Method and system for dynamically generating CRM customer data based on geographic semantic analysis

Abstract

The present application relates to the technical field of customer relationship management, in particular to a CRM customer data dynamic generation method and system based on geographic semantic analysis. The specific implementation steps include: parsing the natural language text input by the user and extracting semantic intent; based on real-time geographic space events and time window factors, combining industry prior semantic knowledge, dynamically weighting semantic reasoning of the semantic intent, generating context-aware geographic semantic query representation; performing semantic association calculation on the query representation and the discourse set containing business entity description, identifying candidate entities; aggregating multi-source heterogeneous discourses associated with the candidate entities, extracting event knowledge, and constructing a business semantic portrait; based on the portrait and the industry chain knowledge path generated by reasoning on prior knowledge, constructing customer knowledge entities in the CRM system. Through the geographic semantic analysis technology, the present application realizes intelligent dynamic generation of CRM customer data, and improves the efficiency and accuracy of customer relationship management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of customer relationship management, in particular to a CRM customer data dynamic generation method and system based on geographic semantic analysis. BACKGROUND

[0002] In the field of customer relationship management, to meet the urgent demand for precise customer development in technology-intensive markets such as the electronics industry, the existing technology realizes the visual retrieval of commercial entities in a specific geographic area by integrating geographic information systems with customer relationship management systems. Based on the deep mining technology of internal CRM data, fine behavior prediction and portrait analysis of inventory customers can be realized. Meanwhile, the behavior modeling scheme using complex algorithms can further deepen the pattern insight of customer interaction time series data. The comprehensive application of these technologies lays a solid technical foundation for customer analysis and management using known data.

[0003] However, the existing technology still has problems in automatically and dynamically converting the commercial entities discovered in geographic space into structured and manageable customer data in the customer relationship management system through semantic analysis. The semantic processing capability of the existing technology cannot effectively bridge the gap between raw information and CRM customer entities. Its semantic analysis capability is insufficient to support accurate potential customer identification, and the information extraction and structuring capability is missing. When facing natural language queries containing complex products, processes or industry chain roles, shallow matching based on keywords or fixed categories cannot accurately locate the target entity. Even if the potential entity is identified, it is also impossible to automatically extract time-sensitive business events from unstructured speech and structure them into the data fields of the customer relationship management system to form a customer portrait that can be immediately followed up and contains dynamic event labels.

[0004] In summary, although the existing technology provides a framework for discovering business opportunities using geographic information, it still has defects in the depth analysis of natural language queries, intelligent extraction of multi-source unstructured information, and automatic structured generation of customer relationship management data, which cannot meet the fine and dynamic needs of modern industry. Therefore, the present application proposes a CRM customer data dynamic generation method and system based on geographic semantic analysis. SUMMARY

[0005] The purpose of the present application is to provide a CRM customer data dynamic generation method and system based on geographic semantic analysis, which realizes intelligent dynamic generation of CRM customer data through geographic semantic analysis technology, and improves the efficiency and accuracy of customer relationship management.

[0006] To achieve the above purpose, the present application provides the following technical solutions:

[0007] The CRM customer data dynamic generation method based on geographic semantic analysis comprises:

[0008] Unstructured natural language text input by a user is acquired, and syntax and semantic level analysis is performed to extract semantic intent.

[0009] Based on real-time geographic space events and time window factors, and in combination with an industry prior semantic knowledge containing industry association relationships and an industrial rule base, the semantic intent is dynamically weighted and semantically expanded to generate a context-aware geographic semantic query representation.

[0010] The geographic semantic query representation is semantically associated with a collection of utterances containing natural language descriptions of business entities and their geographic attributes, and candidate entities that meet the conditions are identified.

[0011] Multiple source heterogeneous utterances associated with the candidate entities are aggregated, event knowledge is extracted according to preset data conflict resolution rules, and a business semantic portrait is constructed.

[0012] A customer knowledge entity for customer relationship management in a CRM system is constructed based on the business semantic portrait and an industry chain knowledge path generated by real-time reasoning on the industry prior semantic knowledge.

[0013] Preferably, the industry prior semantic knowledge comprises:

[0014] The industry prior semantic knowledge is a structured data stored in the form of resource description framework triples and oriented to the electronic industry field; the industry prior semantic knowledge provides a basic, quantifiable initial weight for the association between semantic concepts, and the initial weight is dynamically adjusted according to the real-time geographic space events and time window factors.

[0015] Preferably, the dynamic weighting semantic expansion of the semantic intent based on real-time geographic space events and time window factors, and in combination with an industry prior semantic knowledge containing industry association relationships and an industrial rule base, comprises:

[0016] A time window associated with the business rhythm of the electronic industry field is preset, the time window defines the start and end time of business activities, when the current time of the system falls within the valid period of the time window, the industrial rule base triggers a weighting rule, the rule automatically increases the weight factor of the semantic concept related to the typical business activity in the period; a preset geographic area is continuously monitored using geographic fence technology, when a condition-satisfying event is monitored in the geographic area, the industrial rule base triggers another weighting rule, the rule automatically increases the weight factor of the semantic concept related to the event nature.

[0017] Preferably, the event knowledge comprises:

[0018] Continuously monitoring and aggregating real-time network texts related to the candidate entity; analyzing the texts by natural language processing techniques to extract events representing entity labor demand, capacity changes and new product release status, and converting them into structured event knowledge.

[0019] Preferably, aggregating multi-source heterogeneous utterances associated with the candidate entity, extracting event knowledge according to preset data conflict resolution rules, and constructing a business semantic portrait, comprising:

[0020] When information about the same attribute aggregated from different sources conflicts, select according to the preset credibility priority; when the geographic coordinates obtained from the point of interest data deviate from the geographic coordinates parsed from the official website text of the entity by more than a preset distance threshold, the rule defines that the geographic coordinates parsed from the official website text of the entity are preferred.

[0021] Preferably, based on the business semantic portrait and the industry chain knowledge path generated by real-time reasoning on the industry prior semantic knowledge, comprising:

[0022] In the industry prior semantic knowledge, locate the node representing the newly created customer data entity and the multiple nodes representing the existing customer data entities; take the new customer entity node as the starting point, traverse and reason the relationship path in the industry prior semantic knowledge according to the pre-defined logical rules, and find the relationship path between the new customer entity node and the existing customer entity node; convert the reasoned relationship path into a knowledge path representing the upstream and downstream cooperation relationship and the horizontal competition relationship between entities.

[0023] The CRM customer data dynamic generation system based on geographic semantic analysis comprises:

[0024] Semantic intent analysis module: obtaining natural language input by the user, and analyzing the semantic intent by natural language processing techniques;

[0025] Dynamic semantic expansion module: based on real-time geographic space events and time window factors, and combined with an industry prior semantic knowledge containing industry association relationships and an industry rule library, dynamically weighting semantic expansion on the semantic intent, and generating context-aware geographic semantic query representation;

[0026] Semantic association and recognition module: performing semantic association calculation on the geographic semantic query representation and a collection of utterances containing natural language descriptions of business entities and their geographic attributes, and identifying candidate entities that meet the conditions;

[0027] The multi-source discourse fusion and portrait generation module: aggregates multi-source heterogeneous discourses associated with the candidate entity, extracts event knowledge according to preset data conflict resolution rules, and constructs a business semantic portrait;

[0028] The customer knowledge entity construction and integration module: receives a selection instruction of the user on the candidate entity, generates an industry chain knowledge path based on the business semantic portrait and through real-time reasoning on the industry prior semantic knowledge, and constructs a customer knowledge entity for customer relationship management in a CRM system.

[0029] Compared with the prior art, the present application has the following beneficial effects:

[0030] 1. The present application converts a fuzzy natural language query into a machine-understandable, dynamically weighted semantic paradigm by constructing a context-aware geographic semantic query representation; improves the traditional shallow analysis method based on static keyword matching, and improves the recognition accuracy of complex business intent and the accuracy of potential customer positioning by fusing real-time geographic spatial events and time window factors, and combining industry prior semantic knowledge containing industrial correlation.

[0031] 2. The present application constructs a semantic portrait containing entity real-time business dynamics by automatically extracting event knowledge from multi-source heterogeneous discourses; improves the deficiency of the traditional customer portrait containing only static and basic attributes, and enhances the timeliness and follow-up value of customer data by continuously monitoring and processing network text to extract structured events such as labor demand and capacity changes.

[0032] 3. The present application automatically generates a knowledge path representing the deep industrial relationship between entities by real-time reasoning on industry prior semantic knowledge; reveals the upstream and downstream cooperation and horizontal competition relationship between new customers and existing customer ecology by relationship traversal and logical reasoning, improves the defects of traditional CRM systems that customer relationship depends on manual maintenance and lacks a macro-industrial perspective, and provides a deep decision basis for precise strategic sales and risk assessment. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The figure is a CRM customer data dynamic generation method based on geographic semantic analysis of the embodiment of the present application;

[0034] Figure 2 The figure is a dynamically weighted semantic extension principle diagram of the example of the present application;

[0035] Figure 3 The figure is a business semantic portrait construction diagram of the example of the present application;

[0036] Figure 4 The figure is a system module architecture diagram of the present application. DETAILED DESCRIPTION

[0037] Next, the technical solution of the present invention will be explained in detail and clearly with reference to specific examples and related drawings. It should be noted that the examples described below represent only a part of the present invention, not all of it. Based on these examples, those skilled in the art can deduce other possible embodiments of the present invention without creative effort, and these embodiments are also protected by the present invention.

[0038] Example 1:

[0039] This embodiment provides a method for dynamically generating CRM customer data based on geographic semantic analysis for the electronics industry. This method is implemented through a CRM system platform that integrates a Geographic Information System (GIS) and supports a "map-based customer management + visual retrieval" mode. (Refer to...) Figure 1 The method includes: S1. Acquiring unstructured natural language text input by the user and performing syntactic and semantic parsing to extract semantic intent; S2. Based on real-time geospatial events and time window factors, and combined with an industry prior semantic knowledge containing industry relationships and an industrial rule base, dynamically weighting the semantic intent to generate a context-aware geospatial query representation; S3. Performing semantic association calculations between the geospatial query representation and a discourse set containing natural language descriptions of business entities and their geographical attributes to identify candidate entities that meet the conditions; S4. Aggregating multi-source heterogeneous discourses associated with the candidate entities, extracting event-related knowledge according to preset data conflict resolution rules, and constructing a business semantic profile; S5. Receiving the user's selection instruction for the candidate entities, and constructing a customer knowledge entity for customer relationship management in the CRM system based on the business semantic profile and the industry chain knowledge path generated by real-time reasoning of the industry prior semantic knowledge.

[0040] Furthermore, the unstructured natural language text input by the user is acquired and parsed at the syntactic and semantic levels to extract semantic intent; corresponding to step S1 above, the specific implementation process includes:

[0041] The system receives unstructured natural language text queries input by users. The natural language processing engine within the system first performs syntactic and semantic analysis on the text, including word segmentation, part-of-speech tagging, and named entity recognition. It accurately identifies key semantic elements such as geographical scope, core technologies, core products, target industries, and dynamic events, and organizes these elements into a structured semantic intent representation.

[0042] Furthermore, based on real-time geospatial events and time window factors, and combined with an industry-prior semantic knowledge base containing industry-related relationships and an industrial rule base, the semantic intent is dynamically weighted and semantically expanded to generate a context-aware geospatial query representation; (Refer to...) Figure 2 Corresponding to step S2 above, the specific implementation process includes:

[0043] The system utilizes pre-defined industry-specific semantic knowledge and an industry rule base tailored to the electronics industry to expand the extracted semantic intent. The industry-specific semantic knowledge is structured data stored in the form of resource description framework triples, encompassing information such as enterprise entities, product categories, technical specifications, and supply chain relationships within the electronics industry. This information is stored in subject-predicate-object triples, for example, "Company A - produces - chip B" or "Chip B - applied to - smartphone C".

[0044] The industry-prior semantic knowledge provides a basic, quantifiable initial weight for the association between semantic concepts. This initial weight is dynamically adjusted based on real-time geospatial events and time window factors. Specifically, the system assigns an initial weight value to each pair of related semantic concepts, reflecting the strength of the association between the two concepts. For example, the association weight between "semiconductor manufacturing" and "wafer fab" might be set to 0.9, indicating a high degree of correlation; while the association weight between "semiconductor manufacturing" and "software development" might be set to 0.3, indicating a lower degree of correlation. These initial weight values ​​are not fixed but are dynamically adjusted based on real-time geospatial events and time window factors.

[0045] This application's embodiments utilize prior semantic knowledge from the industry to introduce precise entity and concept associations from the electronics industry as prior constraints into the semantic expansion process. By assigning initial weights to the associations between semantic concepts and dynamically adjusting these weights in conjunction with real-time geospatial events and time window factors, a context-aware geographic semantic query representation is generated that possesses both domain knowledge depth and can reflect real-time changes in the market environment, thereby improving the accuracy and timeliness of subsequent potential customer identification.

[0046] A time window is pre-defined that is associated with the business rhythm of scenarios in the electronics industry, such as "a power component technology exhibition". This time window defines the start and end times of the exhibition. When the system detects that the current time falls into this time window, the industrial rule base will trigger the corresponding weighting rules to automatically increase the weight factors of semantic concepts related to "new product launch", "power semiconductor", and "semiconductor". This makes the system pay more attention to these concepts related to the current business rhythm when performing semantic expansion.

[0047] Meanwhile, the system can set up geofences around an important electronics industrial park and continuously monitor a preset geographical area using geofencing technology. When events such as the start of a new factory, the holding of a technical seminar, or a large-scale recruitment event are detected in the area, the industrial rule base will trigger corresponding weighted rules to increase the weight factors of semantic concepts related to these events (such as "capacity expansion", "technological innovation", "talent demand"), so that the system pays more attention to these concepts related to the current geospatial events when performing semantic expansion.

[0048] When performing dynamically weighted semantic expansion, the ambiguity of user identity can lead to unclear expansion directions. An adaptive semantic focus drift mechanism can be employed. During dynamic weighting, this mechanism analyzes the user's historical query behavior. A pre-trained classifier calculates the association probability between each candidate concept and preset identity tags such as "upstream technology" or "downstream application," selecting the semantic concept cluster corresponding to the highest probability identity tag and embedding it into the query representation as the priority expansion direction. The adaptive semantic focus drift mechanism dynamically adjusts the exploration direction of semantic expansion based on the user's implicit identity, improving the depth of understanding personalized query intent and addressing the issue that general semantic expansion cannot meet the refined needs of expert users.

[0049] This application improves upon the traditional, static semantic expansion method by introducing a dynamic weighting mechanism based on real-time geographic events and business rhythm time windows. This results in a context-aware query representation that can reflect market hotspots and regional dynamics in real time, thereby enhancing the timeliness and accuracy of subsequent entity recognition.

[0050] Furthermore, the geographic semantic query representation is semantically associated with a discourse set containing natural language descriptions of business entities and their geographic attributes to identify candidate entities that meet the criteria; corresponding to step S3 above, the specific implementation process includes:

[0051] A pre-trained deep language model, such as SciBERT or FinBERT (variants of BERT), is selected as the foundational tool for semantic vectorization. This model is then fine-tuned using a publicly available dataset from the electronics industry to enable it to generate semantic vectors that more accurately capture subtle semantic differences within a specific industry. During association computation, the system inputs pre-generated, context-aware geographic semantic query representations into the deep language model. These representations are processed by its multi-layered Transformer encoder, which extracts the output vectors from the final hidden layers to generate a high-dimensional query vector that comprehensively captures the user's complex query intent. Simultaneously, the system processes a massive collection of discourses containing millions of business entity descriptions, inputting each individual text discourse into the model to generate corresponding entity description vectors. All generated entity description vectors and their corresponding entity IDs are constructed into an efficient vector index. Using an approximate nearest neighbor search library such as FAISS or Annoy, when a query vector is received, the system performs a similarity search in the vector index, calculates the cosine similarity between the query vector and each entity description vector in the index, and finally returns a list of entities sorted from high to low by similarity score. Based on a preset semantic similarity threshold, all entities with scores higher than the threshold are selected as candidate entity sets and output, which are then directly visualized on the map of the CRM system's GIS interface as highlighted icons.

[0052] Furthermore, multi-source heterogeneous discourses associated with the candidate entities are aggregated, and event-based knowledge is extracted according to preset data conflict resolution rules to construct a business semantic profile; (referencing...) Figure 3 Corresponding to step S4 above, the specific implementation process includes:

[0053] When a user clicks on a candidate entity icon on the map, the system continuously collects various online texts related to the candidate entity, such as news reports, social media discussions, company announcements, and industry analysis reports, through multiple methods including web crawlers, RSS subscriptions, and API interfaces. These texts come from diverse sources, including official company websites, industry portals, financial media, and social platforms. The system preprocesses these texts, including noise removal, format standardization, and language standardization, to facilitate subsequent analysis and processing.

[0054] The text is analyzed using natural language processing techniques such as named entity recognition, relation extraction, and event detection to extract events that characterize entity labor demand, capacity changes, and new product launch status. These events are then transformed into structured event-based knowledge. For example, recruitment information can be used to extract a company's labor demand (e.g., Company A is recruiting 100 chip design engineers), capacity reports can be used to extract capacity changes (e.g., Company B's wafer fab's monthly capacity has increased from 100,000 wafers to 150,000 wafers), and product launch information can be used to extract new product launch status (e.g., Company C plans to release a new generation of 5G chips next month). This extracted information is then transformed into structured event-based knowledge.

[0055] This embodiment continuously monitors and aggregates multi-source, real-time online text related to candidate entities, and uses natural language processing technology for in-depth analysis. It automatically extracts and structures event-based knowledge of dynamic business changes such as entity labor demand and capacity changes from unstructured discourse, thereby enhancing the timeliness and business insight value of the final generated business semantic profile.

[0056] When conflicting information about the same attribute aggregated from different sources occurs, the system prioritizes information based on a predefined credibility priority. For example, information released by the company's official website has the highest credibility, followed by authoritative media reports, then industry analyst opinions, and finally, social media discussions. When the system obtains conflicting information about the same attribute (such as company size, production capacity data, or technology roadmap) from different sources, it will prioritize the information from the source with higher credibility. For instance, if a company's official website states its annual production capacity is 1 million units, while a social media discussion mentions that the company's annual production capacity is 800,000 units, the system will prioritize the data from the official website.

[0057] When conflicting information about the same attribute aggregated from different sources occurs, simply selecting based on source credibility may lead to misjudgment. A knowledge graph-based cross-validation mechanism can be employed. This mechanism retrieves existing, highly credible fact triples related to the conflicting information from prior industry semantic knowledge. A logical reasoning engine calculates the semantic consistency or logical implication between each conflicting piece of information and these fact triples, selecting the conflicting information with the highest consistency or implication and embedding it into a business semantic profile as the final adopted attribute value. This knowledge graph-based cross-validation mechanism improves the accuracy and credibility of data fusion by introducing domain background knowledge to fact-check conflicting information, addressing the lack of deep logical judgment capabilities in traditional conflict resolution rules.

[0058] The system sets a distance threshold (e.g., 500 meters). When the distance between the company's geographic coordinates obtained from the map's point of interest database and the geographic coordinates converted from the address parsed from the company's official website exceeds this threshold, the system will prioritize the address information provided by the official website. This is because the address information on the official website is usually more accurate and up-to-date, while point of interest data may have issues with outdated updates or inaccurate positioning. For example, if a company has recently moved to a new address, its official website may have updated the new address information, but the map's point of interest database may not have. In this case, the system will prioritize the new address information provided by the official website.

[0059] This embodiment improves upon traditional CRM systems by automatically extracting structured event-based knowledge from multi-source, unstructured real-time discourse, thus creating a business semantic profile that reflects the latest business dynamics of an enterprise and includes time-sensitive event tags, providing real-time business intelligence for sales decisions.

[0060] Furthermore, receiving the user's selection instruction for the candidate entity, and based on the business semantic profile and the industry chain knowledge path generated through real-time reasoning of the industry's prior semantic knowledge, a customer knowledge entity for customer relationship management is constructed in the CRM system; corresponding to step S5 above, the specific implementation process includes:

[0061] On the map of the GIS interface at the system's front end, multiple candidate entities and their business semantic profile summaries generated above are displayed. When the system receives a confirmation selection instruction from the user for one or more candidate entities through interactive methods such as clicking or checking, the system immediately locks the selected new customer and uses its constructed business semantic profile containing dynamic event-based knowledge as data to be processed. At the same time, taking this new customer as the starting entity, the system performs real-time reasoning in the industry's prior semantic knowledge. The system will find the node corresponding to the newly created customer data entity in the industry's prior semantic knowledge base, and locate multiple nodes representing existing customers in the CRM system. These nodes represent different enterprise entities, which are connected by various relational edges to form a complex network structure.

[0062] Starting from the new customer entity node, the system will traverse and reason in the prior semantic knowledge of the industry according to predefined logical rules (such as supplier-customer relationship, competitor relationship, technical cooperation relationship, etc.). This process not only considers the first-order relationship of direct connection, but also the higher-order relationship of multi-hop connection, and explores various possible associations between the new customer and the existing customer. For example, the system may find that the new customer is the upstream supplier of an existing customer, or the technical partner of another existing customer.

[0063] The system then further processes and transforms the relationship paths deduced in the previous step, forming more structured and semantic knowledge paths. These knowledge paths clearly represent the upstream and downstream collaborative relationships (such as supply chain relationships, value chain positions, etc.) and horizontal competitive relationships (such as market competition, technological competition, etc.) between new and existing customers. For example, the system might generate the following knowledge path: New customer A is a core component supplier for existing customer B, and also directly competes with existing customer C in the same market segment. These knowledge paths provide sales and marketing personnel with valuable business insights, helping them better understand the new customer's position and role in the entire industry ecosystem.

[0064] When performing real-time reasoning and generating industry chain knowledge paths, a link prediction mechanism based on Graph Attention Networks (GAT) can be applied to identify potential relationships between new and existing customers that may not yet be explicitly connected. This mechanism inputs the attribute vectors of new customer entity nodes and existing customer entity nodes into the GAT model. The GAT model aggregates neighbor node information and calculates the attention coefficient between nodes, generating a probability of a future link between each pair of "new customer - existing customer" nodes. Potential relationships with probabilities higher than a preset threshold are then selected and embedded into the CRM system as early warning or opportunistic relationship labels. This link prediction mechanism based on graph attention networks improves the foresight of customer relationship management by learning and predicting future relationship evolution from graph structure and node attributes, addressing the limitation of traditional reasoning which can only discover static, existing relationships.

[0065] This embodiment improves upon the traditional CRM model of isolated customer information and manual relationship maintenance by performing real-time reasoning and generating industry chain knowledge paths when constructing customer entities. It automatically reveals the upstream and downstream position and collaborative relationships of customers in the entire industry ecosystem, providing a deep business insight.

[0066] The system integrates all information, creates a new customer record in the CRM system, and populates the corresponding fields with basic information and dynamic event-based knowledge from the business semantic profile. The generated industry chain knowledge path is integrated into the customer record as structured relational data. At the same time, the entity record is bidirectionally associated with the geographic markers on the GIS interface, realizing a refined management model of "managing people with maps". Thus, a customer knowledge entity with rich information, including its own dynamic profile, position and relationship in the industry ecosystem, is successfully built in the CRM system.

[0067] This embodiment introduces a semantic expansion mechanism based on real-time contextual dynamic weighting to achieve accurate and context-aware understanding of the deep intent of users' natural language queries. By automatically extracting structured event-based knowledge from multi-source heterogeneous discourse, it enhances the dynamism and timeliness of traditional customer profiles that only contain static attributes. Furthermore, by performing real-time reasoning during customer entity construction to generate industry chain knowledge paths, it improves the current situation in customer relationship management that relies on manual maintenance and lacks a macro-industry perspective. Ultimately, it constructs an end-to-end technical solution that can respond to natural language queries and automatically discover and generate high-quality customer knowledge entities containing dynamic events and deep industry relationships from open domain information, improving the efficiency and depth of enterprises in conducting refined customer mining and strategic relationship management in target markets.

[0068] Example 2:

[0069] This embodiment provides a CRM customer data dynamic generation system based on geographic semantic analysis. (Refer to...) Figure 4 The system modules include a semantic intent parsing module, a dynamic semantic expansion module, a semantic association and recognition module, a multi-source discourse fusion and profile generation module, and a customer knowledge entity construction and integration module.

[0070] The semantic intent parsing module is responsible for receiving and processing natural language queries input by users. This module employs advanced natural language processing technologies, including word segmentation, part-of-speech tagging, syntactic analysis, and semantic understanding, to comprehensively parse the unstructured text input by users. For example, when a user inputs "find medium-sized enterprises in City A that produce high-end chips," the module will identify key semantic elements, such as the geographical location "City A," the product category "high-end chips," and the enterprise size "medium-sized enterprises," and organize these elements into a structured semantic intent representation.

[0071] The dynamic semantic expansion module receives the output from the semantic intent parsing module and semantically expands it. This module utilizes prior industry semantic knowledge stored in the electronics industry knowledge base, combined with real-time geospatial events and time window factors, to dynamically weight and expand the initial semantic intent, generating a context-aware geospatial query representation that reflects the current business environment. For example, if it's currently the electronics product launch season, the system will automatically increase the weight of semantic concepts related to new product launches; if a large electronics exhibition is detected in the target area, the system will increase the weight of semantic concepts related to the exhibitors.

[0072] The semantic association and recognition module performs semantic association calculations between the geographic semantic query representation generated by the dynamic semantic expansion module and the set of business entity discourses stored in the system. This discourse set contains natural language descriptions of various business entities and their geographic attributes, such as company website introductions, news reports, and industry analysis reports. This module uses a pre-trained deep language model to transform the query representation and discourse text into high-dimensional semantic vectors, and calculates cosine similarity in an efficient vector index library to accurately identify candidate entities that are highly semantically related to the user's query intent. For example, for the above query, the system may identify multiple medium-sized enterprises located in City A, engaged in high-end chip production, such as "Company A," "Company B," and "Company C," as candidate entities.

[0073] The multi-source discourse fusion and profile generation module is responsible for aggregating heterogeneous discourses related to candidate entities from multiple sources and extracting valuable information to construct a business semantic profile. This module collects text data from various sources, including company websites, news media, social media platforms, and industry reports, and uses natural language processing techniques to extract various attributes and event information of entities. Utilizing event detection and relation extraction technologies, the module automatically extracts structured event-based knowledge that represents the dynamic changes of entities from these unstructured texts. When information from different sources conflicts, the system processes it according to preset data conflict resolution rules, ultimately constructing a business semantic profile containing dynamic event tags. For example, the system may learn about Company A's employee size from different sources. If the official website shows "more than 500 people" while a news report states "about 450 people," the system will prioritize the information from the official website.

[0074] The Customer Knowledge Entity Construction and Integration module is the final stage of the system. It is responsible for integrating the information generated by the previous modules into a complete customer knowledge entity and integrating it into the CRM system. This module first receives the user's selection instruction for candidate entities. Then, based on the business semantic profile of the selected entity and combined with the industry chain knowledge path generated through real-time reasoning of prior industry semantic knowledge, it constructs a comprehensive customer knowledge entity. This knowledge entity not only includes the customer's basic information but also deeper information such as its position in the industry chain and its relationships with other customers. It is dynamically created and integrated into the CRM system, while ensuring that the CRM entity record establishes a two-way association with the geographic markers on the GIS interface to achieve a "manage people with maps" management model.

[0075] This application's embodiments organically integrate modules such as semantic intent parsing, dynamic semantic expansion, vectorized association calculation, event-based knowledge extraction, and industry chain knowledge path reasoning to construct an automated, end-to-end processing pipeline from natural language input to structured knowledge entity output. By tightly integrating GIS visualization and interaction modules, it realizes an intuitive operation mode of "visual retrieval" and "managing people with maps." Ultimately, it provides an intelligent system for customer relationship management in the industrial field that can proactively discover, deeply understand, and dynamically generate high-quality potential customer data, improving the accuracy and efficiency of market expansion.

[0076] Although specific examples of the invention have been shown, those skilled in the art will understand that various adjustments, modifications, substitutions, and improvements can be made to these examples without departing from the core principles and spirit of the invention. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A CRM customer data dynamic generation method based on geographic semantic analysis, characterized in that, The method comprises the following steps: acquiring unstructured natural language text input by a user, and performing syntax and semantic analysis to extract semantic intent; based on real-time geospatial events and time window factors, and in combination with an industry prior semantic knowledge comprising industry correlation and an industrial rule base, performing dynamic weighted semantic expansion on the semantic intent to generate context-aware geospatial semantic query representation; the industry prior semantic knowledge is structured data stored in the form of resource description framework triples and oriented to the electronic industry field; the industry prior semantic knowledge provides a basic, quantifiable initial weight for the correlation between semantic concepts, which is dynamically adjusted according to the real-time geospatial events and time window factors; the dynamic weighted semantic expansion on the semantic intent comprises presetting a time window associated with the business rhythm of the electronic industry field, the time window defining the start and end time of business activities, when the current time of the system falls within the valid period of the time window, the industrial rule base triggers a weighting rule, which automatically increases the weight factor of semantic concepts related to typical business activities in the period; using geofencing technology to continuously monitor a preset geographic area, when an event meeting the conditions is monitored in the geographic area, the industrial rule base triggers another weighting rule, which automatically increases the weight factor of semantic concepts related to the nature of the event; performing semantic correlation calculation on the geospatial semantic query representation and a discourse set comprising natural language descriptions of business entities and their geographic attributes, and identifying candidate entities meeting the conditions; aggregating multi-source heterogeneous discourses associated with the candidate entities, extracting event knowledge according to preset data conflict resolution rules, and constructing a business semantic portrait; receiving a selection instruction of the user on the candidate entity, constructing a customer knowledge entity for customer relationship management in a CRM system based on the business semantic portrait and an industry chain knowledge path generated by real-time reasoning on the industry prior semantic knowledge.

2. The CRM customer data dynamic generation method based on geographic semantic analysis according to claim 1, characterized in that, The event knowledge comprises: continuously monitoring and aggregating real-time network texts related to the candidate entity; using natural language processing technology to analyze the real-time network texts, extracting events representing entity labor demand, capacity change and new product release status, and converting them into structured event knowledge.

3. The CRM customer data dynamic generation method based on geographic semantic analysis according to claim 1, characterized in that, aggregating multi-source heterogeneous discourses associated with the candidate entities, extracting event knowledge according to preset data conflict resolution rules, and constructing a business semantic portrait, comprising: when information about the same attribute aggregated from different sources conflicts, selecting according to the preset credibility priority; when the deviation between the geographic coordinates obtained from the point of interest data and the geographic coordinates parsed from the official website text of the entity is greater than the preset distance threshold, the preset data conflict resolution rule defines that the geographic coordinates parsed from the official website text are adopted.

4. The CRM customer data dynamic generation method based on geographic semantic analysis according to claim 1, characterized in that, based on the business semantic portrait and the industry chain knowledge path generated by real-time reasoning on the industry prior semantic knowledge, comprising: In the industry prior semantic knowledge, the node representing the newly created customer data entity and the plurality of nodes representing the existing customer data entities are located; starting from the new customer entity node, the relationship path between the new customer entity node and the existing customer entity nodes is discovered by traversing and real-time reasoning according to a predefined logical rule in the industry prior semantic knowledge; and the real-time reasoned relationship path is converted into an industry chain knowledge path representing the upstream and downstream cooperation relationship and the horizontal competition relationship between entities.

5. A CRM customer data dynamic generation system based on geographic semantic analysis, characterized in that, The method of claim 1 is executed, comprising: a semantic intention analysis module: obtaining natural language input by a user, and analyzing semantic intention by natural language processing technology; a dynamic semantic expansion module: based on real-time geospatial events and time window factors, and in combination with an industry prior semantic knowledge containing industry association relationship and an industrial rule base, performing dynamic weighted semantic expansion on the semantic intention to generate context-aware geospatial semantic query representation; a semantic association and recognition module: performing semantic association calculation on the geospatial semantic query representation and a discourse set containing natural language description of business entities and their geographic attributes, and identifying candidate entities meeting the conditions; a multi-source discourse fusion and portrait generation module: aggregating multi-source heterogeneous discourses associated with the candidate entities, extracting event knowledge according to preset data conflict resolution rules, and constructing a business semantic portrait; a customer knowledge entity construction and integration module: receiving a selection instruction of the user on the candidate entities, constructing a customer knowledge entity for customer relationship management in a CRM system based on the business semantic portrait and an industry chain knowledge path generated by real-time reasoning on the industry prior semantic knowledge.

Images