes in user demand; on the basis of the target query statements of all interaction rounds, the time sequence semantic reasoning capability of the preset large model is used to determine the evolution direction of the problem depth, so that the interaction report can reveal whether the user demand is deepening or generalizing from a dynamic perspective; the round order is converted into the information adoption degree according to the problem depth trend, so that the information value evaluation conforms to the real demand evolution logic of the user, and the core information is prevented from being submerged due to the traditional average allocation of weights; finally, the interaction data, the information adoption degree and the preset report template are deeply fused, the preset large model is used for content generation and structure arrangement, so that the interaction report can retain the original details of user interaction, highlight the key information value, realize format standardization and insight deepening, and provide a high-quality interaction report with high pertinence and high professionalism for the target client. Attached Figure Description

[0016] 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.

[0017] Figure 1 A flowchart of a report generation method based on a large model provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of a report generation device based on a large model provided in Embodiment 2 of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It is understood that, where appropriate, the terms used to distinguish similar objects can be interchanged so that the invention can also be implemented in other embodiments besides the illustrated or described embodiments. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0020] Example 1 like Figure 1 As shown in the figure, this embodiment provides a report generation method based on a large model, which includes the following steps: S100, acquire the interaction data between the target client and the query platform for several interaction rounds, wherein the interaction data for each interaction round includes the target query statement input by the target client into the pre-trained preset large model in the query platform, and the target analysis result text of the preset large model for the target query statement.

[0021] The target client is a terminal initiated by a user to query, such as a mobile phone APP, a web interface, etc., and is used as an entry for human-computer interaction to convert natural language input by the user into a digital signal that can be processed.

[0022] The target query statement is a specific problem text submitted by a user to the query platform through the target client, and is the original input for semantic analysis by the preset large model.

[0023] The query platform is a system carrier integrating a large model, a label library and a knowledge base, and is used to provide a hardware and software environment for query processing and to coordinate the cooperative operation of various task modules such as label generation and knowledge base matching.

[0024] The trained preset large model can be an artificial intelligence model trained by text generation, which can understand natural language instructions and generate text meeting the format requirements. Those skilled in the art know that any large language model and its training method in the prior art fall within the protection scope of the present application, such as GPT series, ERNIE, etc., which will not be described here.

[0025] The target analysis result text is structured answer content generated by the preset large model of the query platform for the target query statement, and is the direct output of the query platform responding to the user's demand.

[0026] The interactive round is a continuous question and answer unit between the user and the query platform through the target client. From the initiation of the first query by the user to the termination of the interaction by the user, at least one set of target query statements and corresponding target analysis result texts are included. As the smallest time sequence unit of the report analysis, the interactive round reflects the dynamic changes of the user's demand through the content association between rounds, and avoids the one-sidedness caused by the isolated question and answer in the interaction report.

[0027] By recording the question and answer content of all interactive rounds, the above avoids the information break caused by the single round of question and answer, and provides high-quality interaction data for analyzing the dynamic changes of the user's demand.

[0028] In a specific embodiment, S100 includes the following steps: S110, inputting the first target query statement of the target client into the trained preset large model of the query platform to obtain the first label and the second label corresponding to the first target query statement, wherein the first label is used to represent the identity field corresponding to the first target query statement, and the second label is used to represent the problem field corresponding to the first target query statement.

[0029] S120, match the first label with a preset label library in the query platform to obtain a third label corresponding to the first target query statement, wherein the preset label library comprises a mapping relationship between the first label and the third label, and each third label corresponds to a plurality of query knowledge bases in the query platform.

[0030] S130, input the first target query statement, the first label corresponding to the first target query statement, the second label and the third label corresponding to the first target query statement into the preset large model to obtain a first target analysis result text corresponding to the first target query statement.

[0031] S140, obtain a j+1th target query statement of the target client for the jth target analysis result text, wherein j>0.

[0032] S150, input the j+1th target query statement into the preset large model to obtain a second label corresponding to the j+1th target query statement.

[0033] S160, input the j+1th target query statement, the second label corresponding to the j+1th target query statement, and the first label and the third label corresponding to the first target query statement into the preset large model to obtain a j+1th target analysis result text corresponding to the j+1th target query statement.

[0034] S170, update j=j+1, and repeat step S140 until the target client ends the interaction with the query platform, to determine the target query statement and the target analysis result text between the target client and the query platform for each interaction round as the interaction data between the target client and the query platform for each interaction round.

[0035] Since the same field problem, similar problem or the same problem has significant differences in attention points due to different user identities, the preset large model needs to give a targeted reply in combination with the identity information of the user who proposes the target query statement, so as to improve the information interaction quality between the user and the large model.

[0036] Therefore, by means of the preset large model, the deep semantic of the target query statement is analyzed in a double-channel manner, and the synchronous identification of the user identity and the problem type is realized. Specifically, the preset large model extracts the identity identifier from the target query statement through the identity feature library learned by massive text training, and simultaneously identifies the core theme of the query relying on the problem field classification capability, and finally outputs the structured first label and second label, thereby laying a foundation for subsequent accurate knowledge base matching.

[0037] The first label is used to accurately depict the identity attributes of the user asking the question, including core features such as occupation, role, and permission level, and is the key basis for identifying user background and potential needs. It can narrow the service range through identity dimension and ensure that the subsequent response of the preset large model meets the user's professional background, knowledge reserve, or permission limit. For example, the first label can be a professional identity such as financial analyst, enterprise employee, pediatrician, high school teacher, and software engineer, a role identity such as student, patient, and enterprise manager, and a permission identity such as VIP user, internal employee, and visitor.

[0038] The second label is used to define the subject ownership of the question asked by the user, focusing on the professional field, business category, or knowledge category involved in the question itself, and is the core basis for matching corresponding knowledge base and professional resources. It can lock the information range through the theme dimension and ensure the professionalism and pertinence of the subsequent response content of the preset large model. For example, the second label can be a subject field such as finance, education, and medicine, a business field such as human resources, law, and technology, and a scenario field such as product consultation, fault troubleshooting, policy interpretation, and process handling.

[0039] The preset label library is a database for storing label mapping rules and knowledge base association relationships, which is constructed using a graph database structure (such as Neo4j) to support efficient label association queries and is used to abstract user identity as a knowledge base access path. The knowledge base is a structured knowledge unit collection organized by domain and theme, stored using a vector database (such as Milvus) to support semantic retrieval.

[0040] The third label is an intermediate semantic layer connecting the user identity (first label) and knowledge resources (knowledge base) set by the query platform, which is used to map the abstract identity features of the user to specific knowledge access paths. Specifically, the preset label library pre-stores a mapping table that maps the first label to the third label, for example, financial analyst is mapped to financial research, individual investor is mapped to investment and finance, bank credit employee is mapped to credit risk control, patient is mapped to health consultation, and education manager is mapped to education decision.

[0041] A third label corresponds to several query knowledge bases in the query platform. For example, financial research corresponds to the macroeconomic database, industry analysis report database, financial policy and regulation database, and academic journal paper database of the financial sector in the query platform. Investment and financial planning correspond to the fund product database, stock market database, and financial planning guide database in the query platform. Credit risk control corresponds to the enterprise credit database, industry risk assessment model database, and credit approval case database in the query platform. Health consultation corresponds to the disease popularization database, medication guide database, and hospitalization process guide database in the query platform. Education decision-making corresponds to the education statistics database, policy and regulation database, and school management case database in the query platform.

[0042] A query knowledge base in the query platform can also be referenced by multiple third labels. For example, the basic medical database corresponds to both clinical diagnosis and treatment and medical education.

[0043] The first label of the target query statement can be mapped to multiple query knowledge bases, thereby filtering the query range from the full knowledge base to the subset strongly related to the user's identity through the identity of the first label, and integrating relevant knowledge resources scattered in multiple databases into a query knowledge base set for a specific identity, avoiding blind search in the full knowledge base.

[0044] The mapping relationship between the first label and the third label can be generated by training the historical interaction data between the user and the preset large model. The correspondence between the third label and the query knowledge base can be set by the implementer in the query platform according to the actual situation.

[0045] The target query statement and the three types of labels are fused and input into the preset large model, providing basic semantics ("what is being queried") through the target query statement, constraining the output perspective ("who's perspective is being answered from") through the first label (user identity), limiting the field range ("in which professional field is the answer being provided") through the second label (query topic), and anchoring the knowledge source ("based on which knowledge base is the content being generated") through the third label. This achieves precise text generation under the four-dimensional constraints of "user's specific needs, user's identity, query topic, and knowledge scene resources." The second target query statement and subsequent target query statements can directly use the first label corresponding to the first target query statement, which represents the user's identity, and the third label, which anchors the knowledge source, thereby eliminating redundant steps and improving data acquisition efficiency.

[0046] The preset large model converts the above four types of inputs into structured generation rules through the pre-trained domain adaptation capability. Specifically, the knowledge base resources associated with the third label are preferentially called, the target query statement is logically disassembled and deeply analyzed around the theme boundary of the second label from the professional perspective corresponding to the first label, and the generated target analysis result text meets the four conditions of "semantic matching with query requirements, perspective matching with user identity, content covering the theme range, and argument source from associated knowledge base", thereby avoiding generating irrelevant or exceeding the authority information, and improving the information interaction quality between the target client and the large model.

[0047] Correspondingly, the target analysis result text is a structured analysis content generated by the preset large model based on the cooperative input of the target query statement, the first label, the second label and the third label, directly responding to the user's query demand after domain knowledge reasoning and multi-dimensional constraint.

[0048] In this embodiment, based on the interaction rule that the user's subsequent query depends on the previous answer, the user's feedback or deepening demand for the jth analysis result is captured, the semantic association between the subsequent query and the previous answer is ensured, and the continuous semantic chain for analyzing the evolution trend of the user's demand is provided by the associated record from the jth result to the j+1th query, thereby avoiding mixing irrelevant queries into the same interaction process.

[0049] The problem domain of the user's subsequent query may be related to the initial domain or locally extended, so the second label needs to be updated to adapt to the new problem domain, but the first label representing the user's identity and the third label anchoring the knowledge source remain unchanged, thereby ensuring the stability of the identity and the knowledge scene by reusing the first label and the third label, and ensuring the response to adapt to the new demand by updating the second label, forming a response mode of stable framework + dynamic content, and avoiding the disconnection between the subsequent answer and the initial identity or knowledge domain.

[0050] The specific condition for the target client to end the interaction with the query platform can be set by the implementer according to the actual situation, for example, the target client can input a preset ending text, or the target client does not input a new problem text within a preset time period, then the interaction is determined to be ended, and the interaction data for each interaction round is constituted by the target query statement and the target analysis result text for each interaction round. The preset ending text and the preset time period can be set by the implementer according to the actual situation, for example, the preset ending text can be texts such as "understood, thank you" and "understood, thank you", and the preset time period can be 3 minutes, 5 minutes, 10 minutes, etc.

[0051] The above, through the two-dimensional positioning of the user identity and the problem theme formed by the first label and the second label, the two synergistically greatly reduces the knowledge base search range. Through the mapping mechanism of mapping the first label in the preset label library to the third label in the query platform, the user identity is accurately associated with the query knowledge base, the search range is reduced from the full knowledge base to the identity-related knowledge base subset, and the relevant knowledge resources scattered in multiple databases are integrated into a query knowledge base set for a specific identity, significantly improving the retrieval efficiency and analysis accuracy of the preset large model. By inputting the four types of input into the preset large model, the preset large model generates a target analysis result text that meets the four conditions of "semantic matching query demand, perspective matching user identity, content covering theme range, and argument derived from associated knowledge base" under the four-dimensional constraints of "user specific demand, user identity, query theme, and knowledge scene resource", so that the analysis result fits the user's cognitive level and professional demand, and avoids generating generalized content that deviates from actual demand, improving the information interaction quality between the target client and the large model. Based on the interactive law that the user's subsequent query depends on the previous answer, the iteration interaction is carried out, and the first label and the third label are reused to ensure the stability of the identity and the knowledge scene in the iteration process, and the second label is updated to ensure the response to new demands, forming a stable framework + dynamic content response mode, avoiding the disconnection between the subsequent answer and the initial identity or knowledge field, and ensuring the high quality of the interaction data, providing a high-quality data basis for generating in-depth, logical, and targeted reports.

[0052] S200, according to all target query sentences corresponding to all interaction rounds, obtaining a problem depth trend corresponding to the target client, wherein the problem depth trend is positive or negative.

[0053] The problem depth trend is the change direction of the semantic complexity and professional depth of the user's question in multiple rounds of interaction, including positive and negative, wherein positive represents that the user's question progresses from surface information to deep demand, that is, the question becomes more and more in-depth, specific, and professional, for example, from "what is high blood pressure" to "medication contraindications for high blood pressure patients". Negative represents that the user's question degenerates from deep demand to surface information, that is, the question becomes more and more superficial, broad, and basic, for example, from "specific treatment plan" to "disease definition". The problem depth trend can be used to quantify the change in the depth value of the information in the user interaction process, and provide a quantitative basis for the information adoption degree corresponding to each interaction round.

[0054] In a specific embodiment, S200 includes the following steps: S210, obtaining the round order of the target query sentence corresponding to each interaction round.

[0055] S220, input all target query statements corresponding to all interaction rounds, round order of each target query statement and preset prompt word into preset large model, and obtain problem depth trend corresponding to the target client.

[0056] The round order is the sequential number of each interaction round in the whole interaction process, which is used as a time axis for semantic analysis, so that the preset large model can determine the evolution direction of the user's problem. The form of the round order can be set by the implementer according to the actual situation. For example, the round order can be the first round, the second round, etc.

[0057] The preset prompt word is a pre-set instruction text for guiding the preset large model to complete the trend judgment task. In this step, the preset prompt word can be "according to the input target query statement and corresponding round order, determine the problem depth trend corresponding to the target client, and output the positive or negative trend result".

[0058] The above, by recording the round order, a time sequence anchor point is provided for trend analysis. By combining the semantic connotation of the target query statement and the time sequence relationship of the round order, the trend direction of the problem depth is determined, the trend misjudgment problem caused by only looking at the content without looking at the order in the traditional keyword analysis is solved, and the positive and negative results are output to reflect the real demand evolution of the user. Key behavior feature data is provided for the information adoption degree corresponding to each interaction round in the subsequent.

[0059] S300, according to the problem depth trend corresponding to the target client, the information adoption degree corresponding to each interaction round is mapped.

[0060] The information adoption degree is the acceptance possibility of the user to the information provided by the query platform in a certain interaction round. The higher the information adoption degree is, the more likely the information of the interaction round is adopted by the user and affects the decision, so that in the report generation, the interaction content with high adoption degree is preferentially displayed or analyzed, thereby solving the limitations of traditional average allocation of information weight or only looking at the single round interaction effect.

[0061] In a specific embodiment, S300 includes the following steps: S310, according to the problem depth trend corresponding to the target client, a third mapping rule corresponding to the target client is obtained.

[0062] S320, according to the third mapping rule, the round order of each target query statement is mapped to the information adoption degree.

[0063] In a specific embodiment, S310 includes the following steps: S311, if the problem depth trend corresponding to the target client is positive, then the round order and the information adoption degree in the third mapping rule corresponding to the target client are positively correlated.

[0064] S312, if the problem depth trend corresponding to the target client is negative, then the round order and the information adoption degree in the third mapping rule corresponding to the target client are negatively correlated.

[0065] The third mapping rule is a rule defining the correspondence between the round order and the information adoption degree, determined by the problem depth trend, and used to ensure that the calculation of the information adoption degree conforms to the user demand evolution logic.

[0066] Specifically, under the positive trend, the user demand gradually focuses, and the later interaction is closer to the core demand, so the later the round order, the higher the information adoption degree. Therefore, the positive correlation mapping under the positive trend enables the interaction report to focus on the core information that the user ultimately focuses on, avoiding interference from the front-end preparatory content.

[0067] Under the negative trend, the user demand diverges or deviates, and the early interaction can better reflect the core demand, so the earlier the round order, the higher the information adoption degree. Therefore, by negative correlation mapping, the information value of the early interaction is preserved, avoiding the core demand being ignored due to the user's later deviation.

[0068] Through the third mapping rule, the round order is converted into a quantifiable information adoption degree. For example, under the positive trend, the information adoption degree X1 = 1 / (1+exp(-B)), and under the negative trend, the information adoption degree X1 = exp(-B), where X1 is the information adoption degree, B is the round order, and exp() is the exponential function with e as the base. Correspondingly, the value range of the information adoption degree is (0, 1).

[0069] Through the above-mentioned dynamic mechanism of determining the mapping rule according to the trend and converting the round order into a quantifiable information adoption degree through the mapping rule, in the positive trend, the positive correlation mapping ensures that the key information in the deepening process of the user demand obtains a high adoption degree, avoiding interference from the front-end preparatory content, and in the negative trend, the negative correlation mapping preserves the information value of the early core demand of the user, avoiding the core information being ignored due to the later deviation, so that the interaction report can accurately capture the core content that the user ultimately focuses on, thereby improving the pertinence and professionalism of the report in summarizing the user interaction.

[0070] S400, input the interaction data, the information adoption degree corresponding to each interaction round, and the preset report template into the preset large model to obtain an interaction report between the target client and the query platform.

[0071] Among them, the interaction data as the original material of the interaction report provides the specific content of the target client's question and the platform's answer, the information adoption degree as the value filter of the interaction report guides the preset large model to preferentially display the interaction content with high adoption degree, and avoids irrelevant information interference, and the preset report template as the structural framework of the interaction report defines the chapter framework and content emphasis of the interaction report, such as the financial report focusing on investment strategy and the medical report focusing on diagnosis and treatment suggestions, so that the interaction report can not only retain the original details of user interaction, but also highlight the key information value.

[0072] The preset report template is a pre-designed framework document that specifies the report structure and content direction, including fixed chapters (such as interaction overview, demand analysis, conclusion and suggestion, etc.) and variable filling areas (such as user question statistics, trend charts, etc.), ensuring that the interaction reports of different clients have a unified format, facilitating reading and comparison.

[0073] Correspondingly, the interaction report is an analysis document generated by the preset large model based on interaction data, information adoption degree and preset template, including user demand interpretation, information value evaluation, trend prediction, etc., providing interaction summary for users, improving user experience, and providing user behavior insight for platform operators, guiding content optimization.

[0074] As mentioned above, based on the information adoption degree, the content is screened to ensure that the interaction report focuses on the core information that users are really interested in, avoiding the deviation caused by subjective selection, and through the semantic analysis capability of the preset large model, the potential patterns in the interaction data are mined to provide high-quality interaction reports with high pertinence and high professionalism for target clients.

[0075] In a specific embodiment, the report generation method based on the large model further includes the following steps: S10, for any first label, obtain the interaction data of a plurality of interaction rounds of a plurality of preset users for the current first label, wherein the interaction data of each interaction round includes a plurality of sets of interaction texts between a preset user and a preset large model, and each set of interaction texts includes question text of the preset user and answer text of the preset large model for the question text.

[0076] S20, extracting keywords from each set of interaction texts to obtain a set of interaction keywords corresponding to each set of interaction texts.

[0077] S30, analyzing all sets of interaction keywords corresponding to each interaction round to obtain a question depth trend corresponding to each interaction round of the preset user, wherein the question depth trend is positive or negative.

[0078] S40, according to the question depth trend corresponding to all interaction rounds, obtaining an influence weight corresponding to each interaction keyword corresponding to the current first label.

[0079] S50, obtaining a target keyword corresponding to the current first label according to an influence weight corresponding to each interaction keyword and a number of occurrences of each interaction keyword.

[0080] S60, mapping a third label corresponding to the target keyword in a preset label library as a third label corresponding to the current first label.

[0081] S70, traversing all first labels to obtain a third label corresponding to each first label.

[0082] The interaction round is a continuous interaction process between the same preset user and the preset large model around a certain theme, including a complete cycle from the first question to the end of the dialogue. Since the preset large model shows that different users with the same identity label have commonalities in the types of questions and focuses when interacting with the preset large model, these commonalities can be extracted from the interaction data. Therefore, by analyzing the dynamic change process of user demand through continuous dialogue, time sequence dimension data for analyzing the problem depth trend is provided.

[0083] The preset user is a typical user selected to train the label mapping relationship, covering the identity field corresponding to the first label, used to extract common rules from the interaction data of the typical user, ensure the universality of the label mapping rule, and avoid interference of abnormal interaction of individual users on the analysis result. In this embodiment, representative interaction samples are screened to ensure that the data covers common interaction scenarios of the identity user, so as to avoid bias caused by single sample in subsequent analysis.

[0084] Using the keyword extraction technology in natural language processing, non-core information is stripped from the question text and answer text, and core words reflecting the interaction theme are retained, so as to ensure that the extracted keywords cover both user demand and model response, and avoid the information being one-sided caused by extracting from the question or the answer alone. Wherein, any keyword extraction technology in the prior art falls within the protection scope of the present application, such as TF-IDF, TextRank, BERT keyword model, etc., which will not be described here.

[0085] In this embodiment, the problem depth trend can represent the change direction of the semantic complexity and professional depth of the user's question in an interaction round, including forward and reverse, which is used to quantify the depth value change of information in the user interaction process and provide a basis for keyword weight calculation.

[0086] The influence weight is used to represent the contribution of the interaction keyword to the mapping relationship between the first label and the third label. The role is to filter redundant keywords and ensure that the target keyword can accurately reflect the core features of the identity field.

[0087] The target keyword is the keyword most representative in a first label, which is obtained by comprehensively considering the influence weight and the frequency of occurrence, and is used as an intermediate bridge for mapping the first label and a third label, so as to improve the matching accuracy of the label. For example, the target keyword extracted from the interaction data of the preset user with the first label of "teacher" can be "lesson plan design" and "classroom management".

[0088] The preset label library stores the mapping relationship between the keyword and the third label, and the third label corresponding to the first label is obtained by matching the target keyword with the preset label library, thereby providing a basis for limiting the professional field range.

[0089] The above method ensures the representativeness of the analysis sample by collecting interaction data of multiple preset users and multiple interaction rounds, thereby providing a reliable basis for subsequent rule extraction and avoiding mapping deviation caused by one-sided data. The method improves the extraction of interaction keywords, simplifies the dimension of text analysis, converts complex natural language into quantifiable core words, and provides structured input for subsequent calculation. The method improves the depth of the introduced problem and captures the dynamic changes of user demand, so that the subsequent analysis is more in line with the actual scene that the user demand changes with the communication in the real interaction, thereby improving the accuracy of the importance judgment of the keyword. The method improves the influence weight of the quantified keyword, avoids screening the keyword only by the frequency of occurrence, ensures that the core word has a substantial impact on the user demand, and then screens the target keyword by combining the influence weight and the frequency of occurrence, balances the importance of universality and demand deepening, and makes the target keyword accurately represent the core concern of the identity user, thereby serving as an intermediate bridge for mapping the first label and the third label, improving the matching accuracy of the label, providing a basis for further limiting the professional field range, and improving the field adaptability and response efficiency of the target analysis result text of the information interaction of different identity users.

[0090] In a specific embodiment, S10 includes the following steps: S101, for any interaction round of any preset user corresponding to the current first label, a first question text input by the current preset user to the preset large model is used to obtain a first answer text corresponding to the preset large language model.

[0091] S102, a (i+1)th question text input by the current preset user to the preset large model for the ith answer text is used to obtain an (i+1)th answer text corresponding to the preset large language model, where i>0.

[0092] S103, i is updated to i+1, and step S102 is repeatedly executed until the current preset user ends the current interaction round with the preset large language model, and all question texts and answer texts of the current first label for the current preset user in the current interaction round are determined as the interaction data of the current first label for the current preset user in the current interaction round.

[0093] S104, traversing all preset users corresponding to the current first label and all interaction rounds, obtaining interaction data of the current first label for several interaction rounds of several preset users.

[0094] Among them, the first question of the preset user is taken as the starting point of the round, and the initial dialogue context is established through the first answer of the preset large model. The subsequent questions of the preset user depend on the previous answers of the preset large model, forming multiple rounds of question and answer dialogue, constituting the semantic progressive chain in the round, and providing interaction data with time sequence dimension for capturing the dynamic evolution of user demand.

[0095] The specific condition for the preset user to end the current interaction round with the preset large language model can be set by the implementer according to the actual situation, for example, the preset user can input the preset ending text, or the preset user does not input new question text within the preset time period, then it is determined that the interaction round is ended, and all question and answer pairs in the interaction round constitute complete interaction data.

[0096] In a specific embodiment, S30 includes the following steps: S301, for any interaction round, obtaining the interaction order of the interaction text corresponding to each group of interaction keywords corresponding to the current interaction round.

[0097] S302, inputting all groups of interaction keywords corresponding to the current interaction round, the interaction order corresponding to each group of interaction keywords and the preset prompt word into the preset large model, obtaining the question depth trend corresponding to the preset user in the current interaction round.

[0098] Among them, the position of each group of interaction keywords in the round is accurately marked through the interaction order, ensuring that the time sequence relationship is not confused. The form of the interaction order can be set by the implementer according to the actual situation, for example, the interaction order can be the first group, the second group, etc.

[0099] The preset large model has time sequence semantic reasoning ability, and can judge the trend direction of the question depth by combining the semantic connotation of the interaction keyword and the time sequence relationship of the interaction order. Specifically, the preset large model understands the depth level of the keyword through the pre-trained domain knowledge, and then outputs a positive or negative conclusion by combining the change of the depth level with the interaction order. Among them, the question depth of the "definition" type keyword can be set as shallow, the question depth of the "application" type keyword can be set as medium, and the question depth of the "principle derivation" type keyword can be set as deep. Correspondingly, when the preset large model analyzes the question depth from shallow to deep, the corresponding question depth trend is positive, and vice versa, the corresponding question depth trend is negative.

[0100] The preset prompt word is a pre-set instruction text for guiding the preset large model to complete the trend judgment task. For example, the preset prompt word is "according to the input keyword and the corresponding interaction sequence, judge the problem depth trend corresponding to the interaction round, and output the positive or negative trend result".

[0101] By recording the interaction sequence, a time sequence anchor point is provided for trend analysis. By combining the semantic connotation of the interaction keyword and the time sequence relationship of the interaction sequence, the trend direction of the problem depth is judged. The trend misjudgment problem caused by only looking at the content without looking at the sequence in traditional keyword analysis is solved. The positive and negative results are output to reflect the real demand evolution of the preset user in the single round interaction. Key behavior feature data is provided for subsequent accurate mapping of labels.

[0102] In a specific embodiment, S40 includes the following steps: S401, for any interaction round, according to the problem depth trend corresponding to the current interaction round, a first mapping rule corresponding to the current interaction round is obtained.

[0103] S402, according to the first mapping rule, the interaction sequence corresponding to each group of interaction keywords in the current interaction round is mapped to the first priority corresponding to each group of interaction keywords in the current interaction round, wherein the first priority is used to represent the relative importance between different interaction keywords in an interaction round.

[0104] S403, according to the second mapping rule, the problem depth trend corresponding to the current interaction round is mapped to the second priority corresponding to each group of interaction keywords in the current interaction round, wherein the second priority is used to represent the relative importance between different interaction rounds.

[0105] S404, according to the first priority and the second priority corresponding to each group of interaction keywords in the current interaction round, the influence weight corresponding to each group of interaction keywords in the current interaction round is obtained.

[0106] In a specific embodiment, S401 includes the following steps: S4011, if the problem depth trend corresponding to the current interaction round is positive, the interaction sequence and the first priority in the first mapping rule corresponding to the current interaction round are in a positive correlation.

[0107] S4012, if the problem depth trend corresponding to the current interaction round is negative, the interaction sequence and the first priority in the first mapping rule corresponding to the current interaction round are in a negative correlation.

[0108] The first mapping rule is a corresponding relationship between the interaction order and the first priority, which is determined by the problem depth trend of the current interaction round, so as to dynamically adapt the keyword importance distribution law under different trends and avoid the deviation caused by the fixed order weight (such as the default that the later keywords are more important).

[0109] Correspondingly, in the positive trend, the user's question is more focused on the core demand as the interaction deepens, therefore, when the problem depth trend is positive, the first mapping rule corresponding thereto has a positive correlation between the interaction order and the first priority, that is, the later the interaction order, the higher the first priority. On the contrary, in the reverse trend, the user's question degenerates and deviates from the core as the interaction deepens, therefore, when the problem depth trend is reverse, the first mapping rule corresponding thereto has a negative correlation between the interaction order and the first priority, that is, the earlier the interaction order, the higher the first priority.

[0110] The first priority is the relative importance score of the keywords of different interaction orders in the same interaction round, which is converted into a quantifiable first priority through the first mapping rule. For example, in the round of the positive trend, the first priority Y1=1 / (1+exp(-C)), and in the round of the reverse trend, the first priority Y1=exp(-C), wherein Y1 is the first priority, C is the interaction order, and exp() is the exponential function with e as the base. Correspondingly, the value range of the first priority is (0, 1).

[0111] The second mapping rule is a corresponding relationship between the problem depth trend and the second priority, which is used to quantify the overall value of different rounds and ensure that the keywords reflecting the user's core demand in the positive round obtain a higher weight, thereby solving the problem of invalid interaction keyword interference caused by the traditional equal treatment of all rounds. Correspondingly, the second priority represents the overall value of different interaction rounds, and the higher the second priority, the greater the reference significance of the round to reflect the user's core demand.

[0112] The second mapping rule can be a fixed score mapping, and the specific score can be set by the implementer according to the actual situation. For example, the corresponding priority of the positive trend is 5, and the corresponding priority of the reverse trend is 3.

[0113] The product of the first priority and the second priority corresponding to each group of interaction keywords in the current interaction round is obtained to obtain the influence weight corresponding to each group of interaction keywords in the current interaction round.

[0114] The dynamic adaptation mechanism of the mapping rule through the trend determines that the keywords of the same interaction order obtain differentiated priority under different trends, solves the weight misjudgment problem caused by the traditional fixed rule in the reverse trend that the default later keyword is more important, quantifies the importance difference in the same round through the first priority, makes the keywords promoting demand deepening and the keywords laying foundation in the round be clearly distinguished, avoids the core information weight shortage caused by equal treatment of all keywords, quantifies the value difference of different rounds through the second priority, makes the keywords of the positive trend round obtain higher basic score, and the keyword weight of the reverse trend round is reasonably inhibited, solves the invalid interaction interference problem caused by the traditional fixed rule of only looking at the importance of a single round and ignoring the value of the round itself, and finally through the comprehensive calculation of the first priority and the second priority, the output influence weight is more close to the real importance order of the user core demand, and provides a precise quantitative basis for subsequent target keyword screening.

[0115] In an embodiment, S50 includes the following steps: S501, the product of the influence weight corresponding to each interaction keyword and the number of occurrences is determined as the importance degree corresponding to each interaction keyword.

[0116] S502, the maximum interaction keyword corresponding to the current first label is determined as the target keyword corresponding to the current first label.

[0117] Among them, through the comprehensive quantification of the influence weight and the number of occurrences, the most representative target keyword is selected from all the interaction keywords of the current first label, which ensures that the target keyword can not only reflect the core demand of the identity user, but also has universal representativeness, and provides a precise anchor point for the mapping of the subsequent first label to the third label, so as to ensure that the interaction of different identity users can automatically match the most relevant knowledge scene, and improve the accuracy and efficiency of the whole information interaction process.

[0118] As described above, by collecting user query and platform response data from all interaction rounds, the interaction data retains the consistency of user identity while dynamically adapting to changes in the problem domain. This provides high-quality interaction data for analyzing the dynamic changes in user needs. Based on the target query statements of all interaction rounds, the temporal semantic reasoning capability of the pre-set large model is used to determine the direction of the problem's depth evolution. This allows the interaction report to reveal from a dynamic perspective whether user needs are deepening or generalizing. Furthermore, the round order is converted into information adoption degree based on the problem depth trend, ensuring that the information value assessment conforms to the evolution logic of users' real needs. This avoids the core information being buried due to traditional average weighting. Finally, by deeply integrating the interaction data, information adoption degree, and pre-set report template, the pre-set large model is used for content generation and structural arrangement. This allows the interaction report to retain the original details of user interactions while highlighting the value of key information. At the same time, it achieves format standardization and insight depth, providing highly targeted, highly professional, and high-quality interaction reports for the target client.

[0119] Example 2 like Figure 2 As shown, this embodiment two provides a report generation device based on a large model, which includes: The data interaction module 21 is used to acquire the interaction data between the target client and the query platform for several interaction rounds. The interaction data for each interaction round includes the target query statement input by the target client into the pre-trained preset large model in the query platform, and the target analysis result text of the preset large model for the target query statement.

[0120] The trend analysis module 222 is used to obtain the question depth trend corresponding to the target client based on all target query statements corresponding to all interaction rounds, wherein the question depth trend is either positive or negative.

[0121] The degree mapping module 23 is used to map the degree of information adoption for each interaction round based on the question depth trend of the target client.

[0122] The report generation module 24 is used to input the interaction data, the information adoption level corresponding to each interaction round, and the preset report template into the preset large model to obtain the interaction report between the target client and the query platform.

[0123] In one specific embodiment, the data interaction module 21 includes: The label obtaining submodule is configured to input the first target query statement of the target client into a preset large model of the query platform to obtain a first label and a second label corresponding to the first target query statement, wherein the first label is used to represent an identity field corresponding to the first target query statement, and the second label is used to represent a problem field corresponding to the first target query statement.

[0124] The third label matching submodule is configured to match the first label with a preset label library in the query platform to obtain a third label corresponding to the first target query statement, wherein the preset label library comprises a mapping relationship between the first label and the third label, and each third label corresponds to a plurality of query knowledge bases in the query platform.

[0125] The first result text obtaining submodule is configured to input the first target query statement and the first label, the second label and the third label corresponding to the first target query statement into the preset large model to obtain a first target analysis result text corresponding to the first target query statement.

[0126] The target query statement obtaining submodule is configured to obtain a (j+1)th target query statement of the target client for a jth target analysis result text, wherein j>0.

[0127] The second label obtaining submodule is configured to input the (j+1)th target query statement into the preset large model to obtain a second label corresponding to the (j+1)th target query statement.

[0128] The second result text obtaining submodule is configured to input the (j+1)th target query statement, the second label corresponding to the (j+1)th target query statement, and the first label and the third label corresponding to the first target query statement into the preset large model to obtain a (j+1)th target analysis result text corresponding to the (j+1)th target query statement.

[0129] The interaction data obtaining submodule is configured to update j=j+1, repeatedly execute the target query statement obtaining submodule until the target client ends the interaction with the query platform, and determine the target query statement and the target analysis result text between the target client and the query platform for each interaction round as interaction data between the target client and the query platform for each interaction round.

[0130] In an embodiment, the large model-based report generation device further comprises: The interaction data acquisition module is configured to acquire, for any first label, interaction data of a plurality of interaction rounds of the current first label for a plurality of preset users, wherein the interaction data of each interaction round includes a plurality of sets of interaction texts between a preset user and the preset large model, and each set of interaction texts includes question text of the preset user and answer text of the preset large model for the question text.

[0131] The interaction keyword acquisition module is configured to perform keyword extraction on each set of interaction texts to acquire a set of interaction keywords corresponding to each set of interaction texts.

[0132] The question depth trend acquisition module is configured to analyze all sets of interaction keywords corresponding to each interaction round to acquire a question depth trend of the preset user corresponding to each interaction round, wherein the question depth trend is positive or negative.

[0133] The influence weight acquisition module is configured to acquire, according to the question depth trends corresponding to all interaction rounds, an influence weight of each interaction keyword corresponding to the current first label.

[0134] The target keyword acquisition module is configured to acquire, according to the influence weight of each interaction keyword corresponding to the current first label and the number of occurrences of each interaction keyword, a target keyword corresponding to the current first label.

[0135] The third label mapping module is configured to map a third label corresponding to the target keyword in a preset label library to a third label corresponding to the current first label.

[0136] The first label traversal module is configured to traverse all first labels to acquire a third label corresponding to each first label.

[0137] In an embodiment, the influence weight acquisition module includes: The first mapping rule acquisition submodule is configured to acquire, for any interaction round, a first mapping rule corresponding to the current interaction round according to the question depth trend corresponding to the current interaction round.

[0138] The first priority acquisition submodule is configured to map, according to the first mapping rule, an interaction order of each set of interaction keywords in the current interaction round to a first priority of each set of interaction keywords in the current interaction round, wherein the first priority is used to represent the relative importance between different interaction keywords in an interaction round.

[0139] The second priority acquisition submodule is configured to map, according to the second mapping rule, the question depth trend corresponding to the current interaction round to a second priority of each set of interaction keywords in the current interaction round, wherein the second priority is used to represent the relative importance between different interaction rounds.

[0140] An influence weight obtaining submodule is configured to obtain an influence weight corresponding to each group of interaction keywords in the current interaction round according to the first priority and the second priority corresponding to each group of interaction keywords in the current interaction round.

[0141] In an embodiment, the trend analysis module 22 comprises: A round order obtaining submodule is configured to obtain a round order of the target query statement corresponding to each interaction round.

[0142] A question depth trend obtaining submodule is configured to input all target query statements corresponding to all interaction rounds, the round order of each target query statement and a preset prompt word into a preset large model to obtain a question depth trend corresponding to the target client.

[0143] In an embodiment, the degree mapping module 23 comprises: A third mapping rule obtaining submodule is configured to obtain a third mapping rule corresponding to the target client according to the question depth trend corresponding to the target client.

[0144] An information adoption degree obtaining submodule is configured to map the round order of each target query statement into an information adoption degree according to the third mapping rule.

[0145] In an embodiment, the third mapping rule obtaining submodule comprises: A first rule determining unit is configured to determine that the round order and the information adoption degree in the third mapping rule corresponding to the target client are in a positive correlation relationship if the question depth trend corresponding to the target client is positive.

[0146] A second rule determining unit is configured to determine that the round order and the information adoption degree in the third mapping rule corresponding to the target client are in a negative correlation relationship if the question depth trend corresponding to the target client is negative.

[0147] Embodiments of the present application also provide a non-transitory computer readable storage medium, which can be arranged in an electronic device to save at least one instruction or at least one program related to a method in the method embodiment, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the method for generating a report based on a large model provided by the above-mentioned embodiment one.

[0148] Embodiments of the present application also provide an electronic device comprising a processor and the aforementioned non-transitory computer readable storage medium.

[0149] It should be noted that the information interaction, execution process and the like between the above modules are based on the same concept as the method embodiments of the present application, and the specific functions and the resulting technical effects can be referred to the method embodiments part, which will not be described here.

[0150] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the technical solution of the present application, and any simple modification, equivalent change and modification of the above embodiments according to the technical essence of the present application are still within the scope of the technical solution of the present application.

Claims

1. A report generation method based on a large model, characterized in that, The report generation method based on large models also includes the following steps: S100, acquire the interaction data between the target client and the query platform for several interaction rounds, wherein the interaction data for each interaction round includes the target query statement input by the target client into the preset large model trained in the query platform, and the target analysis result text of the preset large model for the target query statement; S200, based on all target query statements corresponding to all interaction rounds, obtain the question depth trend corresponding to the target client, wherein the question depth trend is either positive or negative; S300, based on the question depth trend corresponding to the target client, the information adoption degree corresponding to each interaction round is mapped to the result; S400, input the interaction data, the information adoption degree corresponding to each interaction round, and the preset report template into the preset large model to obtain the interaction report between the target client and the query platform.

2. The report generation method based on a large model according to claim 1, characterized in that, S100 includes the following steps: S110, input the first target query statement of the target client into the pre-trained large model in the query platform, and obtain the first label and the second label corresponding to the first target query statement, wherein the first label is used to characterize the identity domain corresponding to the first target query statement, and the second label is used to characterize the question domain corresponding to the first target query statement; S120, the first tag is matched with the preset tag library in the query platform to obtain the third tag corresponding to the first target query statement, wherein the preset tag library includes the mapping relationship between the first tag and the third tag, and each third tag corresponds to several query knowledge bases in the query platform; S130, input the first target query statement and the first label, second label and third label corresponding to the first target query statement into the preset large model to obtain the first target analysis result text corresponding to the first target query statement; S140, obtain the (j+1)th target query statement of the target client for the jth target analysis result text, where j>0; S150, input the (j+1)th target query statement into the preset large model to obtain the second tag corresponding to the (j+1)th target query statement; S160, input the (j+1)th target query statement, the second label corresponding to the (j+1)th target query statement, and the first label and third label corresponding to the first target query statement into the preset large model to obtain the (j+1)th target analysis result text corresponding to the (j+1)th target query statement; S170, update j=j+1, repeat step S140 until the target client ends its interaction with the query platform, and determine the target query statement and target analysis result text between the target client and the query platform for each interaction round as the interaction data between the target client and the query platform for each interaction round.

3. The report generation method based on a large model according to claim 2, characterized in that, The report generation method based on large models also includes the following steps: S10, for any first tag, obtain the interaction data of the current first tag for several interaction rounds for several preset users, wherein the interaction data of each interaction round includes several sets of interaction text between a preset user and a preset large model, and each set of interaction text includes the question text of the preset user and the answer text of the preset large model for the question text. S20, Extract keywords from each set of interactive texts to obtain a set of interactive keywords corresponding to each set of interactive texts; S30, Analyze all groups of interactive keywords corresponding to each interaction round to obtain the question depth trend of the preset user corresponding to each interaction round, wherein the question depth trend is positive or negative; S40: Based on the question depth trend corresponding to all interaction rounds, obtain the influence weight of each interaction keyword corresponding to the current first tag; S50: Based on the influence weight of each interactive keyword corresponding to the current first tag and the number of times each interactive keyword appears, obtain the target keyword corresponding to the current first tag; S60, map the third tag corresponding to the target keyword in the preset tag library to the third tag corresponding to the current first tag; S70, iterate through all the first tags and get the third tag corresponding to each first tag.

4. The report generation method based on a large model according to claim 2, characterized in that, S40 includes the following steps: S401, For any interaction round, obtain the first mapping rule corresponding to the current interaction round based on the question depth trend corresponding to the current interaction round; S402, according to the first mapping rule, the interaction order corresponding to each group of interaction keywords in the current interaction round is mapped to the first priority corresponding to each group of interaction keywords in the current interaction round, wherein the first priority is used to characterize the relative importance between different interaction keywords in an interaction round; S403, According to the second mapping rule, the question depth trend corresponding to the current interaction round is mapped to the second priority corresponding to each group of interaction keywords in the current interaction round, wherein the second priority is used to characterize the relative importance between different interaction rounds; S404. Based on the first and second priorities of each group of interactive keywords in the current interaction round, obtain the influence weight corresponding to each group of interactive keywords in the current interaction round.

5. The report generation method based on a large model according to claim 1, characterized in that, S200 includes the following steps: S210, obtain the round order of the target query statement corresponding to each interaction round; S220, input all target query statements corresponding to all interaction rounds, the round order of each target query statement, and preset prompt words into the preset large model to obtain the question depth trend corresponding to the target client.

6. The report generation method based on a large model according to claim 5, characterized in that, S300 includes the following steps: S310, Based on the question depth trend corresponding to the target client, obtain the third mapping rule corresponding to the target client; S320, according to the third mapping rule, the round order of each target query statement is mapped to the degree of information adoption.

7. The report generation method based on a large model according to claim 6, characterized in that, S310 includes the following steps: S311, if the problem depth trend corresponding to the target client is positive, then the round order and the degree of information adoption in the third mapping rule corresponding to the target client are positively correlated; S312, if the problem depth trend corresponding to the target client is reversed, then the round order and information adoption degree in the third mapping rule corresponding to the target client are negatively correlated.

8. A report generation device based on a large model, characterized in that, The report generation device based on the large model includes: The data interaction module is used to acquire the interaction data between the target client and the query platform for several interaction rounds. The interaction data for each interaction round includes the target query statement input by the target client into the preset large model trained in the query platform, and the target analysis result text of the preset large model for the target query statement. The trend analysis module is used to obtain the question depth trend corresponding to the target client based on all target query statements corresponding to all interaction rounds, wherein the question depth trend is either positive or negative; The degree mapping module is used to map the degree of information adoption for each interaction round according to the question depth trend corresponding to the target client. The report generation module is used to input the interaction data, the information adoption level corresponding to each interaction round, and the preset report template into the preset large model to obtain the interaction report between the target client and the query platform.

9. A non-transitory computer-readable storage medium, wherein the storage medium stores at least one instruction or at least one program segment, characterized in that, The at least one instruction or the at least one program segment is loaded and executed by the processor to implement the large model-based report generation method as described in any one of claims 1-7.

10. An electronic device, characterized in that, Includes a processor and the non-transitory computer-readable storage medium as described in claim 9.

Images