Business consultation dialogue analysis method and big data system based on Internet business

Abstract

The present invention relates to the field of text analysis technology, and specifically provides a method and a big data system for parsing business consultation dialogues based on Internet services. The present invention achieves effective positioning of consultation dialogues by acquiring and screening target consultation dialogue texts in business consultation logs of business affairs interaction terminals. By parsing these texts for conversation interest events, we have a deep understanding of the user's concerns and needs in the consultation process. At the same time, frequent text paragraphs are extracted and keyword vectors are parsed, providing rich feature data for semantic analysis. Finally, based on the conversation interest event information or keyword vector information, semantic annotation processing is performed on members in the demand semantic feature space, and the semantics of business consultation needs are successfully obtained. This series of technical innovations improves the accuracy of understanding consultation needs, provides data support for service optimization and process improvement of enterprises, and significantly enhances the intelligence and personalization level of enterprise operation services.

Description

Technical Field

[0001] The present invention relates to the field of text analysis technology, and in particular to an enterprise consulting dialogue parsing method and a big data system based on Internet business. Background Art

[0002] In current enterprise operation services, with the diversification of customer consultation channels and the complexity of consultation content, how to accurately and efficiently understand and analyze customer consultation needs has become a technical problem that needs to be solved. Traditional consultation demand analysis methods often rely on manual interpretation and statistics, which is not only inefficient, but also difficult to deeply explore potential needs and related information in consultation conversations. In addition, since the text data of consultation conversations usually contains a large amount of unstructured information, how to extract valuable features from it and perform effective semantic annotation is also a major challenge facing current technology. Summary of the invention

[0003] The present invention at least provides a method for analyzing business consultation dialogues based on Internet services and a big data system.

[0004] The present invention provides a method for parsing business consultation dialogues based on Internet business, which is applied to a big data system. The method comprises: obtaining a business consultation log of a business affairs interaction terminal, and determining that the business consultation log contains at least two target consultation dialogue texts of the business affairs interaction terminal; the business consultation log refers to log information containing user dialogue behavior data, and the business consultation log includes multiple multi-terminal consultation dialogue texts; the at least two target consultation dialogue texts belong to the multiple multi-terminal consultation dialogue texts; for each of the target consultation dialogue texts, performing dialogue interest event parsing on the business affairs interaction terminal to obtain dialogue interest event information of the business affairs interaction terminal; extracting frequent text paragraphs of the business affairs interaction terminal from at least part of the target consultation dialogue texts, performing keyword vector parsing on the frequent text paragraphs to obtain keyword vector information of the business affairs interaction terminal; and performing semantic annotation processing on feature space members in a demand semantic feature space based on at least one category of the dialogue interest event information or the keyword vector information to obtain the business affairs consultation demand semantics of the business affairs interaction terminal in the demand semantic feature space.

[0005] The present invention also provides a big data system, comprising a processor and a memory; the processor and the memory are communicatively connected, and the processor is used to read a computer program from the memory and execute it to implement the above method.

[0006] The present invention also provides a computer-readable storage medium on which a computer program is stored, and the computer program implements the above method when running.

[0007] In view of the above background technology, there is an urgent need for a technical solution that can automatically process consultation logs, accurately extract conversation interest events and keyword vectors, and effectively annotate the demand semantic feature space, so as to improve the intelligence and personalization level of enterprise operation services and better meet the diverse needs of customers.

[0008] The present invention achieves effective positioning of consulting conversations by acquiring and screening target consulting conversation texts in the business consulting logs of the enterprise affairs interactive terminal. By parsing these texts for conversation interest events, we have a deep understanding of the user's concerns and needs in the consultation process. At the same time, frequent text paragraphs are extracted and keyword vector parsed, providing rich feature data for semantic analysis. Finally, based on the conversation interest event information or keyword vector information, semantic annotation processing is performed on the members in the demand semantic feature space, and the semantics of the enterprise affairs consulting needs are successfully obtained. This series of technological innovations improves the accuracy of understanding consulting needs, provides data support for the service optimization and process improvement of enterprises, and significantly enhances the intelligence and personalization level of enterprise operation services.

[0009] For a description of the effects of the above-mentioned big data system and computer-readable storage medium, please refer to the description of the above-mentioned method.

[0010] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present invention and are used together with the specification to illustrate the technical solution of the present invention. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 It is a block diagram of a big data system shown in the present invention.

[0013] Figure 2 It is a flow chart of a method for analyzing business consultation dialogue based on Internet business shown in the present invention. DETAILED DESCRIPTION

[0014] Here, exemplary embodiments will be described in detail, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are only examples of devices and methods consistent with some aspects of the present invention.

[0015] Figure 1 The structural diagram of the big data system 10 provided by the present invention includes a processor 102, a memory 104, and a bus 106. Among them, the memory 104 is used to store execution instructions, including internal memory and external memory. The internal memory can also be understood as an internal memory, which is used to temporarily store the operation data in the processor 102 and the data exchanged with the external memory such as the hard disk. The processor 102 exchanges data with the external memory through the internal memory. When the big data system 10 is running, the processor 102 communicates with the memory 104 through the bus 106, so that the processor 102 executes the enterprise consultation dialogue parsing method based on Internet business of the present invention.

[0016] Please combine Figure 2 , Figure 2 It is a flow chart of a method for analyzing business consultation dialogue based on Internet business provided by the present invention, which is applied to a big data system. The method may exemplarily include the following steps 110 to 140.

[0017] In today's era of rapid development of informatization, the amount of data generated in enterprise and institution operation services is growing explosively. In order to effectively manage and utilize this data, this application introduces how to deeply analyze business consulting logs in enterprise and institution operation services and dig out the consulting demand semantics hidden in them.

[0018] Take a large manufacturing company as an example. Its customer service department needs to handle a large number of business consultations from different channels (such as telephone, email, online chat, etc.) every day. These consultations cover product usage, technical support, after-sales service, order inquiries, etc. In order to better understand customer needs, optimize service processes, and improve corporate operational efficiency, the company decided to introduce a big data system to conduct in-depth analysis of the business consultation logs generated by its customer service department.

[0019] Step 110: The big data system obtains the business consultation log of the enterprise affairs interaction terminal, and determines that the business consultation log contains at least two target consultation dialogue texts of the enterprise affairs interaction terminal; the business consultation log refers to log information containing user dialogue behavior data, and the business consultation log includes multiple multi-terminal consultation dialogue texts; the at least two target consultation dialogue texts belong to the multiple multi-terminal consultation dialogue texts.

[0020] First, the big data system obtains the business consultation logs generated by the customer service department in real time through the preset data interface. The business consultation logs record in detail the time, customer ID, customer service representative ID, conversation content and other information of each consultation session, forming a huge data set containing user conversation behavior data.

[0021] The big data system pre-processes these logs, including data cleaning, format unification, deduplication and other steps to ensure the accuracy and efficiency of subsequent analysis. Considering that customers may consult multiple times through different channels and each consultation may involve multiple topics, the system uses advanced text matching and conversation tracking technology to associate these scattered consultation dialogue texts to form a complete consultation conversation chain.

[0022] On this basis, the big data system further selects conversation chains containing at least two consultation dialogue texts as the target consultation dialogue text set. Such screening criteria ensure that there is enough data for subsequent analysis, and also help capture the continuous needs and concerns expressed by customers in multiple interactions.

[0023] Step 120: The big data system performs conversation interest event analysis on the enterprise affairs interaction terminal for each target consultation conversation text, and obtains conversation interest event information of the enterprise affairs interaction terminal.

[0024] For each target consultation dialogue text, the big data system performs dialogue interest event analysis. This step aims to identify specific interest or demand events expressed by customers in the dialogue, such as "asking about product usage methods", "expressing technical support needs", "making after-sales service requests", etc.

[0025] The big data system uses natural language processing (NLP) technology, combined with pre-built knowledge graphs in the field of enterprise operations, to conduct in-depth semantic understanding of the conversation text. Through subtasks such as entity recognition, relationship extraction, and event classification, the big data system can extract key events and their attributes (such as event type, occurrence time, and objects involved) in the conversation.

[0026] For example, when the big data system recognizes a sentence like "I recently encountered some problems when using your product. Can you provide some technical support?", it will automatically mark the conversation interest event "expressing technical support needs" and extract the relevant time information and customer ID.

[0027] Step 130: The big data system extracts frequent text paragraphs of the enterprise affairs interaction terminal from at least part of the target consultation dialogue text, performs keyword vector analysis on the frequent text paragraphs, and obtains keyword vector information of the enterprise affairs interaction terminal.

[0028] In order to more comprehensively capture the conversational characteristics of the customer service department, the big data system also extracts frequently occurring text paragraphs from the target consultation conversation text, that is, those text fragments that appear repeatedly in multiple conversations and may represent common customer concerns or common responses from customer service.

[0029] The big data system performs keyword vector analysis on these frequent text paragraphs and converts the text into a vector representation in a high-dimensional space using word embedding technology, such as Word2Vec, GloVe, or BERT. Through this conversion, the big data system can capture the semantic similarities and differences between texts, thereby further analyzing the potential needs of customers and the response patterns of customer service.

[0030] For example, a big data system may find that keywords such as "operating guide", "troubleshooting", and "parts replacement" appear frequently in multiple conversations. These keyword vectors not only reflect customers' common consulting needs, but also provide an important feature basis for subsequent semantic labeling.

[0031] Step 140: The big data system performs semantic annotation processing on feature space members in the demand semantic feature space based on at least one category of the conversation interest event information or the keyword vector information, and obtains the business consultation demand semantics of the business affairs interaction terminal in the demand semantic feature space.

[0032] Finally, the big data system semantically annotates the feature space members in the demand semantic feature space based on at least one of the conversation interest event information and keyword vector information. The demand semantic feature space is a predefined high-dimensional space used to describe various consulting needs that may be faced in enterprise operation services.

[0033] Each point or vector in the big data system represents a specific consulting need or intention. By mapping conversation interest events and keyword vectors into this space, the big data system can semantically annotate feature members in the space.

[0034] For example, if a conversation interest event is identified as "asking about how to use a product" and its keyword vector is highly correlated with "operation guide" and "usage steps", the big data system will find the corresponding point in the demand semantic feature space and mark it as "product usage consultation demand".

[0035] At the same time, the big data system can also cluster similar consulting needs together according to the similarity of keyword vectors to form different demand groups. In this way, enterprises can more intuitively understand the distribution and trend of customer consulting needs, and thus formulate more accurate service strategies.

[0036] It can be understood that in the process of discussing the in-depth analysis of business consulting logs by big data systems in enterprise operation services, several key concepts are involved, including frequent text paragraphs, conversation interest event information, keyword vector information, demand semantic feature space, feature space members, semantic annotation processing, and enterprise consulting demand semantics.

[0037] First, frequent text segments refer to those text segments that appear repeatedly in multiple consultation conversations. These text segments can represent common concerns of customers or common responses of customer service. For example, in the customer service department of a manufacturing company, text segments such as "operation guide", "troubleshooting" and "parts replacement" may appear frequently in multiple conversations, reflecting the common consultation needs of customers.

[0038] Secondly, the conversation interest event information is the key events and their attributes extracted from the consultation conversation text through natural language processing technology. These events reflect the specific interests or needs expressed by the customer in the conversation. For example, when a customer mentions "I recently encountered some problems when using your product, can you provide some technical support?", the big data system will identify the conversation interest event "expressing technical support needs" and extract the relevant time information and customer ID.

[0039] Then, the keyword vector information is converted into vector representations in a high-dimensional space using word embedding technology. These vectors capture the semantic similarities and differences between texts. For example, the two keywords "operation guide" and "use steps" may be very close in the vector space because they have similar semantics.

[0040] Next, the demand semantic feature space is a predefined high-dimensional space used to describe various consulting needs that may be faced in enterprise operation services. In this space, each point or vector represents a specific consulting need or intention. For example, one point in the space may represent "product usage consulting needs", while another point may represent "after-sales service consulting needs".

[0041] Feature space members refer to specific points or vectors in the demand semantic feature space, which represent different consulting needs or intentions. By mapping the conversation interest event information and keyword vector information, the big data system can convert this information into members in the feature space.

[0042] Semantic annotation processing is the process of mapping conversation interest event information and keyword vector information into the demand semantic feature space and annotating the members of the feature space. For example, if a conversation interest event is identified as "asking about product usage methods" and its keyword vector is highly correlated with "operation guide" and "usage steps", the system will find the corresponding point in the demand semantic feature space and annotate it as "product usage method consultation requirements".

[0043] Finally, the semantics of business consulting needs refer to the semantics of consulting needs obtained through semantic annotation processing and expressed in the semantic feature space of needs. These semantics reflect the specific needs and intentions expressed by customers in the consulting dialogue. For example, "product usage consultation needs" is a business consulting needs semantics, which represents the consulting needs of customers who want to know how to use the product.

[0044] It should be noted that the technical solution recorded in step 140 is the core invention of this application, which realizes the accurate capture and effective representation of the demand semantics generated by the enterprise-business interaction terminal in the business consultation process. Specifically, the big data system performs semantic annotation processing on the feature space members in the demand semantic feature space based on at least one category of the conversation interest event information or keyword vector information, thereby obtaining the enterprise-business consultation demand semantics of the enterprise-business interaction terminal in the demand semantic feature space.

[0045] First, the big data system will collect and process business consultation logs from enterprise and institution interaction terminals, and extract interest event information and keyword vector information in the conversation through advanced text parsing and natural language processing technology. This information is the key to understanding customer needs and optimizing service processes. Secondly, the big data system will build a high-dimensional demand semantic feature space, which is pre-defined and used to comprehensively describe and represent various consulting needs that may be faced in enterprise operation services. In this space, each point or vector represents a specific consulting need or intention, which constitutes a feature space member. Then, the big data system will map the extracted conversation interest event information and keyword vector information to this demand semantic feature space. This mapping process is based on the similarity and relevance of text semantics, which ensures that the specific needs and intentions expressed by customers in the consulting conversation can be accurately represented in the feature space. Finally, the big data system will perform semantic annotation processing on the feature space members mapped to the demand semantic feature space. This processing process is based on the semantic content of the conversation interest event information and keyword vector information, which enables each feature space member to be annotated with the corresponding enterprise and institution consulting demand semantics. For example, if a feature space member is highly correlated with “product usage method”, then it will be labeled as “product usage method consultation demand”.

[0046] In this way, the demand semantics generated by the interactive terminals of enterprises and institutions during the business consultation process are accurately captured and effectively represented, which provides strong data support for enterprises to optimize service processes and improve operational efficiency.

[0047] In more detail, the "demand semantic feature space" is a high-dimensional vector space that is used to comprehensively describe and represent various consulting needs that may be faced in enterprise operation services. Each dimension in this space represents a feature related to the consulting need, such as how to use the product, the need for technical support, and the request for after-sales service. These features together constitute a complex, multi-dimensional semantic space. In this space, "feature space members" are specific points or vectors that represent different consulting needs or intentions. Each feature space member has a corresponding feature vector, and the value of this vector in each dimension reflects the performance or strength of the member in different features.

[0048] Taking the customer service department of a manufacturing enterprise as an example, there is a three-dimensional demand semantic feature space, in which the three dimensions are "product usage consultation", "technical support requirements" and "after-sales service request". In this space, a specific feature space member (i.e., a consulting requirement) can have a feature vector as follows: [0.8, 0.4, 0.2]. This vector indicates that the strength of the consulting requirement in the dimension of "product usage consultation" is 0.8, the strength in the dimension of "technical support requirements" is 0.4, and the strength in the dimension of "after-sales service request" is 0.2.

[0049] Based on the above content, "corporate affairs consulting demand semantics" is obtained through semantic annotation processing and expressed in the demand semantic feature space. It is actually an interpretation or annotation of the feature space member (i.e., feature vector) to explain what kind of consulting demand or intention the member represents. In the above example, "corporate affairs consulting demand semantics" can be "mainly focusing on product usage methods, and also requiring certain technical support, but fewer after-sales service requests." Continuing with the customer service department of a manufacturing enterprise as an example, there is a three-dimensional demand semantic feature space, with the dimensions of "product usage consultation", "technical support requirements" and "after-sales service requests". In this space, a specific "corporate affairs consulting demand semantics" can be represented by a feature vector, such as [0.7, 0.5, 0.1]. Each element of the feature vector corresponds to a dimension in the demand semantic feature space, and its value reflects the intensity or importance of the consulting demand in this dimension. In this example, 0.7 indicates that the consulting demand has a high intensity in the dimension of "consultation on product usage methods", 0.5 indicates that it also has a certain intensity in the dimension of "technical support needs", and 0.1 indicates that the intensity in the dimension of "after-sales service request" is relatively low.

[0050] The present invention achieves effective screening and positioning of consulting dialogues by obtaining the business consulting log of the enterprise affairs interactive terminal and determining at least two target consulting dialogue texts therefrom. By parsing the dialogue interest events of each target consulting dialogue text, the dialogue interest event information of the enterprise affairs interactive terminal is obtained, thereby deeply understanding the user's concerns and needs in the consultation process. At the same time, frequent text paragraphs are extracted from the target consulting dialogue text, and keyword vector parsing is performed to further obtain the keyword vector information of the enterprise affairs interactive terminal, providing rich feature data for subsequent semantic analysis. Finally, according to at least one category of the dialogue interest event information or keyword vector information, semantic annotation processing is performed on the feature space members in the demand semantic feature space, and the enterprise affairs consulting demand semantics of the enterprise affairs interactive terminal in the demand semantic feature space are successfully obtained. In this way, not only the understanding accuracy of consulting needs is improved, but also strong data support is provided for the service optimization and process improvement of the enterprise, which greatly enhances the intelligence and personalization level of enterprise operation services.

[0051] In some optional embodiments, the method further includes: obtaining a consulting dialogue text example configured with a priori learning annotations of the enterprise affairs interaction terminal; using the consulting dialogue text example to perform algorithm optimization, and obtaining a dialogue interest event parsing algorithm for parsing dialogue interest events for the enterprise affairs interaction terminal. Based on this, for each of the target consulting dialogue texts, parsing the dialogue interest events for the enterprise affairs interaction terminal, and obtaining the dialogue interest event information of the enterprise affairs interaction terminal, includes: for each of the target consulting dialogue texts, parsing the dialogue interest events for the enterprise affairs interaction terminal according to the dialogue interest event parsing algorithm, and obtaining the dialogue interest event information of the enterprise affairs interaction terminal.

[0052] In order to further improve the conversation interest event parsing capability of the enterprise affairs interaction terminal, this application also covers the steps of obtaining consultation conversation text examples configured with prior learning annotations of the enterprise affairs interaction terminal and using these examples for algorithm optimization. The purpose of this process is to obtain a more accurate and efficient conversation interest event parsing algorithm to better understand and analyze the user's concerns and needs during the consultation process.

[0053] The consultation dialogue text examples with prior learning annotations are a set of dialogue texts that have been manually annotated and parsed, which contain various events of interest and concerns that users may express during the consultation process. These examples serve as training data to help the algorithm learn how to extract useful information from dialogue texts and perform accurate semantic annotations.

[0054] In practice, obtaining these examples of consulting dialogue texts with prior learning annotations can involve working closely with the customer service department of a manufacturing company. By collecting and analyzing a large amount of historical consulting dialogue data, representative dialogue texts can be selected and carefully annotated manually by a professional annotation team. These annotated data will serve as the basis for subsequent algorithm optimization.

[0055] Next, using these prior learning annotated consultation dialogue text examples for algorithm optimization is a key step. At this stage, various machine learning and deep learning techniques, such as neural networks and support vector machines, can be used to train and optimize the dialogue interest event parsing algorithm. By continuously iterating and adjusting the algorithm parameters, the algorithm can be made more accurate and efficient in parsing dialogue text.

[0056] Specifically, the algorithm optimization process may include the following steps: data preprocessing, feature extraction, model training, performance evaluation, and adjustment optimization. In the data preprocessing stage, the consulting dialogue text examples with prior learning annotations need to be cleaned and formatted for subsequent processing. In the feature extraction stage, useful feature information such as keywords, phrases, sentence structures, etc. need to be extracted from the dialogue text, which will be used as part of the algorithm input. In the model training stage, the extracted feature information and the corresponding annotated data are used to train the dialogue interest event parsing model so that it learns how to extract interest event information from the dialogue text. In the performance evaluation stage, the trained model needs to be performance tested and evaluated to ensure its accuracy and stability. Finally, in the adjustment optimization stage, the model is further adjusted and optimized based on the results of the performance test to improve its performance in practical applications.

[0057] After algorithm optimization, a conversation interest event parsing algorithm for business transaction interaction terminals can be obtained. This algorithm will serve as the core tool for subsequent processing of consultation conversation texts to extract and understand the user's concerns and needs during the consultation process.

[0058] Based on this, for each of the target consultation dialogue texts, the dialogue interest event parsing algorithm can be used to parse the dialogue interest event of the enterprise affairs interaction terminal, thereby obtaining the dialogue interest event information of the enterprise affairs interaction terminal. This process is automated and can greatly improve processing efficiency and accuracy. Specifically, for each target consultation dialogue text, the algorithm will conduct in-depth analysis and parsing, extract key information and interest events, and perform corresponding semantic annotations. This information will serve as the basis for subsequent processing and analysis, providing valuable insights and decision support for enterprises.

[0059] In practical applications, the performance of the conversation interest event parsing algorithm will directly affect the company's ability to understand and analyze user consultation needs. Therefore, various practical factors and challenges need to be fully considered during the algorithm design and optimization process. For example, the diversity and complexity of conversation texts, the ambiguity and uncertainty of user expressions, and the specific needs of different business scenarios. These factors need to be fully considered and addressed during the algorithm design and optimization process.

[0060] In addition, in order to further improve the accuracy and efficiency of the conversation interest event parsing algorithm, some advanced technologies and methods can also be considered. For example, natural language processing technology can be used to enhance the algorithm's ability to understand and analyze conversation texts; deep learning technology can be used to automatically extract high-level feature information from conversation texts; and transfer learning technology can be used to migrate and adapt the algorithm's application in different business scenarios.

[0061] It can be seen that by obtaining consulting dialogue text examples with prior learning annotations configured for enterprise affairs interaction terminals and using these examples for algorithm optimization, a more accurate and efficient conversation interest event parsing algorithm can be obtained. This algorithm will serve as a core tool for subsequent processing of consulting dialogue texts, providing enterprises with more accurate and comprehensive user consulting demand analysis and insights. This will not only help improve the customer service quality and operational efficiency of enterprises, but also provide strong data support and decision-making basis for enterprises to formulate more accurate marketing strategies and business development plans.

[0062] In some optional embodiments, the obtaining of a consulting conversation text example with a priori learning annotations configured for an enterprise affairs interaction terminal includes: obtaining a consulting conversation text example with portrait knowledge information and key entity semantics configured for the enterprise affairs interaction terminal; and processing the consulting conversation text example through an initial NLP algorithm based on the portrait knowledge information and the key entity semantics to obtain the priori learning annotations of the enterprise affairs interaction terminal.

[0063] It is worth mentioning that in order to obtain consulting dialogue text examples with prior learning annotations configured for the enterprise affairs interaction terminal, it is first necessary to clarify that these examples should contain rich portrait knowledge information and key entity semantics, so as to fully reflect the user's concerns and needs in the consulting process. Portrait knowledge information mainly refers to information about the user's basic attributes, behavioral characteristics, preferences and habits, while key entity semantics refers to entity information such as specific products, services, problems or needs mentioned by users in consulting dialogues.

[0064] It is understandable that obtaining examples of consulting dialogue texts with portrait knowledge information and key entity semantics configured with enterprise transaction interaction terminals is a key step. In actual operation, this may require close cooperation with the customer service department of manufacturing enterprises to achieve this by collecting and analyzing a large amount of historical consulting dialogue data. This historical data should contain the various needs and questions expressed by users during the consultation process, as well as the portrait knowledge information and key entity semantics related to them.

[0065] After collecting enough consulting dialogue text examples, the next step is to process these consulting dialogue text examples through the initial NLP algorithm based on the portrait knowledge information and key entity semantics to obtain the prior learning annotations of the enterprise transaction interaction terminal. NLP (natural language processing) algorithm is a technology that can understand and process human language. It can help big data systems extract useful information and knowledge from large amounts of text data.

[0066] At this stage, the main task of the initial NLP algorithm is to conduct in-depth semantic analysis and understanding of the consulting dialogue text examples. Specifically, the algorithm needs to be able to identify key entities in the text (such as product names, service types, problem descriptions, etc.) and understand the semantic relationships and contextual information of these entities in the text. At the same time, the algorithm also needs to classify and annotate the user's consulting behavior based on the profile knowledge information for subsequent learning and optimization.

[0067] For example, there is a sample of a consulting dialogue text about how to use a manufacturing company's product: "I recently purchased your company's CNC machine tool, but encountered some problems during use. How can I adjust the machine tool's cutting parameters to achieve better processing results?" In this example, "CNC machine tool" is a key entity, and "adjusting the machine tool's cutting parameters" and "obtaining better processing results" are the user's specific needs and questions. At the same time, based on the user's profile knowledge information, the big data system can know that this user may be an engineer or technician of a manufacturing company.

[0068] Through the processing of the initial NLP algorithm, the big data system can convert this consulting dialogue text example into a series of numerical feature vectors and perform semantic annotation. Specifically, the numerical feature vector may include the frequency of occurrence of key entities, the degree of matching of user portraits, the type of consulting questions, etc. For example, for the above example text, the frequency of occurrence of the key entity "CNC machine tool" may be 1, and the types of the two questions "adjusting the cutting parameters of the machine tool" and "obtaining better processing results" may be labeled as "product usage consultation". At the same time, based on the user's portrait knowledge information, the big data system can label this example text with a user portrait label of "engineer / technician".

[0069] After completing these processes, the big data system obtains consultation dialogue text examples with prior learning annotations configured for the business transaction interaction terminal. These examples will serve as the basic data for subsequent algorithm optimization and dialogue interest event analysis, helping the big data system to better understand and analyze the user's concerns and needs during the consultation process.

[0070] In practical applications, in order to further improve the accuracy and comprehensiveness of prior learning annotations, the big data system can also continuously iterate and optimize the initial NLP algorithm. For example, the big data system can use more historical consultation dialogue data to train the algorithm to improve its recognition of key entities and user portraits; the big data system can also introduce more semantic features and contextual information to enrich the numerical feature vector to improve the algorithm's understanding and annotation capabilities of consultation questions.

[0071] Therefore, by obtaining examples of consulting dialogue texts configured with the image knowledge information and key entity semantics of the enterprise affairs interaction terminal, and processing these examples through the initial NLP algorithm based on this information and semantics, the big data system can obtain the prior learning annotations of the enterprise affairs interaction terminal. These annotations will serve as an important basis for subsequent processing and analysis of consulting dialogue texts, providing enterprises with more accurate and comprehensive analysis and insights into user consulting needs.

[0072] In the next step, the conversation interest event is parsed on the enterprise affairs interaction terminal according to the conversation interest event parsing algorithm to obtain the conversation interest event information of the enterprise affairs interaction terminal, including: the conversation interest event is parsed on the enterprise affairs interaction terminal according to the conversation interest event parsing algorithm to obtain the business community relationship distribution information of the enterprise affairs interaction terminal and the push response status characteristics of each consultation push record of the business community; the semantic annotation processing is performed on the feature space members in the demand semantic feature space according to at least one category of the conversation interest event information or the keyword vector information to obtain the enterprise affairs consultation demand semantics of the enterprise affairs interaction terminal in the demand semantic feature space, including: the semantic annotation processing is performed on the feature space members in the demand semantic feature space according to the business community relationship distribution information, each of the push response status characteristics, and the keyword vector information to obtain the enterprise affairs consultation demand semantics of the enterprise affairs interaction terminal in the demand semantic feature space.

[0073] Based on this embodiment, according to the optimized conversation interest event parsing algorithm, the big data system will conduct in-depth conversation interest event parsing on the enterprise affairs interaction terminal, aiming to dig out valuable information hidden in the conversation text. This process not only involves understanding the content of the conversation, but also includes a detailed analysis of the distribution of business community relationships and the response status of consultation push records.

[0074] First, the big data system processes the conversation data of the enterprise transaction interaction terminal through the conversation interest event parsing algorithm to obtain the business community relationship distribution information. In the application scenario of manufacturing enterprises, the business community relationship distribution information reveals the interaction mode and degree of association between different business departments or product communities. For example, the algorithm may reveal that there is a high interaction frequency between the CNC machine tool user community and the tool accessories user community, which indicates that the two are closely related in business. This relationship distribution information is expressed in the form of a numerical feature vector, such as (CNC machine tool community, tool accessories community, interaction frequency: 0.7), etc., which provides the big data system with a quantitative description of the interaction between business communities.

[0075] At the same time, the algorithm also analyzes the push response status characteristics of each consultation push record in the business community. This feature reflects the user's feedback after the consultation information is pushed, such as whether it is read, whether it is responded, and the content of the response. These response statuses are recorded in numerical form, for example (push record ID1, reading rate: 0.8, response rate: 0.6, average response time: 5 minutes), providing the big data system with direct evidence of the user's response to the consultation push.

[0076] After obtaining the business community relationship distribution information and push response status characteristics, the big data system further uses this information, combined with the previously extracted keyword vector information, to perform semantic annotation processing on the feature space members in the demand semantic feature space. The demand semantic feature space is an abstract concept that includes all possible consulting demand points, and each feature space member represents a specific demand point or problem.

[0077] In the semantic annotation process, the big data system first analyzes the distribution information of business community relationships to determine which communities or business departments have the most prominent demand points in the current conversation. Then, the big data system combines the push response status characteristics to evaluate the activity of these demand points in the actual consultation process and the user's attention. Finally, the big data system uses keyword vector information to match specific words in the conversation text with members in the demand semantic feature space to further refine the semantic content of the demand points.

[0078] Through this series of processing steps, the big data system can obtain the semantics of business consultation needs in the semantic feature space of business interaction terminals. The result is presented in the form of numerical feature vectors, such as (demand point 1: CNC machine tool operation training, attention: 0.9, activity: 0.8, keyword vector: [CNC machine tool, operation, training]). These values ​​not only quantify the importance and activity of different demand points, but also provide the specific semantic content of the demand points through keyword vectors.

[0079] In this way, the process of parsing conversation interest events for business affairs interaction terminals based on the conversation interest event parsing algorithm is an in-depth and comprehensive data mining process. It not only reveals the interaction patterns and correlation between business communities, but also evaluates the effect of consultation push and user attention. Finally, by combining the business community relationship distribution information, push response status characteristics and keyword vector information, the big data system can accurately mark the business affairs consultation demand semantics of the business affairs interaction terminal in the demand semantic feature space. This result provides strong data support and decision-making basis for the service optimization, process improvement and personalized service of enterprises.

[0080] In some other possible embodiments, the step of obtaining the push response status characteristics of each consultation push record in the business community includes: parsing the conversation interest events of the enterprise business interaction terminal according to the conversation interest event parsing algorithm, obtaining the consultation push attention characteristics of each consultation push record in the business community of the enterprise business interaction terminal, and the distribution characteristics of the involvement portraits of each consultation push record; the involvement portraits of any of the consultation push records include the target portrait in the consultation push record, the first annotated portrait on the first consultation semantic relationship network constituting the consultation push record, and the second annotated portrait on the second consultation semantic relationship network constituting the consultation push record; for each of the consultation push records, determining the interactive mapping relationship network of the consultation push record according to the distribution characteristics of each involvement portrait of the consultation push record; based on the consultation push attention characteristics and the interactive mapping relationship network, obtaining the push response status characteristics of the consultation push record.

[0081] In actual application, in order to obtain the push response status characteristics of each consultation push record of the business community, the big data system first needs to deeply understand the content covered by this feature and its value in the enterprise application scenario. The push response status feature not only reflects the direct feedback of users after the consultation information is pushed, such as whether it is read or responded, but also implies the degree of match between the user portrait and the push content and the interaction mode between user communities. In the context of manufacturing enterprises, this feature is of great significance for optimizing consulting services, improving user satisfaction and guiding product improvements.

[0082] First, the big data system uses the conversation interest event parsing algorithm to deeply analyze the conversation data of the enterprise affairs interaction terminal, aiming to dig out the consultation push attention characteristics of each consultation push record in the business community. This feature quantifies the user's attention to the push consultation, which can be measured by specific behavioral indicators such as the user's click rate, reading time, and scrolling depth after the push. For example, a push record about a new function of a CNC machine tool may have an 80% click rate and an average reading time of 5 minutes. These data constitute the specific value of the consultation push attention feature of the push record.

[0083] At the same time, the big data system also obtains the distribution characteristics of the involved portraits of each consultation push record. The involved portrait is a composite concept, which includes the target portrait in the consultation push record, that is, the user group portrait targeted by the push content; the first annotated portrait on the first consultation semantic relationship network, which reflects the semantic association between the push content and the user's historical consultation or behavior; and the second annotated portrait on the second consultation semantic relationship network, which further reveals the potential connection between the push content and a wider user community or business scenario. These portraits exist in the form of numerical feature vectors, such as (target portrait: CNC machine tool users, similarity: 0.9), (first annotated portrait: users who have recently consulted CNC machine tools, similarity: 0.7), (second annotated portrait: users interested in improving production efficiency, similarity: 0.6).

[0084] Next, for each consultation push record, the big data system determines the interactive mapping network of the consultation push record based on the distribution characteristics of the portraits involved. This network reveals the interaction patterns between the push records and different user portraits, and how these interactions affect the push effect. For example, the big data system may find that the push for CNC machine tool users has a higher response rate among users who have recently consulted about CNC machine tools, and a lower response rate among users who are interested in improving production efficiency. These findings are recorded in the interactive mapping network in numerical form, such as (push record ID, target portrait, response rate: 0.8), (push record ID, first annotated portrait, response rate: 0.9), etc.

[0085] Finally, based on the attention features of the consultation push and the interactive mapping network, the big data system comprehensively evaluates the push response status features of the consultation push record. This feature not only includes direct feedback indicators such as whether the user reads and responds, but also incorporates deep information such as the matching degree between the user portrait and the push content, and the interactive effect between user communities. For example, a push record may have a high reading rate (0.8) and a low response rate (0.3), but through the interactive mapping network big data system, it is found that this push actually has a high potential value in a specific user portrait group (similarity: 0.9, expected response rate: 0.7). Such a comprehensive evaluation provides a more comprehensive push effect analysis for the big data system, and also provides strong data support for the subsequent optimization of consulting services.

[0086] Therefore, in the process of obtaining the push response status characteristics of each consultation push record of the business community, the big data system deeply explores the complex relationship between user behavior, user portraits and push content through the conversation interest event parsing algorithm. This process not only quantifies the user's attention to push consultation, but also reveals the interaction pattern and potential value between push content and different user groups. In the end, the big data system obtains a push response status feature containing multi-level and multi-dimensional information, which provides strong data support and decision-making basis for the optimization of corporate consulting services, improvement of user satisfaction and product improvement.

[0087] In a further embodiment, the push response status characteristics of the consultation push record are obtained based on the consultation push attention characteristics and the interactive mapping relationship network, including: obtaining a first influence weight of the consultation push attention characteristics, and a second influence weight of the interactive mapping relationship network; based on the first influence weight and the second influence weight, feature integration of the consultation push attention characteristics and the interactive mapping relationship network is performed to obtain the push response status characteristics of the consultation push record.

[0088] It is understandable that in order to obtain the push response status characteristics of the consultation push record, the big data system needs to comprehensively consider the information of the consultation push attention characteristics and the interactive mapping relationship network, and integrate the characteristics according to their respective influence weights. This process aims to comprehensively evaluate the effect of the consultation push, while considering the user's attention to the push content and the interaction mode between the push content and different user portraits.

[0089] First, the big data system obtains the first impact weight of the consultation push attention feature. This weight reflects the importance of the consultation push attention feature in evaluating the push response status. In the application scenarios of manufacturing enterprises, user attention is one of the important indicators for measuring the effectiveness of push. For example, if a push record about a new function of a CNC machine tool receives a high click-through rate and reading time, then this push record is likely to have a high degree of attention in the user's mind, thereby having a positive impact on the response status. Therefore, the big data system sets the first impact weight of the consultation push attention feature to a relatively high value, such as 0.6, to reflect its importance.

[0090] At the same time, the big data system also needs to obtain the second influence weight of the interaction mapping network. This weight represents the contribution of the interaction mapping network in evaluating the push response status. The interaction mapping network reveals the interaction patterns and potential values ​​between the push content and different user portraits, which is also crucial to understanding the push effect. For example, even if a push record has a low click-through rate, if it has high potential value in a specific user portrait group, then the push may still be considered effective. Therefore, the big data system also sets the second influence weight of the interaction mapping network to a relatively high value, such as 0.4, to ensure its appropriate contribution in feature integration.

[0091] Next, the big data system integrates the features of the consultation push attention feature and the interactive mapping relationship network based on the first influence weight and the second influence weight. This process aims to fuse information from two different dimensions into a comprehensive push response status feature. Specifically, the big data system multiplies the value of the consultation push attention feature by the first influence weight to obtain the weighted attention feature; similarly, the big data system multiplies the value in the interactive mapping relationship network by the second influence weight to obtain the weighted interaction feature. Then, the big data system adds these two weighted features to obtain the final push response status feature.

[0092] Taking specific values ​​as an example, the attention feature of a consultation push record is (click rate: 0.8, reading time: 5 minutes), and the value in the interactive mapping relationship network is (target portrait similarity: 0.9, expected response rate: 0.7). Then, the weighted attention feature is (click rate: 0.8*0.6=0.48, reading time: 5*0.6=3 minutes), and the weighted interaction feature is (target portrait similarity: 0.9*0.4=0.36, expected response rate: 0.7*0.4=0.28). Finally, the big data system adds these two weighted features to obtain the push response status feature (comprehensive attention: 0.48+0.36=0.84, comprehensive response expectation: 3 minutes+0.28=3.28 minutes).

[0093] Through feature integration, the big data system obtains a push response status feature that contains multi-dimensional information. This feature not only takes into account the user's attention to the pushed content, but also incorporates the interaction mode and potential value between the pushed content and different user portraits. Such a comprehensive evaluation provides the big data system with a more comprehensive analysis of the push effect, and also provides strong data support for the subsequent optimization of consulting services.

[0094] In the process of obtaining the push response status characteristics of the consultation push records based on the consultation push attention characteristics and the interactive mapping relationship network, the big data system comprehensively considers the information of two different dimensions and integrates the characteristics according to their respective influence weights. This process not only quantifies the user's attention to the push content, but also reveals the interaction patterns and potential value between the push content and different user groups. In the end, the big data system obtains a push response status feature containing multi-level and multi-dimensional information, which provides strong data support and decision-making basis for the optimization of corporate consulting services, improvement of user satisfaction and product improvement.

[0095] In some further optional embodiments, the consultation push attention feature and the interactive mapping relationship network are feature integrated based on the first influence weight and the second influence weight to obtain the push response status feature of the consultation push record, including: obtaining an influence weight comparison result through the first influence weight and the second influence weight; if the influence weight comparison result meets the comparison requirements, semantically reconstructing and integrating the consultation push attention feature and the interactive mapping relationship network based on a structured semantic deduction strategy to obtain the push response status feature of the consultation push record; if the influence weight comparison result does not meet the comparison requirements, determining the consultation feedback text semantics corresponding to the selected influence weight between the first influence weight and the second influence weight as the push response status feature of the consultation push record; the consultation feedback text semantics include the consultation push attention feature and the interactive mapping relationship network.

[0096] In practical applications, in order to obtain the push response status characteristics of the consultation push records, the big data system needs to integrate the consultation push attention characteristics and the interactive mapping relationship network based on the first influence weight and the second influence weight. This process involves comprehensive evaluation of information in two different dimensions and selecting an appropriate integration strategy based on the evaluation results.

[0097] First, the big data system obtains the impact weight comparison result through the first impact weight and the second impact weight. This step aims to compare the sizes of the two weights to determine their relative importance in feature integration. In actual applications of manufacturing enterprises, these weights may be set based on historical data, user behavior analysis, or business strategies. For example, if historical data shows that the consultation push attention feature has a higher accuracy in predicting the user's response status, then the first impact weight can be set to a relatively high value, such as 0.65, and the second impact weight can be set to 0.35.

[0098] Next, the big data system selects different feature integration strategies based on the impact weight comparison results. If the impact weight comparison results meet the comparison requirements, that is, the difference between the first impact weight and the second impact weight is within the predetermined threshold range, then the big data system will semantically reconstruct and integrate the consultation push attention features and the interactive mapping relationship network based on the structured semantic deduction strategy. This strategy aims to recombine the information in the two feature vectors in a structured way to form a more comprehensive push response status feature.

[0099] Specifically, the structured semantic deduction strategy can involve the following steps: First, convert the numerical feature vectors in the consultation push attention features and the interactive mapping relationship network into semantic representations, such as converting the click rate into the semantic labels of "high attention" or "low attention". Then, according to the logical relationships between these semantic representations, such as causal relationships, correlation relationships, or sequential relationships, they are combined into a structured semantic representation. Finally, this structured semantic representation is converted back into a numerical feature vector to form the push response status feature.

[0100] For example, the semantic representation of the attention feature of the consultation push is (attention: high, reading time: long), and the semantic representation of the interactive mapping relationship network is (target portrait similarity: high, expected response rate: medium). Based on the structured semantic deduction strategy, the big data system may combine these semantic representations into a structured semantic representation, such as (users with high attention and long reading time have a medium expected response rate to push of high similarity target portraits). Then, the big data system converts this structured semantic representation back into a numerical feature vector, such as (comprehensive attention: 0.85, comprehensive response expectation: 0.6).

[0101] If the impact weight comparison result does not meet the comparison requirements, that is, the difference between the first impact weight and the second impact weight exceeds the predetermined threshold range, then the big data system will select another feature integration strategy. In this case, the big data system determines the consultation feedback text semantics corresponding to the selected impact weight in the first impact weight and the second impact weight as the push response status feature of the consultation push record. This strategy aims to highlight the role of more important feature dimensions in the push response status evaluation.

[0102] Specifically, if the first influence weight is greater than the second influence weight, and the difference between them exceeds the predetermined threshold range, the big data system will select the consultation feedback text semantics corresponding to the consultation push attention feature as the push response state feature. This means that when evaluating the push effect, the big data system will pay more attention to the user's attention to the pushed content. On the contrary, if the second influence weight is greater than the first influence weight, and the difference between them exceeds the predetermined threshold range, the big data system will select the consultation feedback text semantics corresponding to the interactive mapping relationship network as the push response state feature. This means that when evaluating the push effect, the big data system will pay more attention to the interaction pattern and potential value between the pushed content and different user portraits.

[0103] For another example, the first impact weight is 0.7, the second impact weight is 0.3, and the difference between them is within the predetermined threshold range. Then the big data system will semantically reconstruct and integrate the consultation push attention features and the interactive mapping relationship network based on the structured semantic deduction strategy. However, if the first impact weight is 0.8, the second impact weight is 0.2, and the difference between them exceeds the predetermined threshold range, then the big data system will select the consultation feedback text semantics corresponding to the consultation push attention feature as the push response status feature.

[0104] In the process of integrating the attention features and interactive mapping relationship network of consultation push according to the first influence weight and the second influence weight, the big data system first obtains the influence weight comparison result by comparing the sizes of the two weights. Then, according to this comparison result, an appropriate feature integration strategy is selected: if it meets the comparison requirements, semantic reconstruction integration is performed based on the structured semantic deduction strategy; if it does not meet the comparison requirements, the consultation feedback text semantics corresponding to the selected influence weight is selected as the push response status feature. Such a feature integration process not only takes into account the user's attention to the pushed content, but also incorporates the interaction mode and potential value between the pushed content and different user portraits, thereby providing enterprises with a more comprehensive push effect analysis and decision-making basis.

[0105] Under some optional design ideas, the consultation push attention feature and the interactive mapping relationship network are feature integrated based on the first influence weight and the second influence weight to obtain the push response status feature of the consultation push record, including: determining the characteristic thermal values ​​of the consultation push attention feature and the interactive mapping relationship network based on the first influence weight and the second influence weight; there is a quantitative relationship between the characteristic thermal value and the corresponding influence weight; the entity semantic encoding strategy of the consultation push attention feature and the interactive mapping relationship network is the same; based on each of the characteristic thermal values, the consultation push attention feature and the interactive mapping relationship network are processed to obtain the push response status feature of the consultation push record.

[0106] In order to obtain the push response status characteristics of the consultation push record, the big data system needs to integrate the consultation push attention characteristics and the interactive mapping relationship network according to the first influence weight and the second influence weight. This process involves comprehensive evaluation of information in two different dimensions and forms a comprehensive push response status characteristic based on the evaluation results.

[0107] First, the big data system determines the characteristic thermal values ​​of the consultation push attention features and the interactive mapping relationship network based on the first influence weight and the second influence weight. The characteristic thermal value is a quantitative indicator used to indicate the importance of the feature in the push response status evaluation. There is a quantitative relationship between it and the corresponding influence weight, that is, the larger the influence weight, the higher the characteristic thermal value. In the actual application of manufacturing enterprises, these characteristic thermal values ​​may be set based on historical data, user behavior analysis or business strategies.

[0108] For example, the first impact weight of the consultation push attention feature is 0.6, and the second impact weight of the interactive mapping relationship network is 0.4. The big data system can convert these weights into feature thermal values, for example, converting the first impact weight of 0.6 into a feature thermal value of 60, and converting the second impact weight of 0.4 into a feature thermal value of 40. In this way, the consultation push attention feature and the interactive mapping relationship network each have a quantified feature thermal value to indicate their importance in the push response status evaluation.

[0109] Secondly, the big data system needs to ensure that the entity semantic encoding strategy of the consultation push attention features and the interactive mapping relationship network is the same. This is to ensure that the two feature vectors can be effectively semantically matched and fused during the feature integration process. The entity semantic encoding strategy is to convert the entities in the feature vector (such as users, products, push content, etc.) into a unified semantic representation. In the application scenario of manufacturing enterprises, this can involve converting user portraits, product attributes, push content, etc. into the same semantic encoding format for subsequent feature integration processing.

[0110] For another example, the big data system has defined a unified semantic encoding strategy for the entities in the consultation push attention features and interactive mapping relationship network. For example, user portraits may be encoded in the form of "age_gender_occupation", product attributes may be encoded in the form of "type_specification_function", and push content may be encoded in the form of "topic_description_keyword". In this way, the big data system can ensure that during the feature integration process, the entities in the two feature vectors can be effectively semantically matched and fused.

[0111] Then, based on the characteristic heat value, the big data system processes the consultation push attention features and the interactive mapping relationship network to obtain the push response status features of the consultation push record. This processing process aims to integrate the information in the two feature vectors into a comprehensive push response status feature in a quantitative manner. Specifically, the big data system can perform a weighted summation of the two feature vectors according to the characteristic heat value to obtain a comprehensive push response status feature vector.

[0112] For example, the characteristic vector of the consultation push attention is (attention: 0.8, reading time: 5 minutes), and the characteristic vector of the interactive mapping relationship network is (target portrait similarity: 0.9, expected response rate: 0.7). The big data system can perform a weighted sum of these two characteristic vectors based on the previously determined characteristic heat values ​​of 60 and 40. Specifically, the big data system can multiply the attention by the characteristic heat value of 60 to obtain a weighted attention of 48; multiply the reading time by the characteristic heat value of 60 to obtain a weighted reading time of 300 minutes; multiply the target portrait similarity by the characteristic heat value of 40 to obtain a weighted target portrait similarity of 36; and multiply the expected response rate by the characteristic heat value of 40 to obtain a weighted expected response rate of 28. Then, the big data system combines these weighted values ​​into a comprehensive push response state characteristic vector (weighted attention: 48, weighted reading time: 300 minutes, weighted target portrait similarity: 36, weighted expected response rate: 28).

[0113] Through this feature integration process, the big data system obtains a push response status feature vector containing multi-dimensional information. This vector not only takes into account the user's attention to the pushed content, but also incorporates the interaction mode and potential value between the pushed content and different user portraits. At the same time, through the quantitative processing of feature thermal values ​​and the unification of entity semantic encoding strategies, the big data system ensures the effectiveness and accuracy of the feature integration process.

[0114] Finally, the big data system can evaluate the effect of consulting push based on this comprehensive push response status feature vector, and provide strong data support and decision-making basis for the optimization of consulting services, improvement of user satisfaction and product improvement of enterprises. For example, the big data system can evaluate the user's interest in the pushed content based on the weighted attention and weighted reading time in the push response status feature vector; and evaluate the match between the pushed content and the user portrait and the expected response effect based on the weighted target portrait similarity and weighted expected response rate. Through these evaluation indicators, the big data system can better understand the effect of consulting push and provide strong support for the company's subsequent decision-making.

[0115] In this way, the features of the consultation push attention features and the interactive mapping relationship network are integrated according to the first influence weight and the second influence weight, and the push response status features of the consultation push records are obtained. This process involves determining the feature thermal value, unifying the entity semantic encoding strategy, and performing feature integration processing based on the feature thermal value. In this way, the big data system can effectively integrate information from two different dimensions into a comprehensive push response status feature, providing strong data support for the optimization and decision-making of corporate consulting services.

[0116] In some examples, based on at least one of the conversation interest event information or the keyword vector information, semantic annotation processing is performed on feature space members in the demand semantic feature space to obtain the business consultation demand semantics of the business affairs interaction terminal in the demand semantic feature space, including: using the conversation interest event information and the keyword vector information to obtain the global consultation session semantics of the business affairs interaction terminal; based on the global consultation session semantics of the business affairs interaction terminal, semantic annotation processing is performed on feature space members in the demand semantic feature space to obtain the business consultation demand semantics of the business affairs interaction terminal in the demand semantic feature space.

[0117] Furthermore, in order to obtain the semantics of business consultation needs in the demand semantic feature space of the business affairs interaction terminal, the big data system needs to semantically annotate the feature space members in the demand semantic feature space based on at least one of the conversation interest event information or keyword vector information. This process involves in-depth analysis of the consultation conversations of the business affairs interaction terminal, extracting key information therein, and matching and annotating it with the feature space members in the demand semantic feature space.

[0118] First, the big data system uses the conversation interest event information and keyword vector information to obtain the global consultation conversation semantics of the enterprise transaction interaction terminal. This step aims to extract key information and topics from the consultation conversation of the enterprise transaction interaction terminal to form a comprehensive conversation semantic representation. In the actual application of manufacturing enterprises, the conversation interest event information may include product types, production processes, equipment failures, etc. mentioned by users, while the keyword vector information may be key words extracted from the consultation conversation through natural language processing technology, such as "production efficiency", "quality control", "equipment maintenance", etc.

[0119] For example, the big data system extracts the following conversation interest event information and keyword vector information from a consultation session: the conversation interest event information is "the user mentioned the equipment failure problem on the production line", and the keyword vector information is (production efficiency: 0.3, quality control: 0.5, equipment maintenance: 0.7). The big data system can combine this information to form a global consultation session semantics, that is, "the user is concerned about the equipment failure problem on the production line, especially the equipment maintenance aspect."

[0120] Secondly, the big data system semantically annotates the feature space members in the demand semantic feature space based on the global consultation session semantics of the enterprise transaction interaction terminal. The demand semantic feature space is a space containing multiple feature space members, each of which represents a specific demand or problem. In the application scenarios of manufacturing enterprises, these feature space members may include "improving production efficiency", "optimizing quality control processes", "solving equipment failure problems", etc.

[0121] Semantic annotation processing is a process of matching and annotating the global consultation session semantics with the feature space members in the demand semantic feature space. Specifically, the big data system analyzes the key information and topics in the global consultation session semantics, and then compares and matches them with the feature space members in the demand semantic feature space. If the information in the global consultation session semantics is highly correlated with a feature space member, the big data system annotates this feature space member as a demand or question related to the current consultation session.

[0122] For example, the demand semantic feature space contains the following feature space members: "Improve production efficiency" (relevance threshold: 0.4), "Optimize quality control process" (relevance threshold: 0.5), and "Solve equipment failure problems" (relevance threshold: 0.6). The big data system compares and matches the previously obtained global consultation session semantics "Users are concerned about equipment failure problems on the production line, especially equipment maintenance" with these feature space members. By calculating the relevance, the big data system finds that "Solve equipment failure problems" has the highest relevance to the global consultation session semantics, reaching 0.7, which exceeds its relevance threshold of 0.6. Therefore, the big data system marks "Solve equipment failure problems" as a demand or problem related to the current consultation session.

[0123] Finally, the big data system obtains the semantics of business consultation needs of the business transaction interaction terminal in the semantic feature space of needs. This result is a comprehensive annotation result, which reflects the needs and problems that the business transaction interaction terminal is concerned about in the current consultation session. In the actual application of manufacturing enterprises, this result can be used to guide subsequent consulting services, product improvements or business optimization. For example, if the annotation results show that the user is mainly concerned about "solving equipment failure problems", then the enterprise can provide corresponding solutions or services for this demand to improve user satisfaction and production efficiency.

[0124] With this design, the big data system semantically annotates the feature space members in the demand semantic feature space based on at least one category of conversation interest event information or keyword vector information, and obtains the business consultation demand semantics of the business interaction terminal in the demand semantic feature space. This process involves extracting the global consultation session semantics, performing semantic annotation processing, and obtaining comprehensive annotation results. Through such a technical solution, the big data system can more accurately understand the consulting needs of the business interaction terminal and provide strong support for the optimization of consulting services and business decisions of enterprises.

[0125] In other examples, the frequent text paragraphs include business launch transformation consultation texts; the keyword vector information includes transformation mode features; the extracting of frequent text paragraphs of the enterprise business interaction terminal from at least part of the target consultation dialogue text, performing keyword vector analysis on the frequent text paragraphs, and obtaining keyword vector information of the enterprise business interaction terminal includes: when the dialogue interest event information includes a transformation consultation tag, determining the business launch transformation consultation text cluster of the enterprise business interaction terminal in the target consultation dialogue text according to the transformation consultation tag; the business launch transformation consultation text cluster includes local business transformation consultation text and remote business transformation consultation text; determining the transformation consultation preference of the enterprise business interaction terminal according to the text semantic distance between the local business transformation consultation text and the remote business transformation consultation text; the transformation consultation preference includes non-linked operation authentication preference and linked operation authentication preference; parsing the enterprise business interaction terminal according to the transformation consultation semantic parsing algorithm corresponding to the transformation consultation preference to obtain the transformation mode features of the enterprise business interaction terminal; the transformation consultation semantic parsing algorithm includes a linked operation authentication preference parsing algorithm and a non-linked operation authentication preference parsing algorithm.

[0126] This embodiment will explore in depth how to extract frequent text paragraphs from the consulting dialogue text of the enterprise affairs interaction terminal, and through keyword vector parsing technology, reveal the characteristics of the enterprise's transformation model, providing strong support for the digital transformation of manufacturing enterprises.

[0127] First, under the premise that the conversation interest event information contains the transformation consultation tag, the big data system focuses on the target consultation conversation text, which is the first step to mine the enterprise's transformation intention. In these texts, the frequently appearing paragraphs often reflect the key concerns and questions of the enterprise in the transformation process, which is the key to understanding the enterprise's transformation needs. In particular, the business launch transformation consultation text is directly related to how the enterprise adjusts its business model to adapt to the online environment, so it becomes the focus of the big data system analysis.

[0128] The business launch transformation consulting text cluster, as a type of frequent text paragraph, is further subdivided into local business transformation consulting text and remote business transformation consulting text. This classification is based on the regional characteristics of the business operations of enterprises, reflecting the differentiated needs of enterprises in transformation strategies under different market environments. For example, local business may focus more on supply chain optimization and customer service experience improvement, while remote business may focus more on market expansion and cross-cultural communication.

[0129] Next, the big data system quantifies the degree of preference of enterprises for different transformation paths by calculating the text semantic distance between local business transformation consulting texts and remote business transformation consulting texts. Text semantic distance is a numerical indicator that measures the similarity or difference of text content by calculating the cosine similarity or Euclidean distance between vectors based on the keyword vector representation in the text. A smaller semantic distance means a high correlation of the text content, reflecting the specific preference of the enterprise in transformation consulting.

[0130] Transformation consulting preference, as the core output of this analysis, is specifically divided into non-linked operation certification preference and linked operation certification preference. Non-linked operation certification preference refers to the tendency of enterprises to independently perform various operations during the transformation process, such as optimizing production processes and upgrading customer service systems independently, without emphasizing close coordination between various business links. On the contrary, linked operation certification preference indicates that enterprises pay more attention to cross-departmental and cross-regional collaboration during the transformation process, and pursue seamless connection between systems and maximization of overall efficiency.

[0131] In order to reveal the characteristics of the enterprise's transformation model in more depth, the big data system selects the corresponding transformation consulting semantic parsing algorithm to parse the enterprise transaction interaction terminal according to the transformation consulting preference. The non-linked operation authentication preference parsing algorithm focuses on extracting keywords and phrases about independent operations and local optimization in the text, and constructing a numerical feature vector that reflects the local transformation strategy of the enterprise. For example, for local business transformation consulting texts, the algorithm may identify keywords such as "production line automation transformation" and "customer service system upgrade" and convert them into vector form, such as [0.7, 0.3, 0.05], indicating that the importance of production line automation transformation is higher than other transformation measures.

[0132] The linkage operation authentication preference parsing algorithm focuses on analyzing the content of cross-departmental and cross-regional collaboration in the text, and constructs a numerical feature vector that reflects the overall transformation strategy of the enterprise. For example, in the text of remote business transformation consultation, the algorithm may capture keywords such as "global supply chain integration" and "multilingual customer service support" and convert them into vectors [0.4, 0.6, 0.1], indicating that the enterprise focuses on cross-regional market expansion and service capability improvement during the transformation.

[0133] Finally, through the comprehensive analysis of these keyword vectors, the big data system can draw a detailed transformation model feature portrait for the enterprise. This portrait not only shows the specific strategic choices of the enterprise in the transformation process, but also implies the enterprise's judgment and response to future market trends. For manufacturing companies, such analysis undoubtedly provides them with valuable decision-making support, helping them to accurately position themselves in the wave of digital transformation and move forward steadily.

[0134] Extracting frequent text paragraphs from the consulting dialogue text of the enterprise affairs interaction terminal and obtaining the characteristics of the enterprise's transformation model through keyword vector analysis technology is a complex process that combines natural language processing, machine learning and data analysis. It not only requires the big data system to accurately identify and process large amounts of text data, but also requires the big data system to deeply understand the business logic and transformation needs of manufacturing enterprises. Through such technical means, the big data system can tailor transformation strategies for enterprises and help them take a solid step on the road to digital transformation.

[0135] In some other examples, the transformation consulting tag includes a business process transformation tag and an authority authentication transformation tag; determining the business launch transformation consulting text cluster of the enterprise business interaction terminal in the target consulting dialogue text based on the transformation consulting tag includes: determining the positioning distribution information of the business launch transformation consulting text cluster of the enterprise business interaction terminal in the target consulting dialogue text based on the business process transformation tag or the authority authentication transformation tag; determining the text content association range of the business launch transformation consulting text cluster based on the correlation between the business process transformation tag and the authority authentication transformation tag.

[0136] It can be understood that this embodiment involves a specific implementation method of how to determine the business launch transformation consulting text cluster from these consulting dialogue texts based on the transformation consulting tag.

[0137] First, the transformation consulting label mainly includes the business process transformation label and the authority authentication transformation label. These two labels represent the two core aspects that enterprises pay attention to during the transformation process: one is the optimization and adjustment of business processes, and the other is the upgrade and innovation of the authority authentication mechanism.

[0138] Next, the big data system determines the location distribution information of the business online transformation consulting text cluster of the target consulting dialogue text based on the business process transformation label or the authority authentication transformation label. This process is actually a text classification task. The big data system needs to match the consulting dialogue text with the preset transformation label to identify the text paragraphs related to business transformation and authority authentication transformation. To achieve this goal, the big data system can use classification algorithms in machine learning, such as support vector machines (SVM) or naive Bayes classifiers. These algorithms can accurately classify text into corresponding labels based on keywords, phrases, and semantic relationships between them in the text.

[0139] After determining the location distribution information of the business launch transformation consulting text cluster, the big data system further determines the text content association range of the business launch transformation consulting text cluster based on the correlation between the business process transformation label and the authority authentication transformation label. The purpose of this step is to explore the intrinsic connection and mutual influence between texts under different transformation labels. To achieve this goal, the big data system can construct a text relationship network in which nodes represent different text paragraphs and edges represent the degree of association between texts. The degree of association can be calculated based on numerical features such as cosine similarity and co-occurrence frequency between texts.

[0140] Specifically, the big data system can first calculate the cosine similarity between each pair of text paragraphs, which is a value between 0 and 1 that indicates the degree of similarity between the texts. Then, the big data system can selectively build edges in the text relationship network based on the magnitude of the similarity. For example, the big data system can set a threshold, and only when the similarity between the texts exceeds this threshold, will an edge be built between them. In this way, the big data system can obtain a relationship network consisting of closely related text paragraphs.

[0141] In this textual relationship network, the big data system can further analyze the association patterns between texts under different transformation labels. For example, the big data system can find that some text paragraphs are closely related to both the business process transformation and the authority authentication transformation labels, which indicates that these texts not only discuss business transformation but also involve the issue of authority authentication. This cross-label association information is crucial for a comprehensive understanding of the transformation needs of enterprises.

[0142] In order to further enrich the semantic content of the business launch transformation consulting text cluster, the big data system can also use word vector embedding technology. Word vector embedding is a method of converting words or phrases in a text into a high-dimensional vector representation, which can capture the semantic relationship between words. Through word vector embedding, the big data system can convert text paragraphs into a collection of vectors and then calculate the semantic similarity between texts. This vector-based similarity calculation is more accurate and comprehensive than the traditional keyword-based matching method.

[0143] After obtaining the location distribution information and text content correlation range of the business launch transformation consulting text cluster, the big data system can further construct a characteristic portrait of the enterprise's transformation model. This portrait is a multi-dimensional numerical vector, in which each dimension represents a specific transformation feature or attribute. For example, the big data system can set a dimension to represent the degree of importance the enterprise attaches to business process optimization, represented by a value between 0 and 1; another dimension to represent the degree of investment the enterprise has made in upgrading the authority authentication mechanism, also represented by a value between 0 and 1. By filling in the values ​​on these dimensions, the big data system can obtain a characteristic vector that fully describes the enterprise's transformation model.

[0144] Finally, based on this transformation model feature portrait, the big data system can provide enterprises with more accurate and personalized transformation strategy suggestions. For example, if the portrait shows that the enterprise attaches great importance to business process optimization, but invests insufficiently in upgrading the authority authentication mechanism, then the big data system can suggest that the enterprise strengthen the innovation and upgrading of the authority authentication mechanism while maintaining business process optimization to achieve a more comprehensive transformation.

[0145] In some independent embodiments, determining the transformation consulting preference of the enterprise business interaction terminal based on the text semantic distance between the local business transformation consulting text and the remote business transformation consulting text includes: obtaining a first consulting dialogue text set and a second consulting dialogue text set of the local business transformation consulting text and the remote business transformation consulting text; determining the transformation consulting preference of the enterprise business interaction terminal based on the first consulting dialogue text set and the second consulting dialogue text set.

[0146] In this application, the enterprise affairs interaction terminal is a key node connecting internal and external businesses, and its consultation dialogue text contains rich transformation needs and preference information. In order to deeply mine this information, this embodiment provides a technical solution for determining the transformation consultation preference of the enterprise affairs interaction terminal based on text semantic distance.

[0147] First, the big data system needs to obtain two sets of consultation dialogue texts for local business transformation consultation texts and off-site business transformation consultation texts, which are respectively recorded as the first consultation dialogue text set and the second consultation dialogue text set. These two text sets are the basic data for analyzing the transformation consultation preferences of enterprise transaction interaction terminals. In practical applications, the big data system can obtain these text data through web crawling technology, API interface calls, database queries, etc.

[0148] After obtaining the consultation dialogue text sets, the big data system needs to preprocess the texts for subsequent semantic analysis. The preprocessing steps include text cleaning, word segmentation, stop word removal, etc. For example, for a certain text paragraph in the first consultation dialogue text set, the big data system can use natural language processing tools to perform word segmentation to obtain a sequence composed of words or phrases. At the same time, the big data system will remove some common stop words, such as "de" (of), "le" (already), etc., to reduce the complexity of subsequent analysis.

[0149] Next, the big data system needs to convert the preprocessed texts into numerical feature vectors for calculating semantic distances. This step can be achieved through word vector embedding technology. Word vector embedding is a method of converting words or phrases in texts into high-dimensional vector representations, which can capture the semantic relationships between words. For example, the big data system can use pre-trained models such as Word2Vec, GloVe, or BERT to convert each word in the text into a vector with a fixed dimension. In this way, a text paragraph can be represented as a sequence of vectors, and then a vector representing the entire text can be obtained by means of averaging or summing.

[0150] With the vectors representing the local business transformation consultation texts and the off-site business transformation consultation texts, the big data system can calculate the text semantic distance between them. The text semantic distance is a numerical indicator used to quantify the similarity or difference in semantics between two texts. Commonly used methods for calculating text semantic distances include cosine similarity, Euclidean distance, etc. For example, the big data system can use cosine similarity to calculate the angle between two text vectors, and the smaller the angle, the more similar the texts are.

[0151] Based on the calculated text semantic distance, the big data system can further determine the transformation consultation preferences of enterprise transaction interaction terminals. Specifically, the big data system can compare the semantic distances between the local business transformation consultation texts and the off-site business transformation consultation texts and other consultation dialogue texts of the enterprise transaction interaction terminal to judge the preferences of the enterprise transaction interaction terminal in terms of transformation consultation.

[0152] If the semantic distance between the local business transformation consultation text and other consultation dialogue texts of the enterprise affairs interaction terminal is small, while the semantic distance between the remote business transformation consultation text and other consultation dialogue texts of the enterprise affairs interaction terminal is large, then the big data system can believe that the enterprise affairs interaction terminal is more inclined to local business transformation in terms of transformation consultation. Conversely, if the semantic distance between the remote business transformation consultation text and other consultation dialogue texts of the enterprise affairs interaction terminal is small, while the semantic distance between the local business transformation consultation text and other consultation dialogue texts of the enterprise affairs interaction terminal is large, then the big data system can believe that the enterprise affairs interaction terminal is more inclined to remote business transformation in terms of transformation consultation.

[0153] In order to further refine the transformation consulting preferences of the enterprise affairs interaction terminal, the big data system can also consider the numerical range of the text semantic distance. For example, the big data system can set a threshold. When the text semantic distance is less than the threshold, the two texts are considered to be highly similar in semantics; when the text semantic distance is greater than the threshold, the two texts are considered to be semantically different. By adjusting the size of the threshold, the big data system can flexibly control the judgment criteria for transformation consulting preferences.

[0154] Finally, based on the determined transformation consulting preferences, the big data system can provide more personalized transformation strategy recommendations for the enterprise affairs interaction terminal. For example, if the enterprise affairs interaction terminal prefers local business transformation in transformation consulting, then the big data system can suggest that the enterprise focus on the needs and characteristics of the local market when formulating the transformation strategy; if the enterprise affairs interaction terminal prefers remote business transformation in transformation consulting, then the big data system can suggest that the enterprise fully consider the expansion of remote markets and cross-cultural communication issues when formulating the transformation strategy. In this way, the big data system can more accurately reveal the preferences and needs of the enterprise affairs interaction terminal in transformation consulting, and provide more accurate and powerful support for the enterprise.

[0155] In some other independent embodiments, the transformation consulting preference is a non-linked operation authentication preference; the non-linked operation authentication preference parsing algorithm is a long short-term memory algorithm or a deep residual algorithm; the transformation consulting semantic parsing algorithm corresponding to the transformation consulting preference is used to parse the enterprise business interaction terminal to obtain the transformation mode characteristics of the enterprise business interaction terminal, including: determining the consultation status label of the business online transformation consulting text in the business online transformation consulting text cluster; the consultation status label includes a real-time consultation status and a delayed consultation status; if the consultation status label matches the non-linked operation authentication preference parsing algorithm, the business online transformation consulting text is parsed using the non-linked operation authentication preference parsing algorithm to obtain the transformation mode characteristics of the enterprise business interaction terminal; if the consultation status label does not match the non-linked operation authentication preference parsing algorithm, the consultation retelling text of the business online transformation consulting text is obtained, and the consultation retelling text is parsed using the non-linked operation authentication preference parsing algorithm to obtain the transformation mode characteristics of the enterprise business interaction terminal.

[0156] It can be understood that when the transformation consulting preference is determined as a non-linked operation authentication preference, the big data system needs to adopt a specific semantic parsing algorithm to deeply analyze the enterprise transaction interaction terminal and then extract its transformation mode characteristics.

[0157] Among them, the consultation status label is an important metadata, which describes whether the text consultation is conducted in real time or with a delay. In the actual application scenarios of manufacturing enterprises, the real-time consultation status may represent urgent business needs or sudden market changes, while the delayed consultation status may represent more routine or planned business consultations. The big data system concretizes the consultation status label into two values: 0 represents the real-time consultation status, and 1 represents the delayed consultation status.

[0158] Next, the big data system needs to determine whether the consultation status label matches the non-linked operation authentication preference parsing algorithm. The non-linked operation authentication preference parsing algorithm is a key algorithm for parsing the consultation preferences of the enterprise transaction interaction terminal transformation. It can be a long short-term memory algorithm (LSTM) or a deep residual algorithm (ResNet). Both algorithms are good at processing sequence data and can extract deep semantic features from text.

[0159] If the consultation status label matches the non-linked operation authentication preference parsing algorithm, that is, the consultation status is real-time consultation status, the big data system directly uses the non-linked operation authentication preference parsing algorithm to parse the business launch transformation consultation text. In this step, the algorithm analyzes the keywords and phrases in the text and the semantic relationship between them, and then generates a numerical vector representing the transformation mode characteristics of the enterprise transaction interaction terminal. This vector may contain multiple dimensions, each of which represents a specific transformation feature or attribute, and the value on each dimension reflects the strength and importance of the feature or attribute.

[0160] However, if the consultation status label does not match the non-linked operation authentication preference parsing algorithm, that is, when the consultation status is a delayed consultation status, the big data system cannot directly parse the business launch transformation consultation text. Because the text in the delayed consultation status may contain more additional information about historical background, market environment, etc., this information may not be directly relevant to parsing the transformation mode characteristics. Therefore, the big data system needs to first obtain the consultation retelling text of the business launch transformation consultation text. The consultation retelling text is a simplification or reconstruction of the original consultation text. It removes information that is not directly related to the transformation mode characteristics and only retains the core business transformation needs and preferences.

[0161] After obtaining the consultation retelling text, the big data system uses the non-linked operation authentication preference parsing algorithm to parse it. Similar to the parsing process in the real-time consultation state, the algorithm analyzes the keywords and phrases in the retelling text and the semantic relationship between them to generate a numerical vector representing the characteristics of the transformation mode of the enterprise affairs interaction terminal. This vector also contains multiple dimensions, and the value on each dimension reflects the strength and importance of a specific feature or attribute of the enterprise affairs interaction terminal in the transformation process.

[0162] Through the above process, the big data system can obtain the transformation mode characteristics of the enterprise transaction interaction terminal, which are of great significance for guiding enterprises to formulate transformation strategies. For example, if the value of a certain dimension in the numerical vector is high, it means that the enterprise transaction interaction terminal pays great attention to this feature or attribute during the transformation process, and the enterprise should pay more attention and investment when formulating the transformation strategy. With such a design, the big data system can more accurately reveal the transformation mode characteristics of the enterprise transaction interaction terminal and provide enterprises with more accurate and powerful transformation strategy support.

[0163] Furthermore, a readable storage medium is provided, on which a program is stored, and the program implements the above method when executed by a processor.

[0164] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described in the above exemplary embodiment can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided by the present invention, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

Claims

1. A method for analyzing business consultation dialogues based on Internet services, characterized in that: The method is performed by a big data system, and the method includes: Obtaining a business consultation log of an enterprise affairs interaction terminal, and determining that the business consultation log contains at least two target consultation dialogue texts of the enterprise affairs interaction terminal; the business consultation log refers to log information containing user dialogue behavior data, and the business consultation log includes multiple multi-terminal consultation dialogue texts; the at least two target consultation dialogue texts belong to the multiple multi-terminal consultation dialogue texts; Obtaining a consulting dialogue text example configured with a priori learning annotations of an enterprise affairs interaction terminal; optimizing an algorithm using the consulting dialogue text example to obtain a dialogue interest event parsing algorithm for parsing dialogue interest events for the enterprise affairs interaction terminal; For each of the target consultation dialogue texts, the enterprise affairs interaction terminal is parsed for dialogue interest events to obtain dialogue interest event information of the enterprise affairs interaction terminal; wherein, for each of the target consultation dialogue texts, the enterprise affairs interaction terminal is parsed for dialogue interest events to obtain dialogue interest event information of the enterprise affairs interaction terminal, including: for each of the target consultation dialogue texts, the enterprise affairs interaction terminal is parsed for dialogue interest events according to the dialogue interest event parsing algorithm to obtain dialogue interest event information of the enterprise affairs interaction terminal; Extracting frequent text paragraphs of the enterprise affairs interaction terminal from at least part of the target consultation dialogue text, performing keyword vector analysis on the frequent text paragraphs, and obtaining keyword vector information of the enterprise affairs interaction terminal; According to at least one of the conversation interest event information or the keyword vector information, semantic annotation processing is performed on feature space members in the demand semantic feature space to obtain the enterprise affairs consultation demand semantics of the enterprise affairs interaction terminal in the demand semantic feature space; The example of obtaining the consulting dialogue text with the prior learning annotations configured for the enterprise affairs interaction terminal includes: Obtaining the consulting dialogue text examples of the profile knowledge information and key entity semantics of the enterprise affairs interaction terminal; According to the portrait knowledge information and the key entity semantics, the consulting dialogue text example is processed by an initial NLP algorithm to obtain a priori learning annotations of the enterprise affairs interaction terminal; The performing conversation interest event parsing on the enterprise affairs interaction terminal according to the conversation interest event parsing algorithm to obtain conversation interest event information of the enterprise affairs interaction terminal includes: Parsing the conversation interest events of the enterprise affairs interaction terminal according to the conversation interest event parsing algorithm to obtain the business community relationship distribution information of the enterprise affairs interaction terminal and the push response status characteristics of each consultation push record of the business community; The step of performing semantic annotation processing on feature space members in the demand semantic feature space according to at least one of the conversation interest event information or the keyword vector information to obtain the enterprise affairs consultation demand semantics of the enterprise affairs interaction terminal in the demand semantic feature space includes: According to the business community relationship distribution information, each of the push response status characteristics and the keyword vector information, semantic annotation processing is performed on the feature space members in the demand semantic feature space to obtain the enterprise affairs consultation demand semantics of the enterprise affairs interaction terminal in the demand semantic feature space; The steps of obtaining the push response status characteristics of each consultation push record of the business community include: According to the conversation interest event parsing algorithm, the conversation interest event of the enterprise affairs interaction terminal is parsed to obtain the consultation push attention characteristics of each consultation push record in the business community of the enterprise affairs interaction terminal and the distribution characteristics of the involvement portrait of each consultation push record; the involvement portrait of any consultation push record includes the target portrait in the consultation push record and the first annotated portrait on the first consultation semantic relationship network constituting the consultation push record and the second annotated portrait on the second consultation semantic relationship network constituting the consultation push record; wherein the target portrait is the user group portrait targeted by the push content; the first annotated portrait on the first consultation semantic relationship network reflects the semantic association between the push content and the user's historical consultation or behavior, and the second annotated portrait on the second consultation semantic relationship network reveals the potential connection between the push content and the user community or business scenario; For each of the consultation push records, determining an interactive mapping relationship network of the consultation push records according to the distribution characteristics of each of the involved portraits of the consultation push records; Based on the consultation push attention feature and the interactive mapping relationship network, obtaining a push response state feature of the consultation push record; The obtaining the push response state feature of the consultation push record based on the consultation push attention feature and the interactive mapping relationship network includes: Obtaining a first influence weight of the consultation push attention feature and a second influence weight of the interactive mapping relationship network; According to the first influence weight and the second influence weight, feature integration is performed on the consultation push attention feature and the interactive mapping relationship network to obtain a push response state feature of the consultation push record; The step of integrating the consultation push attention feature and the interactive mapping relationship network according to the first influence weight and the second influence weight to obtain the push response state feature of the consultation push record includes: The influence weight comparison result is obtained by comparing the first influence weight and the second influence weight; wherein the first influence weight reflects the importance of the consultation push attention feature in evaluating the push response status; the second influence weight represents the contribution of the interactive mapping relationship network in evaluating the push response status; the influence weight comparison result is obtained for comparing the first influence weight and the second influence weight to determine the relative importance of the first influence weight and the second influence weight in feature integration; and the first influence weight and the second influence weight are set according to historical data, user behavior analysis or business strategy; If the impact weight comparison result meets the comparison requirements, indicating that the difference between the first impact weight and the second impact weight is within a predetermined threshold range, the consultation push attention feature and the interactive mapping relationship network are semantically reconstructed and integrated based on a structured semantic deduction strategy to obtain a push response state feature of the consultation push record; If the influence weight comparison result does not meet the comparison requirement, indicating that the difference between the first influence weight and the second influence weight exceeds a predetermined threshold range, the consultation feedback text semantics corresponding to the selected influence weight of the first influence weight and the second influence weight is determined as the push response state feature of the consultation push record; the consultation feedback text semantics includes the consultation push attention feature and the interactive mapping relationship network; Alternatively, the step of integrating the consultation push attention feature and the interactive mapping relationship network based on the first influence weight and the second influence weight to obtain the push response state feature of the consultation push record includes: According to the first influence weight and the second influence weight, the characteristic thermal values ​​of the consultation push attention feature and the interactive mapping relationship network are determined; there is a quantitative relationship between the characteristic thermal value and the corresponding influence weight, and the larger the influence weight, the higher the characteristic thermal value; the entity semantic encoding strategy of the consultation push attention feature and the interactive mapping relationship network is the same; Based on each of the characteristic thermal values, the consultation push attention feature and the interactive mapping relationship network are processed to obtain the push response state feature of the consultation push record.

2. The method according to claim 1, characterized in that The frequent text paragraphs include business launch transformation consultation texts; the keyword vector information includes transformation mode features; The step of extracting frequent text paragraphs of the enterprise affairs interaction terminal from at least part of the target consultation dialogue text, performing keyword vector analysis on the frequent text paragraphs, and obtaining keyword vector information of the enterprise affairs interaction terminal includes: In the case where the conversation interest event information includes a transformation consultation tag, determining the business launch transformation consultation text cluster of the enterprise affairs interaction terminal in the target consultation conversation text according to the transformation consultation tag; the business launch transformation consultation text cluster includes a local business transformation consultation text and a remote business transformation consultation text; Determining the transformation consulting preference of the enterprise affairs interaction terminal according to the text semantic distance between the local business transformation consulting text and the remote business transformation consulting text; the transformation consulting preference includes a non-linked operation authentication preference and a linked operation authentication preference; According to the transformation consulting semantic parsing algorithm corresponding to the transformation consulting preference, the enterprise affairs interaction terminal is parsed to obtain the transformation mode characteristics of the enterprise affairs interaction terminal; the transformation consulting semantic parsing algorithm includes a linkage operation authentication preference parsing algorithm and a non-linkage operation authentication preference parsing algorithm; The transformation consulting tag includes a business process transformation tag and an authority authentication transformation tag; the business launch transformation consulting text cluster of the enterprise affairs interaction terminal in the target consulting dialogue text is determined based on the transformation consulting tag, including: Determine, according to the business process transformation tag or the authority authentication transformation tag, the location distribution information of the business launch transformation consulting text cluster of the enterprise affairs interaction terminal in the target consulting dialogue text; According to the correlation between the business process transformation tag and the authority authentication transformation tag, the text content correlation range of the business launch transformation consulting text cluster is determined.

3. A big data system, characterized in that: It comprises a processor and a memory; the processor and the memory are communicatively connected, and the processor is used to read a computer program from the memory and execute it to implement the method described in any one of claims 1-2 above.

4. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program implements the method described in any one of claims 1 to 2 when running.

Images