Intelligent early warning reasoning system based on multi-scene analysis

Abstract

The invention discloses an intelligent early warning reasoning system based on multi-scene analysis, and belongs to the technical field of communication service early warning reasoning. The platform end comprises a scene analysis module; the scene analysis module is used for configuring and updating a corresponding scene recognition model for the user side to obtain corresponding scene information; the user side comprises a scene recognition module and an early warning analysis module; the scene recognition module is used for performing real-time scene recognition through a configured scene recognition model to obtain scene information of a user, and generating a dialogue analysis graph according to the scene information; the dialogue analysis graph is sent to an early warning analysis module; the early warning analysis module is used for performing dialogue analysis early warning, setting a dialogue expectation diagram by a user, and performing real-time analysis on a dialogue analysis diagram according to the dialogue expectation diagram to obtain expectation deviation data; the expected deviation data are supplemented into the dialogue analysis graph in real time, and a dialogue monitoring graph is obtained; displaying the dialogue monitoring graph to the user in real time; and performing early warning analysis according to the dialogue monitoring graph.

Description

Technical Field

[0001] The present invention belongs to the technical field of AC service early warning reasoning, and specifically relates to an intelligent early warning reasoning system based on multi-scenario analysis. Background Art

[0002] In the field of human-computer dialogue, with the continuous expansion and complication of application scenarios, traditional dialogue systems have been difficult to meet the needs of users in multi-scenarios. Especially in multiple scenarios such as customer service, business negotiation, online education, and social media monitoring, the dialogue content, user behavior, and potential risks all show a high degree of diversity and complexity. Therefore, it is particularly urgent to develop an intelligent early warning reasoning system that can adapt to multi-scenarios. For example, in the customer service scenario, users may be dissatisfied or file complaints due to product problems, service quality, or improper handling of personal information. Traditional dialogue systems often rely on keyword matching or preset rules to identify user emotions and problems, but this method is inadequate when faced with complex and changeable dialogue content.

[0003] Based on this, in order to achieve intelligent early warning reasoning for communication services, the present invention provides an intelligent early warning reasoning system based on multi-scenario analysis. Summary of the Invention

[0004] In order to solve the problems existing in the above solutions, the present invention provides an intelligent early warning reasoning system based on multi-scenario analysis.

[0005] The object of the present invention can be achieved by the following technical solutions:

[0006] An intelligent early warning reasoning system based on multi-scenario analysis, comprising a platform side and a user side;

[0007] The platform side includes a scenario analysis module;

[0008] The scenario analysis module is used to configure and update a corresponding scenario recognition model for the user side, and the scenario recognition model is used to identify the dialogue scenario of the user in real time and obtain corresponding scenario information.

[0009] Further, the method for configuring a scenario recognition model for the user side includes:

[0010] Presetting several user classifications by the platform party; obtaining scenario material data of the user classifications;

[0011] Establishing a corresponding scenario recognition model for the user classifications according to the scenario material data, and marking corresponding user classification labels for the scenario recognition model;

[0012] Real-time identifying the user information of each user, and marking corresponding user classification labels for the user according to the user information;

[0013] Configure a corresponding scenario recognition model for the user terminal of the user according to the user classification label.

[0014] Furthermore, the method for updating the corresponding scenario recognition model for the user terminal includes:

[0015] Real-time identify the user information corresponding to the user classification, and obtain the scenario optimization data corresponding to the user classification according to the user information;

[0016] Perform update analysis on the scenario recognition model of the user classification according to the scenario optimization data to obtain model update data;

[0017] Update the scenario recognition model of the user terminal corresponding to the user classification according to the model update data.

[0018] The user terminal includes a scenario recognition module and a warning analysis module;

[0019] The scenario recognition module is used to perform real-time scenario recognition through the configured scenario recognition model, obtain the scenario information of the user, generate a dialogue analysis diagram according to the scenario information; and send the dialogue analysis diagram to the warning analysis module.

[0020] Furthermore, when the scenario recognition model encounters an unrecognizable dialogue scenario, it is processed according to the preset processing measures, and scenario optimization data is generated and sent to the platform side.

[0021] Furthermore, the method for performing real-time scenario recognition through the configured scenario recognition model includes:

[0022] The platform party presets dialogue collection items; collect the dialogue data between the user and the target person in real time according to the dialogue collection items;

[0023] Analyze the dialogue data through the scenario recognition model to obtain the corresponding scenario information.

[0024] Furthermore, the method for generating a dialogue analysis diagram according to the scenario information includes:

[0025] The platform party configures a corresponding feature detail list for the user terminal according to the user information, and the feature detail list is used to count the single inference features of the corresponding inference items;

[0026] Extract features from the scenario information according to the feature detail list to obtain the single inference features of the inference items;

[0027] Generate a dialogue analysis diagram according to the single inference features.

[0028] Furthermore, the method for extracting features from the scenario information according to the feature detail list includes:

[0029] Step SA1: Establish a feature matching model. The expression of the feature matching model is:

[0030]

[0031] In the formula: (CR, K ij ) is the input data, CR is the scenario information, K ij represents the single inference feature of the corresponding inference item, i represents the corresponding inference item, i = 1, 2,..., n, where n is the number of types of inference items in the feature list; j represents the single inference feature corresponding to the corresponding inference item, j = 1, 2,..., m, where m is the number of types of single inference features corresponding to the corresponding inference item in the feature list; CR→K ij indicates that the scenario information meets the matching requirements of the single inference feature of the corresponding inference item; the output data is the feature matching value TP(CR, K ij ), and the feature matching value is 1 or 0;

[0032] Step SA2: Generate the input data of the inference item according to the feature list and the scenario information, analyze the input data through the feature matching model, and obtain the corresponding feature matching value;

[0033] When the feature matching value is 1, stop generating the input data of the inference item, and output the single inference feature with the feature matching value of 1 marked with the corresponding inference item label;

[0034] When the feature matching value is 0, continue to generate the input data of the inference item until the feature matching value is 1;

[0035] Step SA3: Loop step SA2 until the single inference feature of each inference item is obtained, and end the analysis.

[0036] The warning analysis module is used to perform dialogue analysis and warning. The user sets a dialogue expectation diagram, and the dialogue analysis diagram is recognized in real time. According to the dialogue expectation diagram, the dialogue analysis diagram is analyzed in real time to obtain the expected deviation data of the corresponding inference item; the expected deviation data is supplemented into the dialogue analysis diagram in real time, and the current dialogue analysis diagram is marked as a dialogue monitoring diagram; the dialogue monitoring diagram is displayed to the user in real time;

[0037] Perform warning analysis according to the dialogue monitoring diagram to obtain dialogue warning data, and supplement the dialogue warning data into the dialogue monitoring diagram.

[0038] Furthermore, the expected deviation data includes a real-time deviation value.

[0039] Furthermore, the expected deviation data also includes a comprehensive deviation value. The method for obtaining the comprehensive deviation value includes:

[0040] Real-time identify the generation time of the dialogue analysis graph, and mark the generation time as T v , where v = 1, 2, ……, z, and z is the number of dialogue analysis graphs;

[0041] Calculate the positioning value of the dialogue analysis graph according to the generation time; preset K positioning intervals and the weight coefficients of the corresponding positioning intervals;

[0042] Identify the real-time deviation value of the dialogue analysis graph, classify the real-time deviation value according to the positioning value and the positioning interval, obtain K positioning classifications, calculate the sum of each real-time deviation value within the positioning classification, and mark it as the positioning classification value;

[0043] Mark the positioning classification value as DR c , where c represents the corresponding positioning classification, c = 1, 2, ……, K;

[0044] Calculate the comprehensive deviation value according to the comprehensive evaluation formula, and the comprehensive evaluation formula is:

[0045]

[0046] In the formula: ZP is the comprehensive deviation value; λ c is the weight coefficient of the corresponding positioning classification.

[0047] Furthermore, the positioning value calculation formula is:

[0048]

[0049] In the formula: DW v is the positioning value; T0 is the current time.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] By constructing an advanced scene recognition model, it can accurately identify key information such as the scene where the dialogue is located, the dialogue theme, and the user's emotions, positions, and interest demands. This function greatly improves the depth and breadth of dialogue understanding, enabling the system to better understand the user's intentions and needs, and thus providing more accurate and personalized services. By real-time analyzing the dialogue content, potential risks and conflicts can be warned. In the customer service scenario, this helps enterprises to timely discover and handle customer dissatisfaction and complaints, improving customer satisfaction and loyalty. In the business negotiation scenario, this helps negotiators to better grasp the positions and interest demands of the other party, and formulate more effective negotiation strategies. In the online education scenario, this helps teachers to timely discover students' learning obstacles and psychological problems, and provide timely tutoring and support. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0053] Figure 1 This is the principle block diagram of the present invention. Detailed implementation manners

[0054] The following will clearly and completely describe the technical solutions of the present invention in combination with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0055] As Figure 1 shown, an intelligent early warning inference system based on multi-scenario analysis includes a platform side and a user side;

[0056] The platform side is used by the platform party to connect to the user sides at each user location and includes a scenario analysis module;

[0057] The scenario analysis module is used to configure and update the corresponding scenario recognition model for the corresponding user side. The scenario recognition model is used to identify the dialogue scenario of the user in real time and obtain the corresponding scenario information; Exemplarily, identify the scenario where the dialogue is located, such as business negotiation, emotional communication, technical support, etc., extract and analyze the key information in each scenario, such as dialogue topic, user emotion, position, interest demand, etc.; that is, the scenario information includes relevant information such as dialogue scenario, dialogue topic, user emotion, position, interest demand, etc.

[0058] The method for configuring the scenario recognition model for the user side includes:

[0059] The platform party presets several user classifications. For example, for customer service for a certain type of product, the platform party prepares to establish a targeted scenario recognition model for the customer service of this product, and regards it as a user classification. That is, the user classifications are mainly set according to the business needs, resources, etc. of the platform party; The platform party can also set the corresponding user classification requirements, and perform intelligent classification based on the current classification method according to the user classification requirements to form several user classifications;

[0060] Obtain scenario material data for the corresponding user classification. The scenario material data is the conversation data collected from multiple sources within the user classification, such as conversations in different scenarios like customer service, business negotiation, online education, social media, etc.; ensure the diversity and representativeness of the data to cover various possible conversation situations and scenarios;

[0061] Establish a scenario recognition model based on the scenario material data and label the corresponding user classification labels for the scenario recognition model;

[0062] Real-time identify the user information of each user, mainly including information related to the user classification, and label the corresponding user classification labels for the user according to the user information;

[0063] Configure the corresponding scenario recognition model for the user side of the corresponding user according to the user classification label.

[0064] In one embodiment, the scenario recognition model is established based on the scenario material data and is established based on existing intelligent technologies. Exemplarily:

[0065] Data preprocessing: Clean the collected scenario material data, remove irrelevant information, noise and redundant data. Perform text preprocessing operations on the conversation text such as word segmentation, stop word removal, and stemming.

[0066] Feature extraction: Use natural language processing techniques to extract key information in the conversation, such as keywords, phrases, sentences, etc. Consider the context information of the conversation to capture the dynamics and coherence of the conversation.

[0067] Feature selection: Select the most representative features according to the objectives and requirements of the conversation recognition model. Feature selection algorithms can be used to automatically determine which features contribute the most to the model performance.

[0068] Model selection: Select a suitable model architecture according to the complexity of the problem and the characteristics of the data. For example, deep learning models (such as convolutional neural network CNN, recurrent neural network RNN and its variants) can be used to process sequence data and capture the time-dependent relationships in the conversation. Consider using ensemble learning methods to improve the robustness and accuracy of the model.

[0069] Model training: Use the preprocessed scenario material data and the selected features to train the model. During the training process, continuously adjust the parameters and architecture of the model to optimize its performance.

[0070] Model evaluation: Use the test data set to evaluate the performance of the model. Evaluation metrics can include accuracy, recall rate, F1 score, etc. Conduct a detailed analysis of the model's performance in different scenarios to determine its advantages and limitations.

[0071] Model Optimization: Optimize the model based on the evaluation results. Optimization methods include adjusting model parameters, increasing the amount of data, improving feature extraction methods, etc. Consider using transfer learning or domain adaptation techniques to improve the generalization ability of the model in new scenarios.

[0072] The method for updating the corresponding scene recognition model for the user side includes:

[0073] Real-time identify the user information corresponding to the corresponding user classification, and obtain the corresponding scene optimization data for the corresponding user classification according to the user information in real time;

[0074] Perform update analysis on the scene recognition model of the corresponding user classification according to the scene optimization data corresponding to the user scene, and obtain model update data; that is, according to the learning optimization method preset in the scene recognition model, optimize according to the scene optimization data, and realize the recognition and analysis of the new scene by the subsequent scene recognition model;

[0075] Update the scene recognition model of the user side corresponding to the corresponding user classification according to the model update data.

[0076] The user side includes a scene recognition module and a warning analysis module;

[0077] The scene recognition module is used to perform real-time scene recognition through the configured scene recognition model, obtain the scene information of the user, generate a dialogue analysis diagram according to the obtained scene information; mark the other party having a conversation with the user as the target person.

[0078] In one embodiment, when an unrecognizable dialogue scene is encountered through the scene recognition model, handle it according to the preset processing measures, and generate scene optimization data and send it to the platform side.

[0079] The preset processing measures are how to handle when an unrecognizable dialogue scene is encountered, such as warning the user, performing risk assessment, and handling according to the risk assessment results, etc. The specific processing measures are set according to the actual situation; the scene optimization data is generated according to the data related to the unrecognizable dialogue scene.

[0080] The method for performing real-time scene recognition through the configured scene recognition model includes:

[0081] Preset dialogue collection items according to the data required for analysis by the scene recognition model; collect the dialogue data between the user and the target person in real time according to the dialogue collection items;

[0082] Analyze the dialogue data through the scene recognition model to obtain the corresponding scene information.

[0083] In one embodiment, the method for generating a dialogue analysis diagram according to the scene information includes:

[0084] The platform party sets corresponding reasoning items for each user. The reasoning items are set according to the data items required for subsequent early warning reasoning, such as reasoning items related to topics, emotions, content, etc. The platform party sets a feature detail list based on the possible scenario information of the user classification on the corresponding reasoning items. For example, various conversation topic data of the simulated determined topic reasoning item are summarized, and various single reasoning features of the topic reasoning item are set. The single reasoning feature is the feature data of the scenario information on the corresponding reasoning item, which is used to standardize the subsequent dialogue analysis diagram presented to the user and facilitate the user to understand the corresponding dialogue analysis diagram. That is, a single reasoning feature can correspond to a feature range. For example, certain content ranges and emotion ranges correspond to a single reasoning feature. For emotions, single reasoning features can be set using emotion values, emotion levels, etc. For topics, etc., the upper-level concepts can also be used to form feature ranges. That is, the feature detail list is used to count the various single reasoning features corresponding to the corresponding reasoning item. The user can adjust the feature detail list by means of adding, deleting, merging, renaming, etc. The corresponding feature detail list is configured for the user terminal according to the user information.

[0085] Feature extraction is performed on the scenario information according to the feature detail list to obtain the single reasoning features of the corresponding reasoning item.

[0086] A corresponding dialogue analysis diagram template is preset for display according to the display methods of the preset various reasoning items. The dialogue analysis diagram is generated according to the dialogue analysis diagram template and the single reasoning features of each reasoning item.

[0087] In one embodiment, when performing feature extraction on the scenario information according to the feature detail list, it can be performed based on the existing feature recognition extraction methods.

[0088] In one embodiment, the method for performing feature extraction on the scenario information according to the feature detail list includes:

[0089] Step SA1: Establish a feature matching model. The feature matching model is used to perform feature matching on the scenario information with the single reasoning features of the corresponding reasoning item to determine the single reasoning features of the corresponding reasoning item. The expression of the feature matching model is:

[0090]

[0091] In the formula: (CR, K ij ) is the input data, CR is the scenario information, K ij represents the single reasoning feature of the corresponding reasoning item, i represents the corresponding reasoning item, i = 1, 2,..., n, n is the number of types of reasoning items in the feature detail list; j represents the single reasoning feature corresponding to the corresponding reasoning item, j = 1, 2,..., m, m is the number of types of single reasoning features corresponding to the corresponding reasoning item in the feature detail list; CR→Kij Indicates that the scenario information meets the matching requirements of the single - inference feature corresponding to the corresponding inference item, that is, it belongs to the feature range corresponding to the single - inference feature; the output data is the feature matching value TP(CR, K ij ), and the feature matching value is 1 or 0;

[0092] Step SA2: Generate the input data of the corresponding inference item according to the feature list and the scenario information, analyze the input data through the feature matching model, and obtain the corresponding feature matching value;

[0093] When the feature matching value is 1, stop generating the input data of this inference item, and mark the single - inference feature with the feature matching value of 1 with the corresponding inference item label for output;

[0094] When the feature matching value is 0, continue to generate the input data of this inference item until the feature matching value is 1;

[0095] Step SA3: Loop step SA2 until the single - inference features of each inference item are determined, and then end the analysis.

[0096] The warning analysis module is used to perform dialogue analysis and warning. The user sets a dialogue expectation graph, and the dialogue expectation graph is used to represent the dialogue analysis graph that the user expects to achieve; real - time identify the dialogue analysis graph, perform real - time analysis on the dialogue analysis graph according to the dialogue expectation graph, and obtain the expected deviation data of the corresponding inference item; supplement the expected deviation data of the corresponding inference item into the corresponding dialogue analysis graph in real - time, that is, supplement the corresponding expected deviation data into the dialogue analysis graph according to the inference item, mark the current dialogue analysis graph as the dialogue monitoring graph; display the dialogue monitoring graph to the user in real - time; perform warning analysis according to the dialogue monitoring graph to obtain dialogue warning data, and supplement the dialogue warning data into the corresponding dialogue monitoring graph.

[0097] In one embodiment, the expected deviation data is the real - time expected deviation value, marked as the real - time deviation value. The method for obtaining the real - time deviation value is: based on the existing method, compare the single - inference feature of the inference item with the corresponding dialogue expectation in the dialogue expectation graph to determine the expected deviation value. The expected deviation value is the degree of deviation from the dialogue expectation, such as the emotional deviation degree, the topic deviation degree. For the expected deviation degree better than the dialogue expectation, it is 0; such as establishing a deviation analysis model based on a CNN network or a DNN network, etc., and training through an artificial method to establish a training set. The training set includes input data and output data. The input data is the dialogue expectation graph and the dialogue analysis graph, and the output data is the expected deviation value of the corresponding inference item.

[0098] In one embodiment, the expected deviation data includes the real - time deviation value and the comprehensive deviation value. Because it is not comprehensive enough to perform warning analysis only relying on the real - time deviation value, the comprehensive deviation value is supplemented for optimization. The method for obtaining the comprehensive deviation value includes:

[0099] Real-time identify the generation time of the corresponding dialogue analysis diagram and mark it as T v , where v is a subscript representing the corresponding dialogue analysis diagram, v = 1, 2,..., z, and z is the number of dialogue analysis diagrams in this conversation;

[0100] Set the corresponding positioning value for the corresponding dialogue analysis diagram according to the obtained generation time. The calculation formula for the positioning value is:

[0101]

[0102] In the formula: DW v is the positioning value; T0 is the current time;

[0103] According to the positioning value calculation formula, it can be known that the value range of the positioning value is [0, 1); the platform divides the value range into K positioning intervals and sets the corresponding weight coefficients for the corresponding positioning intervals. For example, the value range is equally divided into 10 positioning intervals, and it can also be adjusted according to the actual situation, such as [0, e -2 ), [e -2 , e -4 ), [e -4 , e -7 ), [e -7 , e -10 ), [e 10 , e -∞ ); specifically, it is set by the platform according to the slope change of the positioning curve;

[0104] Identify the real-time deviation value of the dialogue analysis diagram, classify the obtained real-time deviation value according to the corresponding positioning value and positioning interval to obtain K positioning classifications, and calculate the sum of each real-time deviation value within the positioning classification, which is marked as the positioning classification value;

[0105] Mark the obtained positioning classification value as DR c , where c represents the corresponding positioning classification, c = 1, 2,..., K, and the number of positioning classifications is the same as the number of positioning intervals;

[0106] Calculate the comprehensive deviation value according to the comprehensive evaluation formula. The comprehensive evaluation formula is:

[0107]

[0108] In the formula: ZP is the comprehensive deviation value; λ c is the weight coefficient of the corresponding positioning classification.

[0109] In one embodiment, early warning analysis is performed according to the conversation monitoring graph. The platform party establishes an early warning analysis model based on a CNN network, a DNN network, etc., and conducts training by manually establishing a corresponding training set. The training set includes input data and output data. The input data is the conversation monitoring graph, and the output data is the conversation early warning data, such as relevant data such as conversation early warning items and suggested conversation directions. Analysis is performed through the successfully trained early warning analysis model to obtain the corresponding conversation early warning data.

[0110] In one embodiment, for the early warning analysis according to the conversation monitoring graph, analysis can also be performed based on other existing methods.

[0111] The above formulas are all calculated by removing the dimension and taking their numerical values. The formula is a formula obtained by collecting a large amount of data for software simulation to be closest to the actual situation. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0112] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent early warning inference system based on multi-scenario analysis, characterized in that, It includes a platform side and a user side; The platform side includes a scenario analysis module; The scenario analysis module is used to configure and update a corresponding scenario recognition model for the user side, and the scenario recognition model is used to identify the user's conversation scenario in real time and obtain corresponding scenario information; The user side includes a scenario recognition module and a warning analysis module; The scenario recognition module is used to perform real-time scenario recognition through the configured scenario recognition model, obtain the user's scenario information, and generate a conversation analysis diagram according to the scenario information; Send the conversation analysis diagram to the warning analysis module; The warning analysis module is used to perform conversation analysis and warning. The user sets a conversation expectation diagram, real-time identifies the conversation analysis diagram, and performs real-time analysis on the conversation analysis diagram according to the conversation expectation diagram to obtain the expected deviation data of the corresponding inference item; Supplement the expected deviation data into the conversation analysis diagram in real time, mark the current conversation analysis diagram as a conversation monitoring diagram; and display the conversation monitoring diagram to the user in real time; Perform warning analysis according to the conversation monitoring diagram to obtain conversation warning data, and supplement the conversation warning data into the conversation monitoring diagram.

2. The intelligent early warning inference system based on multi-scenario analysis according to claim 1, characterized in that, The method for configuring a scenario recognition model for the user side includes: The platform party presets several user classifications; obtains the scenario material data of the user classifications; Establish corresponding scenario recognition models for the user classifications according to the scenario material data, and mark corresponding user classification labels for the scenario recognition models; Real-time identify the user information of each user, and mark corresponding user classification labels for the user according to the user information; Configure a corresponding scenario recognition model for the user side of the user according to the user classification label.

3. An intelligent early warning inference system based on multi-scenario analysis according to claim 2, characterized in that The method for updating a corresponding scenario recognition model for the user side includes: Real-time identify the user information corresponding to the corresponding user classification, and obtain the scenario optimization data corresponding to the user classification in real time according to the user information; Perform update analysis on the scenario recognition model of the user classification according to the scenario optimization data to obtain model update data; Update the scenario recognition model of the user side corresponding to the user classification according to the model update data.

4. An intelligent early warning inference system based on multi-scenario analysis according to claim 1, characterized in that When the scenario recognition model encounters an unrecognizable conversation scenario, it is processed according to a preset processing measure, and scenario optimization data is generated and sent to the platform side.

5. An intelligent early warning inference system based on multi-scenario analysis according to claim 1, characterized in that, The method for performing real-time scenario recognition through a configured scenario recognition model includes: The platform party presets conversation collection items; collects the conversation data between the user and the target person in real time according to the conversation collection items; Analyze the conversation data through the scenario recognition model to obtain corresponding scenario information.

6. An intelligent early warning inference system based on multi-scenario analysis according to claim 1, characterized in that The method for generating a conversation analysis diagram according to scenario information includes: The platform party configures a corresponding feature list for the user side according to the user information, and the feature list is used to count the single inference features of the corresponding inference item; Extract features from the scenario information according to the feature list to obtain the single inference features of the inference item; Generate a conversation analysis diagram according to the single inference features.

7. An intelligent early warning inference system based on multi-scenario analysis according to claim 6, characterized in that, The method for extracting features from scenario information according to a feature list includes: Step SA1: Establish a feature matching model, and the expression of the feature matching model is: Where: (CR, K ij ) is the input data, CR is the scenario information, and K ij represents the single-inference feature of the corresponding inference item, i represents the corresponding inference item, i = 1, 2,..., n, where n is the number of types of inference items in the feature list; j represents the single-inference feature corresponding to the corresponding inference item, j = 1, 2,..., m, where m is the number of types of single-inference features corresponding to the corresponding inference item in the feature list; CR→K ij indicates that the scenario information meets the matching requirements of the single-inference feature of the corresponding inference item; the output data is the feature matching value TP(CR, K ij ), and the feature matching value is 1 or 0; Step SA2: Generate the input data of the inference item according to the feature breakdown list and the scenario information, and analyze the input data through the feature matching model to obtain the corresponding feature matching value; When the feature matching value is 1, stop generating the input data of the inference item, and output the single-inference feature with the feature matching value of 1 marked with the corresponding inference item label; When the feature matching value is 0, continue to generate the input data of the inference item until the feature matching value is 1; Step SA3: Loop Step SA2 until the single-inference feature of each inference item is obtained, and end the analysis.

8. An intelligent early warning inference system based on multi-scenario analysis according to claim 1, characterized in that, The expected deviation data includes the real-time deviation value.

9. An intelligent early warning reasoning system based on multi-scenario analysis according to claim 8, characterized in that, The expected deviation data also includes the comprehensive deviation value, and the method for obtaining the comprehensive deviation value includes: Real-time identify the generation time of the dialogue analysis graph, and mark the generation time as T v , where v = 1, 2, ……, z, and z is the number of dialogue analysis graphs; Calculate the positioning value of the dialogue analysis graph according to the generation time; preset K positioning intervals and the weight coefficients of the corresponding positioning intervals; Identify the real-time deviation value of the dialogue analysis graph, classify the real-time deviation value according to the positioning value and the positioning interval to obtain K positioning classifications, calculate the sum of the real-time deviation values within the positioning classification, and mark it as the positioning classification value; Mark the positioning classification value as DR c , where c represents the corresponding positioning classification, c = 1, 2, ……, K; Calculate the comprehensive deviation value according to the comprehensive evaluation formula, and the comprehensive evaluation formula is: where: ZP is the comprehensive deviation value; λ c is the weight coefficient of the corresponding positioning classification.

10. An intelligent early warning inference system based on multi-scenario analysis according to claim 9, characterized in that, The positioning value calculation formula is: Where: DW v is the positioning value; T0 is the current time.

Images