An intelligent communication assistance method and system for medical academic popularization

Abstract

The application relates to the technical field of medical academic digital popularization, and specifically discloses an intelligent communication auxiliary method and system for medical academic popularization, which comprises the following steps: acquiring multi-source communication data and analyzing and extracting business characteristics, constructing a cooperation partner portrait, converting the portrait and business project demand into a semantic vector, calculating the similarity to screen a candidate cooperation partner set, calculating the score through a comprehensive scoring function, generating a cooperation partner recommendation list, real-time identification of compliance risks in business communication and pushing prompt information, automatic generation of targeted popularization language, collection of business feedback data for dynamic adjustment of the weight coefficient of the scoring function. Through the closed-loop mechanism of multi-source data fusion, semantic matching, comprehensive scoring, compliance control and dynamic optimization, the application solves the problems of low screening efficiency, insufficient matching accuracy, high labor cost and high compliance risk of traditional methods, improves the business efficiency and communication quality of medical academic popularization, reduces the compliance risk, and realizes the intelligentization of the whole popularization process.

Description

Technical Field

[0001] This invention relates to the field of digital promotion technology for medical academic research, and in particular to an intelligent communication assistance method and system for promoting medical academic research. Background Technology

[0002] As the scale of pharmaceutical academic promotion business continues to expand, the amount of partner master data accumulated by the company has increased significantly, and the size of the business development team has also grown accordingly. In the early stages, the company relied on traditional business development management systems to carry out its business. These systems stored partner information in structured data, and business development personnel would screen candidate partners through keyword searches, followed by manual communication and recording of communication content. Finally, the head of the business development department would coordinate resource allocation based on manually compiled information.

[0003] However, with business development, traditional systems also have many technical problems: First, partner screening relies on manual experience and simple conditional searches, lacking in-depth data analysis capabilities, resulting in low screening efficiency, insufficient matching accuracy, and difficulty in accurately identifying suitable partners. Second, operations such as communication record entry and cooperation intention maintenance rely on manual completion, with low automation. Sales personnel need to invest a lot of time, and the system lacks communication support mechanisms, making it difficult for sales personnel to obtain effective support in a timely manner, resulting in low communication efficiency. Third, the compliance requirements for pharmaceutical academic promotion are becoming increasingly strict, and traditional systems lack standardized communication prompts, which can easily lead to non-standard information expression and potential compliance risks.

[0004] Therefore, there is an urgent need for an intelligent communication assistance method and system for promoting medical and pharmaceutical academic research in order to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent communication assistance method for medical academic promotion, comprising the following steps: Acquire multi-source communication data, parse the multi-source communication data, extract business characteristic information of partners, and construct partner profiles; Based on the partner profiles and investment project requirements, corresponding semantic vectors are generated, and vector similarity is calculated to filter out a set of candidate partners. The multi-dimensional business characteristics of each candidate partner in the candidate partner set are obtained, the comprehensive matching score of each candidate partner is calculated based on the preset comprehensive scoring function, and a partner recommendation list is generated based on the scoring results. The system acquires real-time communication content between investment promotion personnel and target partners, identifies compliance risks based on a pre-set compliance rule base and large language model, generates compliance prompts, and automatically generates promotional script suggestions based on the profile of the target partner and the needs of the investment promotion project. Collect business feedback data and dynamically adjust the weight coefficients in the comprehensive scoring function based on the business feedback data.

[0006] Furthermore, the present invention also discloses an intelligent communication assistance system for medical academic promotion, comprising: The acquisition module is used to acquire multi-source communication data, parse the multi-source communication data, extract the business characteristic information of partners, and construct partner profiles; The calculation module is used to generate corresponding semantic vectors based on the partner profile and investment project requirements, and calculate the vector similarity to filter out a set of candidate partners. The list generation module is used to obtain the multi-dimensional business characteristics of each candidate partner in the candidate partner set, calculate the comprehensive matching score of each candidate partner based on the preset comprehensive scoring function, and generate a partner recommendation list based on the scoring results. The suggestion generation module is used to obtain the real-time communication content between investment promotion personnel and target partners, identify compliance risks based on a preset compliance rule library and large language model, generate compliance prompt information, and automatically generate promotional script suggestions based on the profile of the target partner and the needs of the investment promotion project. The dynamic adjustment module is used to collect business feedback data and dynamically adjust the weight coefficients in the comprehensive scoring function based on the business feedback data.

[0007] Furthermore, the computing module includes: The acquisition unit is used to acquire multiple business feature information from the partner profile and convert the multiple business feature information into a corresponding first semantic vector through an embedding model; The conversion unit is used to obtain project information from the investment project requirements and convert it into a corresponding second semantic vector through an embedding model. A calculation unit is used to calculate the semantic similarity between the first semantic vector and the second semantic vector; The first filtering unit is used to select an initial candidate set based on the semantic similarity. The second screening unit is used to perform a secondary screening of the initial candidate set based on a preset business matching factor to obtain the final candidate partner set.

[0008] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described intelligent communication assistance method for medical academic promotion.

[0009] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described intelligent communication assistance method for promoting medical academic research.

[0010] The beneficial effects of this application are as follows: Firstly, this invention can improve the efficiency and accuracy of partner screening and matching. By collecting multi-source communication data to build a complete partner profile, and combining semantic vector matching and a multi-dimensional comprehensive scoring mechanism, it can achieve deep matching and screening of partners and investment projects, replacing traditional manual retrieval and subjective judgment, greatly reducing ineffective communication, and thus improving the accuracy of investment promotion.

[0011] Secondly, this invention can reduce manual operation costs and realize automatic parsing of multi-source data, automatic extraction of business features, and automatic generation of recommendation lists, reducing manual operations in communication record entry, partner screening, and other links, thereby reducing the time and manpower costs of investment promotion personnel. At the same time, it provides communication support through intelligent script generation, improving communication efficiency.

[0012] Third, this invention constructs a real-time compliance detection mechanism based on a compliance rule base and a large language model, accurately identifies the risk of violations in communication and pushes reminder information, standardizes the expression in promotional communication, and effectively reduces compliance risks in the process of medical academic promotion.

[0013] Fourth, this invention can clarify communication priorities through automated recommendation and scoring ranking, provide data support for resource allocation, reduce manual statistical analysis, and realize intelligent management and optimized allocation of cooperative resources. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of a method flow proposed in an embodiment of this application.

[0015] Figure 2 This is a schematic diagram of the system structure proposed in an embodiment of the present invention.

[0016] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] like Figure 1 As shown, this application provides an intelligent communication assistance method for medical academic promotion, including the following steps: S1. Acquire multi-source communication data, parse the multi-source communication data, extract business characteristic information of partners, and construct partner profiles; S2, Based on the partner profile and investment project requirements, generate corresponding semantic vectors and calculate vector similarity to select a set of candidate partners; S3, obtain the multi-dimensional business characteristics of each candidate partner in the candidate partner set, calculate the comprehensive matching score of each candidate partner based on the preset comprehensive scoring function, and generate a partner recommendation list based on the scoring results; S4. Obtain the real-time communication content between the investment promotion personnel and the target partners, identify compliance risks based on the preset compliance rule library and large language model, generate compliance prompt information, and automatically generate promotional script suggestions based on the profile of the target partners and the needs of the investment promotion project. S5, collect business feedback data, and dynamically adjust the weight coefficients in the comprehensive scoring function based on the business feedback data.

[0019] As described in steps S1-S5 above, this invention constructs a full-process intelligent communication assistance system that integrates data collection and analysis, semantic matching and filtering, comprehensive scoring and recommendation, real-time compliance management and control, and dynamic optimization and iteration. This system enables precise screening of partners, efficient communication and compliance risk prevention during the promotion of pharmaceutical academics, forming a continuously self-optimizing intelligent recommendation closed loop. It comprehensively solves the core problems of low partner screening efficiency, insufficient matching accuracy, high manual operation costs, weak communication assistance capabilities and insufficient compliance risk prevention in existing technologies.

[0020] As the scale of pharmaceutical academic promotion business expands, the amount of partner data surges. Existing technologies lack intelligent support for the entire process, and the various links are isolated from each other. Data processing, matching and screening, communication support and model optimization cannot form an effective linkage, resulting in low investment promotion efficiency, serious waste of resources and prominent compliance risks. Therefore, it is necessary to establish a complete technical system covering data processing and model optimization, and realize the intelligent upgrade of the entire investment promotion process through the collaborative operation of each link.

[0021] Traditional solutions employ a decentralized, manual operation and simplistic system-assisted approach. Data collection and analysis rely on manual input, partner selection depends on keyword searches and subjective judgment, communication lacks compliance prompts and precise wording support, and there is no continuous optimization mechanism. This results in low efficiency and poor coordination across all stages, making it unsuitable for large-scale, high-precision business development needs. This invention integrates six core elements: multi-source data processing, semantic vector matching, multi-dimensional comprehensive scoring, real-time compliance detection, intelligent wording generation, and dynamic model optimization. This forms a closed-loop technical solution that specifically addresses the problems of fragmented processes, low intelligence, and lack of collaborative optimization in traditional methods.

[0022] The core principle of this invention is as follows: First, by collecting and analyzing multi-source communication data, key business features are extracted to construct a comprehensive partner profile, providing a data foundation for subsequent matching. Next, the partner profile and investment project requirements are converted into semantic feature vectors. Core candidates are selected through similarity calculation. Then, a comprehensive quantitative scoring is performed based on multi-dimensional business features to generate an accurate partner recommendation list. During the communication process between investment personnel and partners, compliance risks are detected in real time and prompts are pushed. At the same time, personalized promotional messages are generated. Finally, by collecting business feedback data, the weight coefficients of the scoring model are continuously adjusted to achieve dynamic optimization of the recommendation model, forming a complete intelligent closed loop of data input, processing, output, feedback, and optimization.

[0023] The core objective of this invention is to achieve intelligent and precise operation of the entire process of pharmaceutical academic promotion and investment attraction. By using multi-source data fusion and semantic matching technology to improve the accuracy of partner matching, by using a comprehensive scoring mechanism to clarify communication priorities, by using real-time compliance management to reduce the risk of violations, and by using a dynamic optimization mechanism to maintain the long-term adaptability of the model, the invention ultimately significantly reduces the cost of manual operation, improves investment attraction efficiency and communication quality, and ensures that the promotion process complies with industry compliance requirements, thus providing comprehensive intelligent support for pharmaceutical academic promotion.

[0024] In one embodiment, step S1, which involves acquiring multi-source communication data, parsing the multi-source communication data, extracting business characteristic information of partners, and constructing a partner profile, specifically includes: S11, acquire multi-source communication data between investment promotion personnel and potential partners through the data acquisition module. The multi-source communication data includes telephone communication recording data, enterprise WeChat chat text data, and image and document data transmitted during the communication process. S12, Automatic speech recognition processing is performed on the telephone communication recording data based on the speech recognition model to obtain the first text data; The image and document data are parsed and processed based on the multimodal understanding big data model to obtain the second text data; S13, the first text data, the second text data and the enterprise WeChat chat text data are merged to construct a complete set of business communication texts; S14, based on the large language model, perform semantic analysis and information extraction on the business communication text set, and extract the business characteristic information of the potential partners, including cooperation attitude, expertise in relevant departments, hospital resources and potential cooperation intentions; S15, the extracted business feature information is structured and associated with the basic information of the potential partners for storage, thereby constructing a partner profile containing multiple dimensions.

[0025] As described in steps S11-S15 above, comprehensive collection, accurate analysis, effective integration, and structured extraction of business characteristics of multi-source communication data can be achieved. Ultimately, a multi-dimensional partner profile can be constructed, providing a data foundation for subsequent intelligent matching of partners and investment projects, generation of promotional scripts, and compliance risk assessment.

[0026] Because existing technologies typically store partner information only in structured data form, there is a lack of in-depth mining and integration of unstructured data generated during communication between investment promotion personnel and potential partners. This results in partner profiles being one-dimensional and incomplete, failing to fully reflect the business capabilities and cooperation intentions of partners, and thus affecting the matching accuracy of partner screening. Therefore, it is necessary to comprehensively acquire multi-source communication data and process it systematically to extract key business features and solve the problem of insufficient partner profile construction.

[0027] Traditional solutions rely solely on manual input of basic structured information from partners, failing to handle unstructured data such as recorded phone conversations, images, and documents. Furthermore, the limited data collection dimensions result in incomplete business feature extraction, making it impossible for user profiles to support precise matching needs. This solution addresses the shortcomings of traditional methods—single data source, insufficient unstructured data processing capabilities, and incomplete feature extraction—through a coherent technical approach encompassing multi-source data collection, categorized parsing and transformation, data fusion, semantic analysis and information extraction, and structured storage.

[0028] Specifically, the process begins with S11, which uses a data acquisition module to obtain multi-source communication data between sales personnel and potential partners. This data originates from actual business communication between the two parties. The telephone communication recordings are voice records generated during telephone conversations between sales personnel and potential partners. The WeChat chat text data are text information generated through online communication between the two parties via WeChat. The image and document data are multimedia materials related to business transmitted during the communication process. The collection of multi-source data ensures comprehensive information coverage and provides data support for the subsequent extraction of complete business characteristics.

[0029] Next, S12 processing is performed. For telephone communication recording data, automatic speech recognition processing is performed based on a speech recognition model. For example, this speech recognition model can use the Doubao large model recording file recognition standard version interface. At the same time, a pre-built medical business knowledge base and domain professional corpus are introduced as prior knowledge. Through domain adaptive optimization, the model's semantic understanding ability of medical professional terms and business context is improved, thereby accurately converting the speech data into first text data and ensuring the effective extraction of speech information. For image and document data, based on the Doubao Volcano Ark multimodal understanding model, combined with prompt word templates built for medical academic promotion scenarios, the model is semantically guided through business knowledge, enhancing the model's ability to extract key business information from multimodal data, converting unstructured image and document data into standardized second text data, and realizing unified parsing of different types of unstructured data.

[0030] Then, S13 is executed to uniformly format the first text data, the second text data, and the WeChat chat text data to construct a complete set of business communication texts. This process eliminates the format differences of data from different sources through data format standardization rules, realizes data fusion and unification, and ensures that subsequent semantic analysis can be carried out based on complete communication information, avoiding feature extraction omissions caused by data dispersion.

[0031] Then, in S14, semantic analysis and information extraction are performed on the business communication text set based on a large language model. The large language model can deeply understand the business logic and potential information in the text and accurately extract the business characteristic information of potential partners. These business characteristic information include cooperation attitude, expertise in specific departments, hospital resources, and potential cooperation intentions. Cooperation attitude is determined by the model's recognition of the semantic tendency of the text. Expertise in specific departments is obtained through keyword matching and semantic classification. Hospital resources are quantitatively extracted based on information such as hospital level and cooperation coverage mentioned in the text. Potential cooperation intentions are obtained by comparing semantic similarity analysis with a preset cooperation intention expression template. Comprehensive business characteristic extraction provides core support for building accurate partner profiles.

[0032] Finally, step S15 is executed to structure the extracted business feature information, organize the feature content according to the preset data structure specifications, and associate and store the structured business feature information with the basic information of potential partners. The basic information of potential partners comes from the existing partner master data in the business system database. After association and storage, a partner profile containing multiple dimensions is constructed. This profile can comprehensively and systematically reflect the business attributes and cooperation potential of partners, providing a reliable data foundation for semantic vector construction, similarity calculation and accurate recommendation in subsequent steps. This directly supports the present invention in solving the core problems of low efficiency and insufficient matching accuracy in partner screening.

[0033] In one embodiment, step S2, which involves generating corresponding semantic vectors based on the partner profile and investment project requirements, and calculating vector similarity to filter out a candidate partner set, specifically includes: S21, obtain multiple business feature information from the partner profile, and convert the multiple business feature information into a corresponding first semantic feature vector through an embedding model. Each component in the first semantic feature vector is used to represent the feature value of the partner in multiple dimensions such as academic cooperation activity, specialty department characteristics and hospital resource level. S22, obtain information on the indications, key promotion departments and target hospital types of pharmaceutical products in the requirements of the investment promotion project, and convert them into corresponding second semantic feature vectors through an embedding model; S23, Based on the vector database, the semantic similarity between the first semantic feature vector and the second semantic feature vector is calculated using the cosine similarity algorithm. The specific calculation process is as follows: First, the dot product of the first semantic feature vector and the second semantic feature vector is calculated. Then, the magnitude of the first semantic feature vector and the magnitude of the second semantic feature vector are calculated respectively. Finally, the dot product is divided by the product of the magnitude of the first semantic feature vector and the magnitude of the second semantic feature vector to obtain the semantic similarity. S24, Select partners whose semantic similarity is higher than a preset threshold or whose ranking is in the top N as the initial candidate set; S25, based on the initial candidate set, department matching degree and hospital resource matching degree are introduced as screening factors to perform a second screening on the initial candidate set to obtain the final candidate partner set.

[0034] As described in steps S21-S25 above, it is possible to achieve accurate semantic matching between partner profiles and investment project requirements. Through semantic vector conversion, similarity calculation and secondary screening, a set of candidate partners that are highly compatible with investment projects can be efficiently selected, providing high-quality candidates for subsequent comprehensive scoring and recommendation. This directly solves the core problems of low partner screening efficiency and insufficient matching accuracy in existing technologies.

[0035] Because existing technologies rely on human experience and simple conditional searches for partner selection, they can only match superficial keywords and cannot deeply explore the semantic relationship between partners and investment projects. This results in low matching between the selection results and actual needs, and a large amount of ineffective communication increases investment costs. Therefore, it is necessary to establish a semantic matching mechanism to achieve accurate selection by quantifying similarity, thereby improving selection efficiency and matching accuracy.

[0036] Traditional methods for screening partners rely on keyword searches combined with manual judgment. This approach fails to capture the deep semantic connections between business characteristics and project requirements, and its reliance on singular screening criteria makes it difficult to comprehensively consider multi-dimensional matching factors, resulting in poor accuracy and low efficiency. By converting partner profiles and project requirements into semantic feature vectors, and using a cosine similarity algorithm to quantify the degree of semantic matching, a secondary optimization is performed using business-dimensional screening factors. This creates a two-layer screening mechanism combining initial semantic matching and refined business feature screening, specifically addressing the shortcomings of traditional methods in capturing insufficient semantic connections and relying on a single screening dimension.

[0037] Specifically, step S21 is executed first to obtain multiple business feature information from the partner profile. This business feature information comes from the partner profile and covers multiple dimensions such as academic cooperation activity, specialty department characteristics, hospital resource level, historical cooperation success rate, and communication frequency. These business feature information are then converted into corresponding first semantic feature vectors using an embedding model. Numerical features such as historical cooperation success rate and communication frequency are directly used as vector components after normalization. Categorical features such as specialty department characteristics are converted into vector representations using the embedding model. Each component of the final first semantic feature vector corresponds to the partner's feature value across various business dimensions, achieving a unified semantic expression for different types of business features and laying the foundation for subsequent cross-dimensional matching.

[0038] Next, step S22 is performed to obtain information on the indications for pharmaceutical products, key departments for promotion, and target hospital types from the investment project requirements. This information comes from the investment project knowledge base, which is constructed by the system after structuring relevant information. This project requirement information is then converted into corresponding second semantic feature vectors using an embedding model. The conversion process is consistent with that of the first semantic feature vectors, ensuring that both exist in the same vector space and satisfying the prerequisites for similarity calculation. This achieves a semantically quantified expression of the investment project requirements.

[0039] Then, step S23 is executed, which calculates the semantic similarity between the first and second semantic feature vectors based on the vector database. The cosine similarity algorithm is used, specifically: first, the dot product of the first and second semantic feature vectors is calculated; then, the magnitudes of the first and second semantic feature vectors are calculated separately; finally, the dot product is divided by the product of the magnitudes of the two vectors to obtain the semantic similarity. The calculation formula is as follows: sim(x,y)=(x·y) / (||x||·||y||); Where x represents the first semantic feature vector, y represents the second semantic feature vector, and sim(x,y) represents the semantic similarity. This algorithm can accurately quantify the angle relationship between two vectors in the semantic space. The closer the similarity value is to 1, the higher the semantic fit between the partner and the investment project. Through the efficient retrieval capability of the vector database, it can achieve rapid semantic matching of massive partners, greatly improving the screening efficiency.

[0040] Then, in step S24, an initial candidate set is selected based on the semantic similarity calculation results. Two screening methods can be used: first, selecting partners with semantic similarity higher than a preset threshold (set based on historical cooperation data and business experience) to ensure the selected partners have a basic semantic fit; second, selecting the top N partners in semantic similarity (N determined based on the actual needs of the investment promotion business) to ensure the initial candidate set is of appropriate size. Both methods can quickly screen partners with strong semantic relevance to the investment promotion project, completing the initial screening and reducing the amount of data processed subsequently.

[0041] Finally, step S25 is executed, introducing department matching degree and hospital resource matching degree as screening factors for secondary screening based on the initial candidate set. Department matching degree is determined by comparing the degree of fit between the partner's specialty departments and the key departments promoted in the investment project. For example, if the key department promoted in the investment project is cardiology, and the partner's specialty departments are cardiology and respiratory medicine, then the department matching degree is calculated according to preset rules. Hospital resource matching degree is determined by comparing the degree of fit between the partner's hospital resource level and the target hospital type of the investment project. By setting reasonable screening thresholds, partners whose department matching degree or hospital resource matching degree does not meet the standards are eliminated. The final candidate partner set has both high semantic fit and meets the matching requirements of the core business dimensions, providing higher-quality candidates for subsequent comprehensive scoring.

[0042] In one embodiment, step S3, which involves obtaining the multi-dimensional business characteristics of each candidate partner in the candidate partner set, calculating the comprehensive matching score of each candidate partner based on a preset comprehensive scoring function, and generating a partner recommendation list based on the scoring results, specifically includes: S31, obtain the multi-dimensional business characteristics of each candidate partner in the candidate partner set, including department matching degree, hospital resource matching degree, historical cooperation success rate and communication activity; S32, calculate the comprehensive matching score of each candidate partner based on the preset comprehensive scoring function. The specific calculation process is as follows: configure a preset weight coefficient for the department matching degree, hospital resource matching degree, historical cooperation success rate and communication activity, then multiply the value of each feature index with its corresponding weight coefficient, and finally add all the product results to obtain the comprehensive matching score. S33, obtain the comprehensive matching score of all partners in the candidate partner set, find the minimum score value and the maximum score value, and then for each partner, divide the difference between the comprehensive matching score of the partner and the minimum score value by the difference between the maximum score value and the minimum score value to obtain the normalized comprehensive matching score; S34, Sort the partners in the candidate partner set in descending order according to the normalized comprehensive matching score, and select a preset number of the top-ranked partners to generate a partner recommendation list; S35, the recommended partner list is displayed to the business development personnel. The recommended partner list includes the partner name, the normalized comprehensive matching score, and a summary of key business characteristics related to the reasons for the recommendation.

[0043] As described in steps S31-S35 above, the objects in the candidate partner set can be quantitatively evaluated and accurately ranked according to multi-dimensional business characteristics, generating a partner recommendation list containing core information. This provides clear guidance on communication priorities for investment promotion personnel, further improving the accuracy of partner selection and investment promotion efficiency, and solving the problems of lack of quantitative scoring mechanism and lack of targeted recommendation results in existing technologies.

[0044] Since the set of candidate partners after semantic matching and secondary screening still needs to be further differentiated in terms of suitability, the existing technology lacks a comprehensive quantitative evaluation of multi-dimensional business characteristics. Relying solely on a single dimension or manual judgment and ranking makes it difficult for investment promotion personnel to quickly identify the best partners, and communication priorities are unclear, affecting the efficiency of investment promotion. Therefore, it is necessary to establish a scientific comprehensive scoring mechanism to achieve accurate ranking of candidate partners through multi-dimensional feature weighted calculation and normalization processing.

[0045] Traditional solutions rank candidates based on single features or subjective experience, failing to comprehensively consider multi-dimensional core business characteristics such as departmental matching and historical success rates. This results in rankings lacking objectivity and specificity, hindering efficient decision-making by investment promotion personnel. By extracting key business characteristics of candidate partners and constructing a comprehensive scoring function with pre-defined weights, this solution normalizes the data and sorts it in descending order, generating a recommendation list containing core information. This comprehensive approach addresses the shortcomings of traditional methods, including single-dimensional scoring, insufficient objectivity, and incomplete recommendation information.

[0046] The specific approach involves first executing S31 to obtain multi-dimensional business characteristics of each candidate partner in the candidate partner set. These business characteristics are derived from the partner profiles and investment project demand matching results already constructed by the system. Among them, the department matching degree is obtained by the degree of fit between the partner's expertise and the key departments promoted in the investment project; the hospital resource matching degree is obtained by the degree of compatibility between the partner's hospital resource level and the target hospital type of the investment project; the historical cooperation success rate comes from the statistical data of the partner's past cooperation completion status stored in the business system database; and the communication activity level is obtained based on the statistical data of the historical communication frequency between investment personnel and partners. The selection of multi-dimensional business characteristics comprehensively covers the core dimensions of cooperation compatibility, providing comprehensive data support for the overall scoring.

[0047] Next, step S32 is performed, calculating the comprehensive matching score for each candidate partner based on a preset comprehensive scoring function. The system assigns preset weight coefficients to department matching degree, hospital resource matching degree, historical cooperation success rate, and communication activity level. These weight coefficients can be preset based on business experience or optimized and adjusted through model training based on historical cooperation data. The specific calculation process involves multiplying the specific value of each feature indicator by its corresponding weight coefficient, and then summing all the products to obtain the comprehensive matching score. For example, if the weight coefficients for department matching degree, hospital resource matching degree, historical cooperation success rate, and communication activity level are 0.3, and a partner's department matching degree is 0.9, hospital resource matching degree is 0.8, historical cooperation success rate is 0.7, and communication activity level is 0.6, then the partner's comprehensive matching score is 0.9×0.3+0.8×0.2+0.7×0.3+0.6×0.2. This weighted calculation achieves a comprehensive quantification of multi-dimensional features, objectively reflecting the compatibility between the partner and the investment project.

[0048] Then, step S33 is executed to normalize the overall matching score. First, the overall matching scores of all partners in the candidate partner set are obtained, and the minimum and maximum scores are selected. Then, for each partner, the difference between that partner's overall matching score and the minimum score is calculated. This difference is divided by the difference between the maximum and minimum scores to obtain the normalized overall matching score. The calculation formula is as follows: Score_norm=(Score-Score_min) / (Score_max-Score_min); Here, Score_norm represents the normalized overall matching score, Score represents the overall matching score of the partner, Score_min represents the minimum score value in the candidate set, and Score_max represents the maximum score value in the candidate set. Normalization maps all partner scores to the range of 0 to 1, eliminating the impact of differences in feature dimensions on the ranking results, making the scores more comparable, and providing a unified standard for subsequent ranking.

[0049] Then, in step S34, the partners in the candidate partner set are sorted in descending order based on the normalized comprehensive matching score, with partners whose scores are closer to 1 ranking higher. A preset number of top-ranked partners are then selected based on the actual needs of the business development operations to generate a partner recommendation list. This preset number can be flexibly set according to the daily business development communication plan or business needs. The sorting ensures that the most suitable partners are recommended first, helping business development personnel quickly focus on key communication targets and reducing ineffective screening time.

[0050] Finally, step S35 is executed, displaying the partner recommendation list to the investment promotion personnel. The list includes the partner's name, a normalized overall matching score, and a summary of key business characteristics related to the reasons for the recommendation. This summary focuses on core strengths such as departmental compatibility and historical success rates, clearly presenting the basis for the recommendation. This presentation method provides investment promotion personnel with comprehensive and accurate decision-making references, enabling them to quickly grasp the core suitability highlights of partners before communication, thus improving the targeting and efficiency of communication.

[0051] In one embodiment, step S4, which involves acquiring real-time communication content between investment promotion personnel and target partners, identifying compliance risks based on a preset compliance rule base and a large language model, generating compliance alerts, and automatically generating promotional script suggestions based on the target partner's profile and the investment project's needs, specifically includes: S41: Real-time acquisition of voice or text content generated during communication between investment promotion personnel and target partners, and conversion of it into text data to be analyzed; S42, Construct a compliance rule base, which includes violation expression patterns, sensitive terms and risk scenarios extracted from historical communication cases and pharmaceutical industry regulations; S43, the compliance rule base is input into the large language model in the form of prompt words, the large language model is used to perform semantic analysis on the text data to be analyzed, identify whether there is any violation risk, and generate a risk level assessment result, the risk level assessment result including high risk, medium risk and low risk; S44, When a violation risk is identified, a compliance alert message containing the risk type and modification suggestions is sent to the investment promotion personnel; S45, invoke the intelligent promotional script generation module to automatically generate targeted promotional communication scripts based on the profile information of the target partner and the academic characteristics of the current investment project. The promotional scripts include suggestions for academic promotion entry points, introduction of the core academic advantages of the product, and suggestions for cooperation entry strategies.

[0052] As described in steps S41-S45 above, real-time compliance risk management and targeted promotional script support can be achieved during the investment promotion communication process. This not only avoids the risk of violations in advance, but also provides precise communication guidance for investment promotion personnel, solving the core problems of insufficient communication compliance and inadequate communication support capabilities in existing technologies. At the same time, it improves the professionalism and communication efficiency of pharmaceutical academic promotion.

[0053] As compliance requirements in the pharmaceutical academic promotion industry become increasingly stringent, existing technologies lack mechanisms for identifying and alerting to compliance risks in real-time communication content. This can easily lead to non-standard information expression and compliance risks. Furthermore, sales personnel often struggle to obtain targeted communication strategies in a timely manner, impacting communication quality and efficiency. Therefore, it is necessary to establish a real-time compliance detection and intelligent script generation mechanism to simultaneously address the dual needs of compliance risk prevention and control and improved communication efficiency.

[0054] Traditional solutions rely on the expertise and experience of sales personnel to judge the compliance of communications, lacking standardized risk identification criteria and resulting in inconsistent effectiveness in preventing violations. Furthermore, communication scripts require advance preparation by sales personnel, making it impossible to dynamically adjust them based on partner characteristics and project needs, thus lacking specificity. This step, through a coherent mechanism of real-time communication content analysis, compliance rule base constraints, large language model risk identification, compliance prompts, and personalized script generation, forms a dual support system combining risk prevention and communication assistance. It specifically addresses the weaknesses of traditional methods in compliance risk prevention and the lack of dynamic adaptability in communication scripts.

[0055] First, S41 is executed to acquire real-time audio and text content generated during communication between sales personnel and target partners. Audio content is collected in real-time from recorded phone conversations, while text content is synchronized in real-time from WeChat chats. The audio content is converted into text data using a speech recognition model, and the text content is directly used as the text data to be analyzed. This ensures that the communication content can be accurately parsed by the system, providing real-time data input for subsequent compliance risk identification and script generation.

[0056] Next, S42 is implemented to build a compliance rule base. The rule base is based on actual communication cases accumulated by the company in the long-term process of pharmaceutical academic promotion, combined with the relevant regulatory requirements of the pharmaceutical industry. It is formed after being reviewed and organized by compliance experts and legal counsel. It includes illegal expression patterns, sensitive words, and typical risk scenarios. Among them, illegal expression patterns cover specific types such as non-compliant drug promotion statements and inappropriate efficacy promises. Sensitive words clearly define the statements that are prohibited or restricted by the industry. Typical risk scenarios sort out communication scenarios that are prone to compliance issues, providing a standardized basis for compliance risk identification.

[0057] Then, step S43 is executed, inputting the compliance rule base as prompt words into the large language model. The large language model then performs semantic analysis on the text data to be analyzed. Under the constraints and guidance of the compliance rule base, the large language model accurately identifies whether there are any violations in the text data to be analyzed, and determines the risk level based on the severity of the violation, the clarity of the violation content, and the frequency of its occurrence.

[0058] When the communication content contains explicit expressions of violation or highly sensitive and risky language, it is judged as high risk; When there are suggestive or non-standard expressions, it is judged as medium risk; When no obvious violations are detected but there is a potential risk, it is judged as low risk. The quantification of risk level provides a precise basis for subsequent compliance prompts.

[0059] Then, S44 is executed. When a violation risk is identified, the system immediately pushes a compliance alert to the investment promotion personnel. The alert includes the risk type and modification suggestions. The risk type clearly indicates the category of violation, and the modification suggestions provide examples of compliant expressions to help investment promotion personnel quickly adjust their communication and prevent the violation risk from escalating further, thus achieving real-time control of compliance risks.

[0060] Finally, step S45 is executed, invoking the intelligent promotional script generation module to obtain the target partner's profile information and the academic characteristics of the current investment project. The target partner's profile information includes core characteristics such as their areas of expertise and hospital resources, while the investment project's academic characteristics include information such as the pharmaceutical product's indications and core academic advantages. Based on this information, targeted promotional communication scripts are automatically generated, including suggestions for academic promotion entry points, introductions to the product's core academic advantages, and suggestions for cooperation entry strategies. The academic promotion entry point suggestions are determined by the alignment between the partner's areas of expertise and the product's indications; the introduction of the product's core academic advantages focuses on the project's key academic value; and the suggestions for cooperation entry strategies are adapted to the partner's cooperation intentions and resource capabilities. The generated scripts undergo automatic compliance verification by the system to ensure they conform to pharmaceutical academic promotion standards, providing investment personnel with ready-to-use communication solutions and enhancing the targeting and professionalism of their communication.

[0061] In one embodiment, step S5, which involves collecting business feedback data and dynamically adjusting the weight coefficients in the comprehensive scoring function based on the business feedback data, specifically includes: S51, Collect business feedback data generated by investment promotion personnel during business follow-up. The business feedback data includes whether a cooperative relationship has been established, the progress of cooperation, and the investment promotion personnel's evaluation information on the intention to cooperate. S52, Use the business feedback data as positive or negative samples to construct a training set for model optimization; S53, Based on the model optimization training set, with the goal of minimizing the difference between the scoring result output by the comprehensive scoring function and the actual feedback result, the weight coefficients corresponding to each feature index in the comprehensive scoring function are iteratively updated; S54 deploys the updated weight coefficients to the recommendation calculation module for the next comprehensive partner matching score calculation, forming a continuously optimized intelligent recommendation closed loop.

[0062] As described in steps S51-S54 above, a dynamic optimization mechanism based on business feedback data can be constructed to continuously adjust the weight coefficients in the comprehensive scoring function, enabling the recommendation model to adapt to the dynamic changes in business scenarios, continuously improve the accuracy and stability of partner recommendations, improve the entire intelligent recommendation closed loop, and solve the problem that recommendation model parameters are fixed and cannot be continuously optimized with changes in business in the existing technology.

[0063] Since the business scenarios, partner needs, and market environment of pharmaceutical academic promotion are all dynamically changing, fixed weight coefficients will cause the comprehensive scoring function to gradually deviate from actual business needs, reduce the adaptability of the recommendation model, and make it difficult to maintain recommendation accuracy after long-term use. Therefore, it is necessary to establish a parameter optimization mechanism based on actual business feedback, and ensure that the model always keeps in line with the actual business by continuously iterating and updating the weight coefficients.

[0064] In traditional solutions, the weight coefficients of scoring models are often initially preset or rely solely on one-time experience adjustments, lacking linkage with actual business feedback and failing to dynamically optimize based on collaboration results. This leads to poor model adaptability and insufficient long-term effectiveness of recommendation results. By implementing a complete process of business feedback data collection, training set construction, iterative updates to weight coefficients, and optimized parameter deployment, this approach adjusts model parameters based on actual results, specifically addressing the problems of model rigidity and lack of continuous optimization capabilities inherent in traditional methods.

[0065] First, S51 is executed to collect business feedback data generated by sales personnel during business follow-up. This data comes from the entire process of subsequent business development between sales personnel and partners, and is obtained by the system through a combination of automatic collection and manual input. Specifically, it includes data on whether a cooperative relationship has been established, stage data on the progress of cooperation, and evaluation information on cooperation intentions given by sales personnel based on actual communication and cooperation docking. This data directly reflects the degree to which the recommendation results match the actual business needs, providing core basis for model optimization.

[0066] Next, in step S52, business feedback data is used as positive or negative samples to construct a training set for model optimization. Feedback data indicating established partnerships, smooth progress, and positive feedback from sales personnel are categorized as positive samples, representing that the corresponding weighting coefficients effectively filter out high-quality partners. Feedback data indicating no established partnerships, hindered progress, or negative feedback are categorized as negative samples, indicating that the current weighting coefficients have room for optimization. This classification and organization of positive and negative samples forms a structured training set for model optimization, providing data support for adjusting the weighting coefficients.

[0067] Then, step S53 is executed to iteratively update the weight coefficients based on the model optimization training set. The optimization objective is to minimize the difference between the scoring result output by the comprehensive scoring function and the actual feedback result. The weight coefficients corresponding to each feature indicator in the comprehensive scoring function are adjusted through the model training algorithm. For example, if the historical cooperation success rate feature of multiple positive samples in the training set has a stronger correlation with the actual cooperation result, the system will appropriately increase the weight coefficient corresponding to the historical cooperation success rate through iterative calculation. If the department matching degree feature shows a large deviation from the actual fit in some negative samples, its weight coefficient will be adjusted accordingly to make the weight coefficient configuration more closely match the adaptation logic of actual business.

[0068] Finally, step S54 is executed to deploy the updated weight coefficients to the recommendation calculation module, directly applying them to the next partner comprehensive matching score calculation. The new weight coefficients will be adjusted based on the impact of the latest business feedback on each feature indicator, making the comprehensive score result more accurately reflect the actual suitability value of the partner. This forms a continuous optimization loop of recommendation, feedback, optimization, and re-recommendation, ensuring that the recommendation model can continuously absorb business experience and maintain high recommendation accuracy and stability in the long term.

[0069] like Figure 2 As shown, the present invention also discloses an intelligent communication assistance system for medical academic promotion, comprising: The acquisition module is used to acquire multi-source communication data, parse the multi-source communication data, extract the business characteristic information of partners, and construct partner profiles; The calculation module is used to generate corresponding semantic vectors based on the partner profile and investment project requirements, and calculate the vector similarity to filter out a set of candidate partners. The list generation module is used to obtain the multi-dimensional business characteristics of each candidate partner in the candidate partner set, calculate the comprehensive matching score of each candidate partner based on the preset comprehensive scoring function, and generate a partner recommendation list based on the scoring results. The suggestion generation module is used to obtain the real-time communication content between investment promotion personnel and target partners, identify compliance risks based on a preset compliance rule library and large language model, generate compliance prompt information, and automatically generate promotional script suggestions based on the profile of the target partner and the needs of the investment promotion project. The dynamic adjustment module is used to collect business feedback data and dynamically adjust the weight coefficients in the comprehensive scoring function based on the business feedback data.

[0070] In one embodiment, the computing module includes: The acquisition unit is used to acquire multiple business feature information from the partner profile and convert the multiple business feature information into a corresponding first semantic vector through an embedding model; The conversion unit is used to obtain project information from the investment project requirements and convert it into a corresponding second semantic vector through an embedding model. A calculation unit is used to calculate the semantic similarity between the first semantic vector and the second semantic vector; The first filtering unit is used to select an initial candidate set based on the semantic similarity. The second screening unit is used to perform a secondary screening of the initial candidate set based on a preset business matching factor to obtain the final candidate partner set.

[0071] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described intelligent communication assistance method for medical academic promotion.

[0072] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described intelligent communication assistance method for promoting medical academic research.

[0073] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0074] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0075] The above description is merely a preferred embodiment of the present invention and does not limit the scope of this application. Any equivalent results or equivalent process transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of protection of this application.

Claims

1. An intelligent communication assistance method for medical academic promotion, characterized in that, Includes the following steps: Acquire multi-source communication data, parse the multi-source communication data, extract business characteristic information of partners, and construct partner profiles; Based on the partner profiles and investment project requirements, corresponding semantic vectors are generated, and vector similarity is calculated to filter out a set of candidate partners. The multi-dimensional business characteristics of each candidate partner in the candidate partner set are obtained, the comprehensive matching score of each candidate partner is calculated based on the preset comprehensive scoring function, and a partner recommendation list is generated based on the scoring results. The system acquires real-time communication content between investment promotion personnel and target partners, identifies compliance risks based on a pre-set compliance rule base and large language model, generates compliance prompts, and automatically generates promotional script suggestions based on the profile of the target partner and the needs of the investment promotion project. Collect business feedback data and dynamically adjust the weight coefficients in the comprehensive scoring function based on the business feedback data.

2. The intelligent communication assistance method for medical academic promotion according to claim 1, characterized in that, The steps of acquiring multi-source communication data, parsing the multi-source communication data, extracting business characteristic information of partners, and constructing partner profiles specifically include: Acquire multi-source communication data between investment promotion personnel and potential partners, including telephone communication recordings, WeChat chat text data, and images and documents transmitted during the communication process; The recorded telephone conversation data is processed by speech recognition to obtain first text data, and the image and document data are processed by multimodal parsing to obtain second text data; The first text data, the second text data, and the enterprise WeChat chat text data are merged to construct a business communication text set; Semantic analysis and information extraction are performed on the set of business communication texts to extract the business characteristic information of the potential partners; The extracted business feature information is structured to construct a partner profile with multiple dimensions.

3. The intelligent communication assistance method for medical academic promotion according to claim 1, characterized in that, The step of calculating vector similarity to filter out the candidate partner set specifically includes: Multiple business feature information from the partner profile is obtained, and the multiple business feature information is converted into a corresponding first semantic vector through an embedding model; Obtain project information from the investment promotion project requirements and convert it into a corresponding second semantic vector through an embedding model; Calculate the semantic similarity between the first semantic vector and the second semantic vector; An initial candidate set is selected based on the semantic similarity. The initial candidate set is further filtered based on preset business matching factors to obtain the final candidate partner set.

4. The intelligent communication assistance method for medical academic promotion according to claim 1, characterized in that, The step of generating a partner recommendation list based on the scoring results specifically includes: Obtain the multi-dimensional business characteristics of each candidate partner in the candidate partner set, wherein the multi-dimensional business characteristics include multiple matching degree feature indicators; Each matching feature index is assigned a corresponding weight coefficient, and the comprehensive matching score of each candidate partner is calculated based on the value of each feature index and its corresponding weight coefficient. The comprehensive matching scores of all partners in the candidate partner set are normalized to obtain the normalized comprehensive matching scores. The partners in the candidate partner set are sorted according to the normalized comprehensive matching score, and a preset number of the top-ranked partners are selected to generate a partner recommendation list. Output the recommended list of partners.

5. The intelligent communication assistance method for medical academic promotion according to claim 1, characterized in that, The steps for automatically generating promotional script suggestions specifically include: Real-time acquisition of communication content generated between investment promotion personnel and target partners during the communication process, and conversion of it into text data to be analyzed; Construct a compliance rule base, which includes illegal expression patterns, sensitive terms, and risk scenarios; The compliance rule base is input into the large language model in the form of prompt words. The large language model is used to perform semantic analysis on the text data to be analyzed, identify whether there is any violation risk, and generate a risk level assessment result. When a violation risk is identified, a compliance alert message is generated; Based on the profile information of the target partners and the academic characteristics of the current investment projects, promotional communication scripts are automatically generated.

6. The intelligent communication assistance method for medical academic promotion according to claim 1, characterized in that, The step of collecting business feedback data and dynamically adjusting the weight coefficients in the comprehensive scoring function based on the business feedback data specifically includes: Collect business feedback data generated by investment promotion personnel during business follow-up; The business feedback data is used as a sample to construct a training set for model optimization; Based on the model optimization training set, with the goal of minimizing the difference between the scoring result output by the comprehensive scoring function and the actual feedback result, the weight coefficients corresponding to each feature index in the comprehensive scoring function are iteratively updated. The updated weighting coefficients will be deployed to the recommendation calculation module for the next calculation of the comprehensive partner matching score.

7. An intelligent communication assistance system for medical academic promotion, characterized in that, include: The acquisition module is used to acquire multi-source communication data, parse the multi-source communication data, extract the business characteristic information of partners, and construct partner profiles; The calculation module is used to generate corresponding semantic vectors based on the partner profile and investment project requirements, and calculate the vector similarity to filter out a set of candidate partners. The list generation module is used to obtain the multi-dimensional business characteristics of each candidate partner in the candidate partner set, calculate the comprehensive matching score of each candidate partner based on the preset comprehensive scoring function, and generate a partner recommendation list based on the scoring results. The suggestion generation module is used to obtain the real-time communication content between investment promotion personnel and target partners, identify compliance risks based on a preset compliance rule library and large language model, generate compliance prompt information, and automatically generate promotional script suggestions based on the profile of the target partner and the needs of the investment promotion project. The dynamic adjustment module is used to collect business feedback data and dynamically adjust the weight coefficients in the comprehensive scoring function based on the business feedback data.

8. The intelligent communication assistance system for medical academic promotion according to claim 7, characterized in that, The computing module includes: The acquisition unit is used to acquire multiple business feature information from the partner profile and convert the multiple business feature information into a corresponding first semantic vector through an embedding model; The conversion unit is used to obtain project information from the investment project requirements and convert it into a corresponding second semantic vector through an embedding model. A calculation unit is used to calculate the semantic similarity between the first semantic vector and the second semantic vector; The first filtering unit is used to select an initial candidate set based on the semantic similarity. The second screening unit is used to perform a secondary screening of the initial candidate set based on a preset business matching factor to obtain the final candidate partner set.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images