Opportunity mining method and device based on structured hints

Abstract

The application relates to the technical field of business opportunity mining, and provides a business opportunity mining method and device based on structured prompts, which comprises the following steps: constructing a structured business query library; receiving a user's business opportunity mining query; using a large language model, processing the business opportunity mining query based on a predefined structured prompt template to obtain an intent recognition result, and generating a business opportunity query sentence combined with business rules and dictionary constraints according to the intent recognition result; obtaining an initial query result from the business query library according to the business opportunity query sentence; based on the intent recognition result, performing standardized business opportunity analysis on the initial query result, and outputting a final business opportunity mining result according to a predefined format specification. The application effectively overcomes the technical defects of illusion, unstable output and uncontrollability when directly using a large language model to process business opportunity information, so that the business opportunity mining is more accurate, controllable, flexible, efficient and the like.

Description

Technical Field

[0001] This application relates to the field of business opportunity mining technology, and in particular to a business opportunity mining method and apparatus based on structured prompts. Background Technology

[0002] With the advancement of information technology development, the procurement needs of government agencies and enterprises are growing daily, generating a massive amount of bidding and procurement announcements. This information is a crucial source for businesses to acquire opportunities. However, these announcements are typically published on various websites and platforms, with inconsistent formats, unstructured content, and a large amount of irrelevant information. Traditional manual screening and analysis methods are not only inefficient and costly, but also prone to errors due to subjective judgment, making accuracy difficult to guarantee.

[0003] With the widespread application of large language models, people have begun to use them for business opportunity mining. For example, the business opportunity mining and recommendation device with patent publication number CN118840134A is based on a large language model to perform structured parsing of bidding information and extract key elements to generate business opportunity mining information for external companies. The business opportunity recommendation method, system, electronic device and medium based on LSTM model with patent publication number CN111931065A uses LSTM, CFR and Lambdamart models to obtain the final ranking of business opportunities for each target user.

[0004] However, the existing business opportunity discovery techniques have the following problems: 1. Reliance on large-scale "end-to-end parsing": Directly submitting raw, heterogeneous bidding information to a large language model for end-to-end parsing and element extraction. This approach heavily relies on the model's generalization ability. However, the model has limited understanding of complex hierarchical relationships and the deep meanings of professional terms in specific domains, which can easily lead to information extraction errors or illusions, thus reducing the accuracy and reliability of the mining results.

[0005] 2. Rigid Interaction of Traditional Recommendation Models: While existing models such as LSTM and CFR are effective for specific ranking tasks, their recommendation logic relies on structured user behavior data and pre-defined feature engineering, making the system relatively rigid. They cannot process or understand users' natural language query intent; users can only passively receive recommended business opportunities. After receiving these opportunities, users can only rely on their own experience to analyze and expand upon them, resulting in a low interactive experience and low information acquisition efficiency.

[0006] Therefore, there is currently a lack of a precise, controllable, flexible, and efficient method for identifying business opportunities. Summary of the Invention

[0007] To address the aforementioned problems in the prior art, this application provides a business opportunity mining method and apparatus based on structured prompts, making business opportunity mining more accurate, controllable, flexible, and efficient.

[0008] To achieve the above objectives, the technical solution adopted in this application is as follows: Firstly, this application provides a business opportunity mining method based on structured prompts, including: Step S1: Construct a structured business query database based on historical business data; Step S2: Receive the user's business opportunity mining query, use a large language model, based on a predefined structured prompt template, process the business opportunity mining query to obtain intent recognition results, and generate a business opportunity query statement that combines business rules and dictionary constraints based on the intent recognition results, and obtain the initial query results from the business query database based on the business opportunity query statement. Step S3: Based on the intent recognition result, perform standardized business opportunity analysis on the initial query result, and output the final business opportunity mining result according to the predefined format specification.

[0009] The beneficial effects of this application are as follows: This application controls the large language model through predefined structured prompt templates, ensuring that it strictly follows a controlled process of intent recognition, query construction, standard analysis, and standardized output during opportunity mining, and strictly limits the execution of this process to the scope of a pre-built structured business query library. This method transforms the large language model from a general open-domain text generator into a predictable query and analysis engine for closed-domain opportunity data. It effectively overcomes the technical shortcomings of directly using the large language model to process opportunity information, such as numerous illusions, unstable output, and uncontrollable behavior. It achieves end-to-end controllability and reliability in the opportunity mining process, from natural language interaction to structured result output, making opportunity mining more accurate, controllable, flexible, and efficient.

[0010] Optionally, step S2 involves processing the business opportunity mining query to obtain intent recognition results, including: The business opportunity mining query is matched with predefined intent keywords and logical judgment rules to obtain the matched deterministic intent; The key entity information associated with the deterministic intent is parsed from the business opportunity mining query to obtain a structured description of the intent recognition result, which includes the deterministic intent and key entity information.

[0011] As described above, by matching user queries with predefined keywords and rules, and forcing them to be categorized into a limited number of deterministic intent categories, the system achieves standardized and deterministic understanding of users' fuzzy and diverse natural language queries. This avoids deviations in the entire analysis chain caused by misjudgment of intent, and further improves the accuracy and controllability of business opportunity mining.

[0012] Optionally, step S2, which generates a business opportunity query statement combining business rules and dictionary constraints based on the intent recognition result, includes: Based on the deterministic intent, the mandatory business rule corresponding to the deterministic intent is called from the preset rule base; The key entity information is matched and validated using a predefined field dictionary, which defines the queryable data fields and their corresponding legal data acquisition rules. The mandatory business rules, the key entity information after matching and verification, and the fixed query field templates are combined and concatenated to generate a structured database query statement as a business opportunity query statement.

[0013] As described above, by invoking mandatory business rules based on deterministic intent and using a field dictionary to validate and constrain entity information, the correctness of the generated database query statements in terms of business logic and the security of data access scope are ensured. This prevents the large language model from generating invalid, unauthorized, or non-business-logical queries, thus guaranteeing the accuracy and compliance of the results from the source of data retrieval.

[0014] Optionally, step S3, based on the intent recognition result, performs standardized business opportunity analysis on the initial query result, including: Based on the intent recognition results, select the corresponding standardized analysis task template. The analysis output items defined by the standardized analysis task template include at least one or more of the following: project information overview list, key business opportunity value assessment, adaptive solution strategy suggestions, and industry trend insight report. Based on the initial query results, the large language model is controlled to perform analysis according to the selected analysis task template.

[0015] Optionally, step S3, which involves performing standardized opportunity analysis on the initial query results based on the intent recognition result, further includes: The structured prompt template sends constraint instructions to the large language model, which require that it must and can only be based on the factual information contained in the initial query results, prohibiting the introduction of external knowledge or fictitious data, and requiring the large language model to clearly indicate insufficient data in the output when the initial query results are insufficient to draw a definite conclusion.

[0016] As described above, by imposing strict constraints on the large language model during the analysis phase, requiring it to reason solely based on the initial query results, the final business opportunity mining results are built upon verifiable data facts, significantly improving the credibility and interpretability of the analysis conclusions.

[0017] Optionally, step S2 further includes: When the key entity information is missing a preset key field, the missing key field is set according to the user's relevant information; When the content of the key entity information cannot find a valid match in the predefined field dictionary, an output field recognition error message will be displayed.

[0018] Optionally, step S1 includes: Based on a pre-configured industry keyword library and negative keyword library, historical business data is collected from multiple public data sources; The historical business data is preprocessed; Based on a pre-defined industry classification system and administrative division knowledge base, the pre-processed historical business data is identified and categorized according to industry and regional attributes. Calculate and identify the characteristic indicators of historical business data in four dimensions: time, space, technology, and scenario. The structured data containing the aforementioned feature indicators is persistently stored to form a commercial query library that serves as the sole data source for subsequent queries and analysis.

[0019] Optionally, the predefined format specifications in step S3 include: field name format specifications, query field sorting rules, and query data format specifications.

[0020] Optionally, the deterministic intent includes at least one of business opportunity search, case search, and trend analysis.

[0021] Secondly, this application provides a business opportunity mining device based on structured prompts, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the business opportunity mining method based on structured prompts provided in the first aspect.

[0022] The technical effects of the business opportunity mining device based on structured prompts provided in the second aspect are described in the relevant description of the business opportunity mining method based on structured prompts provided in the first aspect. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the business opportunity mining method based on structured prompts according to an embodiment of this application. Figure 2 This is a schematic diagram of the system composition of the business opportunity mining method based on structured prompts according to an embodiment of this application; Figure 3 This is a schematic diagram of the business opportunity mining device based on structured prompts according to an embodiment of this application.

[0024] Explanation of reference numerals in the attached figures: 1. A business opportunity discovery device based on structured prompts; 2. Processor; 3. Memory. Detailed Implementation

[0025] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application can be understood more clearly and thoroughly, and that the scope of this application can be fully conveyed to those skilled in the art.

[0026] This application applies to scenarios where enterprises intelligently mine potential business opportunities from massive, heterogeneous public bidding and procurement information. Existing technologies suffer from problems such as uncontrollable large-scale model parsing and rigid interaction of traditional models.

[0027] In various embodiments of this application, a structured business query library is constructed; a user's business opportunity mining query is received; using a large language model and based on a predefined structured prompt template, the business opportunity mining query is processed to obtain an intent recognition result; and a business opportunity query statement combining business rules and dictionary constraints is generated based on the intent recognition result; initial query results are obtained from the business query library based on the business opportunity query statement; based on the intent recognition result, standardized business opportunity analysis is performed on the initial query results; and the final business opportunity mining result is output according to a predefined format specification. This achieves end-to-end controllability and reliability of the business opportunity mining process from natural language interaction to structured result output, making business opportunity mining more accurate, controllable, flexible, and efficient.

[0028] The present application will now be described in conjunction with the accompanying drawings and specific embodiments.

[0029] Please refer to Figure 1 Opportunity discovery methods based on structured prompts include: Step S1: Construct a structured business query library based on historical business data.

[0030] In this embodiment, a structured commercial query library needs to be pre-built to serve as a unified and standardized underlying data support. Specifically, in this embodiment, step S1 includes: Step S11: Based on the pre-configured industry keyword library and negative keyword library, collect historical business data from multiple public data sources.

[0031] In this embodiment, the publicly available data sources include public websites such as the China Government Procurement Network, whose historical business data includes bidding announcements and winning bid announcements, making the data sources more authentic and reliable.

[0032] Step S12: Preprocess historical business data.

[0033] The preprocessing process includes extracting explicit fields such as project code, project name, procuring entity, announcement time, bid opening time, budget amount, and project location from historical business data, and performing duplicate removal, missing data repair, and format standardization.

[0034] Step S13: Based on the preset industry classification system and administrative division knowledge base, identify and classify the industry attributes and regional attributes of the preprocessed historical business data.

[0035] The industry attribute identification and classification adopts a four-level classification system consisting of nine vertical industries, atomic industries, subcategories of atomic industries, and an industry tree. The unique industry affiliation is determined by matching the name of the purchasing unit with industry keywords.

[0036] Among them, the identification and classification of regional attributes are determined by matching with the national administrative division knowledge base to determine the information of provinces, cities, and counties.

[0037] Step S14: Calculate the feature indicators of the identified and categorized historical business data in the four dimensions of time, space, technology, and scenario.

[0038] Among them, the characteristic indicators under the four dimensions of time, space, technology, and scenario include, but are not limited to, the growth rate of the number of projects, the regional coverage index, the technical keyword TF-IDF, and the scenario penetration rate, etc.

[0039] Step S15: Persistently store the structured data containing feature indicators to form a commercial query library that serves as the sole data source for subsequent queries and analysis.

[0040] The structured data includes, but is not limited to, industry bidding data tables, industry trend analysis result tables, and bidding scenario detail tables. These result tables are provided to the large model system in the form of an SQL database or similar structured storage, serving as the only trusted data source and not directly exposing the original text.

[0041] Step S2: Receive the user's business opportunity mining query, use a large language model, based on a predefined structured prompt template, process the business opportunity mining query to obtain intent recognition results, and generate a business opportunity query statement that combines business rules and dictionary constraints based on the intent recognition results, and obtain the initial query results from the business query database based on the business opportunity query statement.

[0042] Specifically, when receiving a business opportunity mining query from a user described in natural language, this application does not directly feed the question to the large language model. Instead, it first encapsulates the user's question with a pre-defined structured prompt template and inputs it as a whole into the large language model. The large language model then processes the query under the constraints of the structured prompt template.

[0043] In the following embodiments of this application, the first example will be used for illustration. In the first example, the business opportunity mining query is: help me see if there are any cloud platform projects with a value of over one million in the education industry in Yunnan Province recently, and provide analysis and suggestions.

[0044] The intent recognition result identifies the business opportunities that the user wants to explore, thereby determining the direction of subsequent queries.

[0045] Among them, business rules and dictionary constraints are the backend rules corresponding to deterministic intents. In the first example, the backend rules for applying business opportunity queries include: bid opening time filtering, default geographical location rules, and output format specifications. Combined with the parsed user key entity information, a SQL query statement that fully conforms to the user query and the preset format and logic is generated. The result data is retrieved from the pre-built structured database as the initial query result.

[0046] Step S3: Based on the intent recognition results, perform standardized business opportunity analysis on the initial query results, and output the final business opportunity mining results according to the predefined format specifications.

[0047] The predefined structured suggestion templates also perform corresponding analyses on the initial query results based on the deterministic intent. In the first example, the opportunity analysis includes an overview of the opportunity information, value analysis of key opportunities, and suggested solutions and strategies.

[0048] The predefined formatting specifications include: field name formatting specifications, query field sorting rules, and query data formatting specifications. Specifically, the field name formatting specifications require that query outputs consistently use Chinese field names, prohibiting the display of English database field names, and that the meaning of each field and the correspondence between its Chinese and English names be confirmed before querying. The query field sorting rules require that all query results be arranged in descending order according to the preset fields. The query data formatting specifications require that when outputting budget amount fields, the unit, such as ten thousand yuan, must be added. This ensures that regardless of how the user's question is phrased, the final output delivered to the user is a consistent, standardized, and easily understandable professional report.

[0049] Therefore, this application controls the large language model through a predefined structured prompt template, ensuring that it strictly follows a controlled process of intent recognition, query construction, standard analysis, and standardized output during the opportunity mining process. The execution of this process is strictly limited to the scope of a pre-built structured business query library, thereby effectively overcoming the technical defects of direct use of the large language model to process business opportunity information, such as numerous illusions, unstable output, and uncontrollability. This achieves end-to-end controllability and reliability in the opportunity mining process, from natural language interaction to structured result output, making opportunity mining more accurate, controllable, flexible, and efficient.

[0050] In one embodiment, step S2 processes the opportunity mining query to obtain intent recognition results, and generates an opportunity query statement that combines business rules and dictionary constraints based on the intent recognition results, including: Step S21: Match the business opportunity mining query with predefined intent keywords and logical judgment rules to obtain the matched deterministic intent.

[0051] This application predefines categories of deterministic intents and corresponding intent keywords. The categories of deterministic intents include: Business opportunity search: Search for bidding projects that can be secured now or in the future; Case search: Search for historical winning bids or typical cases; Trend Analysis: Statistical analysis of trends in a specific industry / region / technology; Hybrid intent: When complex needs such as business opportunities and trends are included, it is handled in the form of main intent + secondary intent.

[0052] The intent keywords include: If a business opportunity search contains keywords such as "tender," "project," "business opportunity," or "online listing," it is considered a business opportunity search. If a business opportunity search contains keywords such as "case study", "typical project", or "success story", it is considered a case study search. If the business opportunity mining query contains keywords such as "trend", "development", "analysis", or "trend", it is determined to be a trend analysis. If multiple types of keywords exist simultaneously, select the main image according to priority or frequency of occurrence, and explicitly label the main image and secondary intent in the output mixed intent.

[0053] In the first example, the deterministic intent of the opportunity mining query is opportunity query.

[0054] Step S22: Parse the key entity information associated with the deterministic intent from the business opportunity mining query to obtain a structured description of the intent recognition result that includes the deterministic intent and key entity information.

[0055] Different deterministic intents correspond to different key entity information. In the first example, key entity information includes: time (last 3 months), industry (education), region (Yunnan), amount, and product information.

[0056] Specifically, once a deterministic intent and key entity information are identified, the intent recognition result is returned in a strictly fixed JSON / structured format.

[0057] In the first example, the intent recognition result is: { "intent_type":"Business Opportunity Inquiry", "sub_intent":["Trend Analysis"], "entities":{ "region":"Yunnan Province", "industry":"education industry", "budget_threshold":"over 1 million yuan", "time_range":"The past 3 months" } In this embodiment, if the model's confidence level is insufficient, such as when it cannot clearly determine whether a query is a case query or a business opportunity query, the structured prompt template requires the large language model to ask the user for clarification, rather than making inferences on its own. By writing intent recognition rules into the structured prompt and requiring output in a fixed machine-readable format, the programmatic verification and control of the large model's intent understanding results are achieved, thus eliminating ambiguity for subsequent queries.

[0058] Therefore, by matching user queries with predefined keywords and rules, and forcing them to be categorized into a limited number of deterministic intent categories, the system achieves standardized and deterministic understanding of users' fuzzy and diverse natural language queries, avoids deviations in the entire analysis chain caused by misjudgment of intent, and further improves the accuracy and controllability of business opportunity mining.

[0059] Step S23: According to the deterministic intention, call the mandatory business rules corresponding to the deterministic intention from the preset rule library.

[0060] In this embodiment, the recognized deterministic intention and key entity information are transformed into a strictly restricted structured query, and various business rules, dictionaries, and default conditions corresponding to the intention recognition result are forcibly introduced.

[0061] Specifically, the mandatory business rules corresponding to different deterministic intentions include: Business opportunity query: Specifically refers to future business opportunities (1) It is necessary to automatically add a screening condition that the bid opening time is later than the current time, and its structured description is: STR_TO_DATE(bid_opening_time,'%Y-%m-%d %H:%i')>NOW(); (2) It is necessary to filter out dirty data for the bid opening time field first, that is, records with non-standard date formats do not participate in the calculation.

[0062] Case query: (1) By default, sort by budget amount from high to low and limit the number of records, such as the first 20 records.

[0063] Trend analysis: (1) Default time window: the last 30 days or the last N months. If the user does not specify, a reasonable default will be given according to the rules in the structured prompt template.

[0064] Step S24: Perform field matching and data verification on the key entity information through a predefined field dictionary, and the field dictionary defines the queryable data fields and their legal corresponding data acquisition rules.

[0065] Among them, the above steps also include: When a preset keyword field is missing in the key entity information, set the missing keyword field according to the relevant information of the user; When the content in the key entity information cannot find a legal matching item in the predefined field dictionary, an unrecognizable field prompt will be output.

[0066] For example, the preset keyword field is region, such as Yunnan is specified in the first example. However, if the user does not specify the geographical location when querying, the system will default to inject Yunnan Province into the region field according to the user's historical query content and personal information.

[0067] For example, a predefined field dictionary includes a vertical industry name field, limited to ten options: government and political affairs, education, healthcare, agriculture and commerce, internet, transportation, finance, military-civilian integration, and industry and energy. The first example explicitly states the education industry. However, if the user's input cannot directly match any of these ten industry types, the structured suggestion template requires the large language model to first attempt a fuzzy match using the "keyword category / subcategory" field. If this also fails, the model must explain to the user that the industry cannot be identified, rather than subjectively categorizing it randomly. This ensures that all generated queries are within the valid data range, avoiding invalid queries and data inconsistencies.

[0068] Therefore, by invoking mandatory business rules based on deterministic intent and using field dictionaries to validate and constrain entity information, the correctness of the generated database query statements in terms of business logic and the security of data access scope are ensured. This prevents invalid, unauthorized, or non-business-logical queries from being generated by the large language model, thus guaranteeing the accuracy and compliance of the results from the source of data retrieval.

[0069] Step S25: Combine and concatenate the mandatory business rules, the key entity information after matching and verification, and the fixed query field template to generate a structured database query statement as a business opportunity query statement.

[0070] It includes a pre-defined set of fixed query field templates that combine the entire field to generate a structured database query statement as a business opportunity query statement.

[0071] In one embodiment, step S3 involves performing standardized opportunity analysis on the initial query results based on the intent recognition results, including: Step S31: Based on the deterministic intent, select the corresponding standardized analysis task template. The analysis output items defined in the standardized analysis task template shall include at least one or more of the following: project information overview list, key business opportunity value assessment, adaptive solution strategy suggestions, and industry trend insight report.

[0072] Step S32: Based on the initial query results, control the large language model to perform analysis according to the selected analysis task template.

[0073] Specifically, after the structured query results are returned, the large language model needs to perform role-based and hierarchical insight analysis on the results under cue engineering constraints. Specifically, under cue engineering constraints, the large language model must output corresponding information step by step, which can be tailored according to the type of deterministic intent. This embodiment is explained in detail below: 1) Project Information Overview The query results will be displayed in a table or list format, with fields including: project code, project name, procuring entity, announcement date, bid opening time, budget amount, project location, and vertical industry.

[0074] 2) Key Business Opportunity Analysis Filter the results and mark projects with budgets exceeding a certain threshold as key business opportunities; Analyze its potential value, such as: whether it belongs to an emerging technology direction; whether it is located in a key expansion area; whether it is a comprehensive project across product lines, etc.

[0075] 3) Suggested solutions Based on the industry, scenario, and company product portfolio, provide tailored suggestions for each key business opportunity, including: architectural solutions, key product portfolios, and customer communication or competitive strategies.

[0076] 4) Interpretation of Typical Cases Select several projects with relatively large amounts and recent dates from the results, and analyze them one by one; summarize their industry value, policy background and replicable experience.

[0077] 5) Industry Trend Insights Based on the trend analysis table of correlation, a comprehensive interpretation is made of the changing trends of investment scale, popular technology directions, and application scenario diffusion in the target industry / region.

[0078] Step S3 further includes: The structured prompt template sends constraint instructions to the large language model, which requires that it must and can only rely on the factual information contained in the initial query results, prohibiting the introduction of external knowledge or fictitious data. Furthermore, when the initial query results are insufficient to draw a definite conclusion, the large language model is required to explicitly indicate insufficient data in the output.

[0079] Specifically, the structured prompt template explicitly requires: 1) When performing any analysis, summary, or trend inference, the large language model must not fabricate non-existent items or fields; 2) Large language models can only make inferences based on initial query results and known indicators. If the data is insufficient to support a certain conclusion, it must be explicitly stated that "the data is insufficient to support a clear conclusion". 3) For speculative judgments, they should be expressed with words such as possible, tend to, or based on preliminary data. Inferences should not be treated as factual statements.

[0080] Therefore, this embodiment breaks down the analysis task into clear sub-tasks and provides strict descriptions in a structured prompt template. This makes the output of the large language model readable and insightful, while minimizing logical jumps and data illusions, and significantly improving the credibility and interpretability of the analysis conclusions.

[0081] From the perspective of system implementation, the structured suggestion-based opportunity mining methods described in the above embodiments refer to... Figure 2 It can be seen that the corresponding system can be divided into corresponding functional modules, specifically: The five modules of the data layer correspond to the five sub-steps S11 to S15 in step S1.

[0082] The two modules in the computation layer are used to execute steps S2 and S3 and cache the corresponding results.

[0083] In the large model layer, the intent recognition engine corresponds to the two sub-steps S21 and S22 in step S2, the query construction and constraint engine corresponds to the three sub-steps S23 to S25 in step S2, the insight analysis engine corresponds to the sub-step S31 in step S3, and the output standardization controller corresponds to the sub-step S32 in step S3. Thus, all four engines in the large model layer are controlled through a unified structured prompting engineering, achieving controllable orchestration of the entire natural language interaction process.

[0084] The interaction layer is used to interact with users, conduct corresponding session management, and record logs.

[0085] Please refer to Figure 3 The embodiments of this application also provide a business opportunity mining device 1 based on structured prompts, including a memory 3, a processor 2, and a computer program stored on the memory 3 and executable on the processor 2. When the processor 2 executes the computer program, it implements the steps of the business opportunity mining method based on structured prompts in the above embodiments.

[0086] As can be seen from the above embodiments, this application has the following advantages: 1. It achieves precise control over the application behavior of large language models in professional fields, fundamentally suppressing the illusion problem of large language models.

[0087] By designing and implementing a complete four-dimensional structured prompting engineering system encompassing intent recognition, query construction and constraint, insight analysis, and standardized output, this application strictly confines the large language model to a closed framework defined by business rules, dictionary constraints, and a structured knowledge base. This transforms the model from an open text generator into a query and analysis engine with controlled access to a business query library. The entire process of query generation, analysis execution, and output organization is subject to auditable constraints based on predefined rules, effectively preventing fictitious information, incorrect attribution, and professional biases caused by the model's unchecked actions, ensuring high factual consistency and reliability of the business opportunity mining results.

[0088] 2. An intelligent interaction paradigm that integrates flexible interaction with natural language and in-depth exploration of structured data has been constructed.

[0089] This application overcomes the limitations of traditional recommendation models, which can only provide static lists and cannot understand user intent. Users can freely ask questions, filter, and follow up using natural language, and the system transforms this flexible intent into precise queries and in-depth analysis of a structured knowledge base through structured prompting engineering. Ultimately, users do not receive a simple list, but a comprehensive report containing project overviews, key business opportunity value analysis, solution strategy recommendations, and industry trend insights, greatly improving information acquisition efficiency and decision support depth.

[0090] 3. It provides a set of reusable and engineerable large-scale model control protocols, realizing the feasibility and portability of intelligent applications.

[0091] This application concretizes the innovative control mechanism into a series of reusable and configurable structured prompt templates. This enables the entire system to have a clear technical implementation path, good maintainability, and adaptability to different industries or data sources, providing a solid engineering foundation for building reliable and trustworthy enterprise-level AI applications.

[0092] Since the systems / devices described in the above embodiments of this application are systems / devices used to implement the methods of the above embodiments of this application, those skilled in the art can understand the specific structure and modifications of the system / devices based on the methods described in the above embodiments of this application, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of this application fall within the scope of protection of this application.

[0093] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0095] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The words "a" or "an" preceding a component do not exclude the presence of a plurality of such components. This application can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0096] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0097] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0098] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if these modifications and variations fall within the scope of the claims of this application and their equivalents, then this application should also include these modifications and variations.

Claims

1. A business opportunity mining method based on structured prompts, characterized in that, include: Step S1: Construct a structured business query database based on historical business data; Step S2: Receive the user's business opportunity mining query, use a large language model, based on a predefined structured prompt template, process the business opportunity mining query to obtain intent recognition results, and generate a business opportunity query statement that combines business rules and dictionary constraints based on the intent recognition results, and obtain the initial query results from the business query database based on the business opportunity query statement. Step S3: Based on the intent recognition result, perform standardized business opportunity analysis on the initial query result, and output the final business opportunity mining result according to the predefined format specification.

2. The business opportunity mining method based on structured prompts according to claim 1, characterized in that, Step S2 involves processing the business opportunity mining query to obtain intent recognition results, including: The business opportunity mining query is matched with predefined intent keywords and logical judgment rules to obtain the matched deterministic intent; The key entity information associated with the deterministic intent is parsed from the business opportunity mining query to obtain a structured description of the intent recognition result, which includes the deterministic intent and key entity information.

3. The business opportunity mining method based on structured prompts according to claim 2, characterized in that, Step S2, which generates a business opportunity query statement combining business rules and dictionary constraints based on the intent recognition result, includes: Based on the deterministic intent, the mandatory business rule corresponding to the deterministic intent is called from the preset rule base; The key entity information is matched and validated using a predefined field dictionary, which defines the queryable data fields and their corresponding legal data acquisition rules. The mandatory business rules, the key entity information after matching and verification, and the fixed query field templates are combined and concatenated to generate a structured database query statement as a business opportunity query statement.

4. The business opportunity mining method based on structured prompts according to claim 1, characterized in that, Step S3, based on the intent recognition result, performs standardized business opportunity analysis on the initial query result, including: Based on the intent recognition results, select the corresponding standardized analysis task template. The analysis output items defined by the standardized analysis task template include at least one or more of the following: project information overview list, key business opportunity value assessment, adaptive solution strategy suggestions, and industry trend insight report. Based on the initial query results, the large language model is controlled to perform analysis according to the selected analysis task template.

5. The business opportunity mining method based on structured prompts according to claim 4, characterized in that, Step S3, which involves performing standardized business opportunity analysis on the initial query results based on the intent recognition results, further includes: The structured prompt template sends constraint instructions to the large language model, which require that it must and can only be based on the factual information contained in the initial query results, prohibiting the introduction of external knowledge or fictitious data, and requiring the large language model to clearly indicate insufficient data in the output when the initial query results are insufficient to draw a definite conclusion.

6. The business opportunity mining method based on structured prompts according to claim 1, characterized in that, Step S2 also includes: When the key entity information is missing a preset key field, the missing key field is set according to the user's relevant information; When the content of the key entity information cannot find a valid match in the predefined field dictionary, an output field recognition error message will be displayed.

7. The business opportunity mining method based on structured prompts according to any one of claims 1 to 6, characterized in that, Step S1 includes: Based on a pre-configured industry keyword library and negative keyword library, historical business data is collected from multiple public data sources; The historical business data is preprocessed; Based on a pre-defined industry classification system and administrative division knowledge base, the pre-processed historical business data is identified and categorized according to industry and regional attributes. Calculate and identify the characteristic indicators of historical business data in four dimensions: time, space, technology, and scenario. The structured data containing the aforementioned feature indicators is persistently stored to form a commercial query library that serves as the sole data source for subsequent queries and analysis.

8. The business opportunity mining method based on structured prompts according to any one of claims 1 to 6, characterized in that, The predefined format specifications in step S3 include: field name format specifications, query field sorting rules, and query data format specifications.

9. The business opportunity mining method based on structured prompts according to claim 2 or 3, characterized in that, The deterministic intent includes at least one of business opportunity search, case search, and trend analysis.

10. A business opportunity mining device based on structured prompts, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the business opportunity mining method based on structured prompts as described in any one of claims 1 to 9.

Images