Event portrait generation method and related device
By combining generative models and error correctors, and utilizing machine learning and deep learning models to generate event profiles, the problem of insufficient accuracy and reliability of event profiles in existing technologies is solved, and efficient and accurate event profile generation is achieved.
Patent Information
- Application Number
- CN202511042212.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-28
- Publication Date
- 2025-10-31
Smart Images
Figure CN120873167A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of information extraction technology, and in particular to an event profile generation method and related apparatus. Background Technology
[0002] As informatization deepens, it brings new challenges and opportunities to the prevention of various events, enabling the prediction of events through more diverse data and indicators. For a given event, there are often multiple analytical dimensions, such as the urgency, scope of impact, and probability of occurrence. By analyzing these different dimensions to describe the event, an event profile is created, helping risk control personnel gain a more comprehensive understanding of the event situation.
[0003] Currently, there are various methods for obtaining event profiles, commonly including named entity recognition (NER) and deep learning. NER extracts structured information from raw text by recognizing named entities and combines it with statistically calculated metrics to assemble an event profile. This approach relies heavily on the capabilities of named entity recognition, requires significant model training costs, and is highly proprietary, making it difficult to apply to text processing across different domains. Deep learning prediction methods can obtain structured information and metrics that are difficult to statistically calculate, such as event severity, but prediction accuracy is low, and the extracted structured information is highly random, resulting in insufficient accuracy. Summary of the Invention
[0004] In view of the above problems, this application provides an event profile generation method and related apparatus to improve the accuracy and reliability of event profiles. The specific solution is as follows:
[0005] The first aspect of this application provides a method for generating event profiles, including:
[0006] Based on the generative model, structural data is extracted from the event description text to obtain the first structural data;
[0007] The structural data of the event description text is extracted based on the correction device, and the extraction result is fused with the first structural data to obtain the second structural data;
[0008] The second structured data is input into a machine learning model to extract event metrics, thereby obtaining the first event metric data.
[0009] The event description text, the second structural data, and the first event indicator data are input into a deep learning model to extract the event indicators and obtain the second event indicator data.
[0010] The first event indicator data and the second event indicator data are fused together and combined with the second structure data to obtain the event profile of the event description text.
[0011] In one possible implementation, the extraction of structured data from the event description text based on the generative model to obtain first structured data includes:
[0012] The extraction template and the event description text are input into the generation model, so that the generation model extracts the structural data from the event description text according to the structural data items contained in the extraction template, and obtains the first structural data.
[0013] In one possible implementation, the bias corrector includes: a first neural network model and a logic analyzer. The bias corrector is used to extract structural data from the event description text, and the extracted results are fused with the first structural data to obtain second structural data, including:
[0014] The first structural data and the event description text are input into the first neural network model to extract the structural data, thereby obtaining the third structural data.
[0015] The third structural data, the first structural data, and the event description text are input into the logic analyzer, and the third structural data and the first structural data are fused to obtain the second structural data.
[0016] In one possible implementation, the step of inputting the third structural data, the first structural data, and the event description text into the logic analyzer, and fusing the third structural data and the first structural data to obtain the second structural data, includes:
[0017] The logic analyzer determines whether each structural data value in the first structural data is in the event description text, and determines that the first target structural data value does not exist in the event description text;
[0018] The second target structure data value in the third structure data replaces the first target structure data value and is added to the first structure data. The second target structure data value and the first target structure data value correspond to the same structure data item.
[0019] In one possible implementation, the training process of the bias corrector includes:
[0020] The fourth structural data output by the generative model and the event description text are input into the correction device to obtain the fifth structural data;
[0021] The first loss value between the fifth structural data and the original structural data is determined based on the first loss function. After processing the first loss value based on the gradient descent algorithm, the parameters of the bias corrector are updated. The original structural data is the structural data corresponding to the pre-set event description text.
[0022] In one possible implementation, the training process of the machine learning model includes:
[0023] The sixth structural data output by the corrector and the event description text are input into the machine learning model to obtain the third event index data;
[0024] The second loss value between the third event indicator data and the original event indicator data is determined based on the second loss function. After processing the second loss value based on the gradient descent algorithm, the parameters of the machine learning model are updated. The original event indicator data is the event indicator data corresponding to the pre-set event description text.
[0025] In one possible implementation, the training process of the deep learning model includes:
[0026] The seventh structural data output by the corrector, the event description text, and the fourth event index data output by the machine learning model are input into the deep learning model to obtain the fifth event index data.
[0027] The third loss value between the fifth event index data and the original event index data is determined based on the third loss function. After processing the third loss value based on the gradient descent algorithm, the parameters of the deep learning model are updated. The original event index data is the event index data corresponding to the pre-set event description text.
[0028] A second aspect of this application provides an event profile generation apparatus, comprising:
[0029] The structural data extraction module is used to extract structural data from the event description text based on the generative model to obtain the first structural data.
[0030] The structural data correction module is used to extract the structural data from the event description text based on the correction device, and fuse the extraction result with the first structural data to obtain the second structural data;
[0031] The first indicator extraction module is used to input the second structured data into the machine learning model to extract event indicators and obtain the first event indicator data.
[0032] The second indicator extraction module is used to input the event description text, the second structured data, and the first event indicator data into a deep learning model to extract the event indicators, thereby obtaining the second event indicator data; and...
[0033] The event profile generation module is used to fuse the first event indicator data and the second event indicator data, and then combine them with the second structured data to obtain the event profile of the event description text.
[0034] A third aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the event profile generation method of the first aspect or any implementation thereof.
[0035] A fourth aspect of this application provides an electronic device, including at least one processor and a memory connected to the processor, wherein:
[0036] The memory is used to store computer programs;
[0037] The processor is used to execute the computer program so that the electronic device can implement the event profile generation method of the first aspect or any implementation thereof.
[0038] The fifth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to perform the event profile generation method described in the first aspect or any implementation thereof.
[0039] By employing the above technical solutions, the event profile generation method provided in this application can extract structural data from event description text based on a generative model to obtain first structural data. Based on this, a bias corrector is used to extract structural data from the event description text, and the extracted results are fused with the first structural data to obtain second structural data. This achieves the extraction of structured data from unstructured data and ensures the accuracy of the extracted structural data through data bias correction. Event indicators are extracted using a machine learning model to obtain highly accurate first event indicator data. Then, a deep learning model is used to predict second event indicator data. Finally, the first and second event indicator data are fused and combined with the second structural data to obtain an event profile of the event description text. This event profile generation method, by combining a generative model and a bias corrector, can extract existing and missing structured information and obtain indicator values through traditional machine learning and deep learning, resulting in a highly credible and readable event profile. Attached Figure Description
[0040] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.
[0041] Figure 1 An architecture diagram of an event profile generation system provided in this application;
[0042] Figure 2 A flowchart of an event profile generation method provided in this application;
[0043] Figure 3 Architecture diagram generated for the event profile provided in this application;
[0044] Figure 4 Architecture diagrams for training the various models provided in this application;
[0045] Figure 5 A structural diagram of an event portrait generation device provided in this application;
[0046] Figure 6 This is a structural diagram of an electronic device provided in this application. Detailed Implementation
[0047] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.
[0048] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.
[0049] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.
[0050] See Figure 1 , Figure 1A schematic diagram of the architecture of an event profile generation system is shown. The system may include a terminal 100 and a server 200. The server 200 can provide the event profile generation method provided in this embodiment to one or more terminals.
[0051] The terminal 100 may be equipped with an event profile generation application. The application and webpage can provide an interface. The terminal 100 can receive relevant parameters input by the user on the event profile generation interface and send the parameters to the server 200. The server 200 can obtain the processing result based on the received parameters and return the processing result to the terminal 100.
[0052] It should be understood that in some optional implementations, the terminal 100 can also complete the action of obtaining the processing result based on the received parameters on its own, without the need for the server to cooperate. This application embodiment is not limited to this.
[0053] The following description Figure 1 The product form of the mid-terminal 100;
[0054] The terminal 100 in this application embodiment can be a mobile phone, tablet computer, in-vehicle device, laptop computer, ultra-mobile personal computer (UMPC), netbook, personal digital assistant (PDA), etc., and this application embodiment does not impose any restrictions on it.
[0055] Terminal 100 may include a radio frequency unit, memory, input unit, display unit, camera (optional), audio circuitry (optional), speaker (optional), microphone (optional), headphone jack (optional), processor, external interface, power supply, and other components. Those skilled in the art will understand that the above-mentioned components are merely examples and do not constitute a limitation on the terminal or multifunctional device; it may include more or fewer components, or a combination of certain components, or different components.
[0056] The input unit can be used to receive input numeric or character information, and to generate key signal inputs related to user settings and function control of the portable multi-functional device. Specifically, the input unit may include a touchscreen (optional) and / or other input devices. Other input devices may include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.
[0057] Among them, the input device can receive input data, etc.
[0058] The display unit can be used to display information input by the user or information provided to the user, various menus of the terminal, interactive interfaces, file display, and / or playback of any multimedia file. In the embodiments of this application, the display unit can be used to display the interface for generating event profiles, processing results, etc.
[0059] The memory can be used to store software code related to the event profile generation method, the processor can execute the steps of the event profile generation method, and can also schedule other units (such as the above-mentioned input unit and display unit) to achieve the corresponding functions.
[0060] This radio frequency unit (optional) can be used to receive and send signals during information transmission or calls.
[0061] In this embodiment of the application, the radio frequency unit can send data to the server 200 and receive the processing results sent by the server 200.
[0062] It should be understood that this radio frequency unit is optional and can be replaced with other communication interfaces, such as a network port.
[0063] Terminal 100 also includes a power source (such as a battery) for supplying power to the various components.
[0064] Terminal 100 also includes an external interface, which can be a standard Micro USB interface or a multi-pin connector, which can be used to connect terminal 100 to other devices for communication or to connect a charger to charge terminal 100.
[0065] Server 200 includes a bus, a processor, a communication interface, and memory. The processor, memory, and communication interface communicate with each other via the bus.
[0066] The memory can be used to store software code related to the event profile generation method, the processor can execute the steps of the chip's event profile generation method, and can also schedule other units to achieve the corresponding functions.
[0067] Current technologies for generating event profiles typically employ either deep learning or traditional machine learning methods for analysis, then combine the analytical conclusions to form a complete event profile. Deep learning is generally used to process unstructured data, such as event description text and user comments, to obtain sentiment and other indicators. Traditional machine learning is generally used to process structured data, such as event severity, urgency, and frequency, to predict the probability of event occurrence. However, in practical applications, information such as event severity is often embedded in the text description and is difficult to obtain directly.
[0068] When using deep learning methods, only qualitative analysis of indicators such as text sentiment can be performed. When predicting indicators such as the probability of events, the prediction accuracy is often lower than that of traditional machine learning methods, such as XGBoost.
[0069] To address the aforementioned problems, this application provides an event profile generation method. The event profile generation method of this application embodiment will be described in detail below with reference to the accompanying drawings.
[0070] Reference Figure 2 , Figure 2 This is a flowchart illustrating an event profile generation method provided in an embodiment of this application, as follows: Figure 2 As shown, the event profile generation method provided in this application embodiment may include steps 201 to 205, which are described in detail below.
[0071] 201. Based on the generative model, structural data is extracted from the event description text to obtain the first structural data.
[0072] Specifically, refer to Figure 3 As shown, the information extractor can employ a generative model. By utilizing the text understanding and generation capabilities of the generative model, the extraction template and event description text are input into the generative model, which then extracts the structural data from the event description text according to the structural data items contained in the extraction template, thus obtaining the first structural data.
[0073] For example, refer to Figure 3 As shown, the input event description text is denoted as... Extract the template text and denote it as The information extractor is denoted as Among them, extracting template text Specify the attribute names to be extracted, such as "event domain, event time, event severity," and define the data to be extracted. The generator model outputs normalized, structured data. Input the event text and the extraction template text into the generator model to obtain structured data.
[0074] (1)
[0075] in Let be the i-th structured information extracted, and n be the total number of structured information extracted.
[0076] It is understood that those skilled in the art can choose a suitable generative model as an information extractor as needed, which will not be elaborated here.
[0077] 202. Based on the correction device, structural data is extracted from the event description text, and the extraction results are fused with the first structural data to obtain the second structural data.
[0078] Specifically, the correction mechanism here may include a first neural network model and a logic analyzer. The first structural data and event description text are input into the first neural network model for structural data extraction, resulting in third structural data. The third structural data, the first structural data, and the event description text are then input into the logic analyzer, which fuses the third and first structural data to obtain second structural data. The logic analyzer determines whether each structural data value in the first structural data exists in the event description text, identifying first target structural data values that are not present in the event description text. The second target structural data value from the third structural data replaces the first target structural data value and is added to the first structural data, with the second target structural data value corresponding to the same structural data item.
[0079] For example, refer to Figure 3 As shown, the correction device is used to generate and verify structured information, denoted as... The first neural network model uses Neural network, denoted as This is combined with the structure of logical analysis. Since the structured information to be generated is relatively independent, therefore, it utilizes... The characteristics of neural networks allow for the parallel generation of multiple target values, improving generation efficiency. The structured information extracted from formula (1) is concatenated with the event text and input into... In this process, new structured information is obtained:
[0080] (2)
[0081] in Let be the i-th structured information extracted, and n be the total number of structured information extracted.
[0082] get and Then, both, along with the event description text, are input into the logic analyzer, denoted as... This yields the final structured information. Due to the generative model... The extracted structured information is highly accurate for information already present in the original text; however, it cannot accurately predict information that is absent or missing from the original text. Therefore, using... Neural networks amplify structured information, generating more accurate predictions, but Neural network on the original text The existing structured information may have been predicted, leading to information distortion. Therefore, the logic analyzer combines... and Judge in sequence For each value in, if In the event description text If it already exists, then use ,like In the original text If it does not exist, then use By combining the information from the two parts, structured information that is as accurate as possible can be generated:
[0083]
[0084] in The i-th final structured information extracted is denoted as n, where n is the total number of structured information extracted.
[0085] 203. Input the second structure data into the machine learning model to extract event indicators and obtain the first event indicator data.
[0086] Specifically, refer to Figure 3 As shown, a machine learning model is used to extract indicator data, denoted as... This involves predicting event metrics such as probability of occurrence and severity, using the XGBoost model as the machine learning metric extractor. Input the machine learning metric extractor to obtain the probability of event occurrence:
[0087]
[0088] in The probability of an event occurring. The value represents the severity of the event and ranges from 0 to 1.
[0089] It is understood that those skilled in the art can select and adjust the types of machine learning models used above as needed, and no restrictions are imposed here.
[0090] 204. Input the event description text, the second structure data, and the first event indicator data into the deep learning model to extract the event indicators and obtain the second event indicator data.
[0091] Specifically, refer to Figure 3 As shown, to improve the accuracy of the event metric data, after obtaining the first event metric data mentioned above, a deep learning metric extractor is then used, denoted as... This information is then processed to obtain new metrics. The deep learning metrics extractor still employs... Neural networks can simultaneously generate multiple independent metrics. (This refers to the ability to translate event description text into a single, coherent sentence.) Structured information ,index and Simultaneously, the data is input into the deep learning metric extractor to obtain new metrics:
[0092]
[0093] in The probability of an event occurring. This represents the severity of the event, with values ranging from 0 to 1. It combines metrics generated through machine learning. and and metrics generated by deep learning and This allows us to obtain more comprehensive reference information on indicators.
[0094] 205. After fusing the first event indicator data and the second event indicator data, combine them with the second structural data to obtain the event profile of the event description text.
[0095] Specifically, the extracted structured information, the metrics obtained from the machine learning metric extractor, and the metrics obtained from the deep learning metric extractor are finally input into the profile generator, denoted as... These are combined to form the final event profile, denoted as... .
[0096]
[0097] This event profiling method leverages a generative model's ability to understand text, extracting key information from the text based on requirements to form structured data, thus converting unstructured data into structured data. A bias corrector then validates the extracted data and supplements missing data from the original text, resulting in highly accurate and complete structured data. Based on this, a machine learning model processes the structured data to obtain highly accurate predictive indicators. Finally, deep learning is used to summarize existing indicators, resulting in highly usable indicators. The final structured data and various indicators are then arranged to generate the final event profiling. This method not only provides structured profiling information, numerical indicators, and categorical indicators but also improves accuracy and can supplement missing data, thus having a wider range of applications.
[0098] In another embodiment, this application also provides a method for training a bias corrector, a machine learning model, and a deep learning model, wherein the training process of the bias corrector may specifically include:
[0099] Step 11: Input the fourth structural data output by the generated model and the event description text into the correction device to obtain the fifth structural data.
[0100] Step 12: Determine the first loss value between the fifth structural data and the original structural data based on the first loss function, and after processing the first loss value based on the gradient descent algorithm, update the parameters of the bias corrector. The original structural data is the structural data corresponding to the pre-set event description text.
[0101] For details, please refer to Figure 4 As shown, the training data during the error correction phase may include: event description text. The corresponding structured data is denoted as and the structured data extracted by the information extractor. .Will and Input to the correction device In neural networks, structured data predicted by the bias corrector is obtained. Then calculate using the loss function and The loss value between the two values is used, and then gradient descent is applied to gradually update the value. The parameters in.
[0102] The training process for a machine learning model includes:
[0103] Step 21: Input the sixth structure data and event description text output by the corrector into the machine learning model to obtain the third event index data.
[0104] Step 22: Determine the second loss value between the third event indicator data and the original event indicator data based on the second loss function, and update the parameters of the machine learning model after processing the second loss value based on the gradient descent algorithm. The original event indicator data is the event indicator data corresponding to the pre-set event description text.
[0105] Specifically, refer to Figure 4 As shown, the machine learning extractor uses [this method] during the training phase. The actual probability of an event occurring Severity of the incident and the output of the correction device As training data, and Input to In the middle, we get Calculated through loss function and The loss value between the two is updated using gradient descent. The parameters in [the dataset]. These can be adjusted through gradient ascent. near .
[0106] The training process of a deep learning model includes:
[0107] Step 31: Input the seventh structural data output by the corrector, the event description text, and the fourth event index data output by the machine learning model into the deep learning model to obtain the fifth event index data.
[0108] Step 32: Determine the third loss value between the fifth event index data and the original event index data based on the third loss function, and update the parameters of the deep learning model after processing the third loss value based on the gradient descent algorithm. The original event index data is the event index data corresponding to the pre-set event description text.
[0109] Specifically, refer to Figure 4 As shown, the deep learning extractor uses [this method] during the training phase. The actual probability of an event occurring Severity of the incident and the output of the correction device Output of machine learning extractor As training data, , and The input is fed into the deep learning extractor to obtain... Calculated through loss function and The loss value between the two is updated using gradient descent. The parameters in.
[0110] As a specific application of the above-mentioned event profiling generation method, taking the input event description text as: "On January 1, 2030, Company XX experienced some machine malfunctions due to an earthquake; the event category is natural disaster, and the field is IT industry," the extracted template text would be: "Event Field, Event Time, Event Category, and Event Impact Scope." For example:
[0111] Extract structured information from the input:
[0112]
[0113] Input the event description text "On January 1, 2030, Company XX experienced some machine malfunctions due to an earthquake. The event category is natural disaster, and the field is IT industry" and the structured information "IT industry", "2030-01-01", "natural disaster", and "not found" into the correction device. The predicted structured information obtained includes "IT", "2030-01-02", "natural disasters", and "enterprise level", among which... The randomness of the generated results led to inaccuracies in the extracted structured data for "IT" and "2030-01-02". The specific value for the event's impact scope, "enterprise level," was successfully completed. Then, the logic analyzer was used. The structured data is evaluated sequentially. Since "IT industry," "2030-01-01," and "natural disasters" all exist in the event text, they are retained. However, the scope of the event's impact does not exist in the event text, so it is discarded. The scope of the generated event impact is defined as "enterprise level". The final structured data obtained is: "IT industry", "2030-01-01", "natural disaster", "enterprise level".
[0114] Then, the keywords "IT industry", "2030-01-01", "natural disasters", and "enterprise level" are input into the machine learning model. The probability of the event occurring is 0.6, and the severity of the event is 0.3. The event text, structured data, event probability, and event severity are then input into the deep learning model. The probability of the event occurring is 0.8, and the severity of the event is 0.3.
[0115] Finally, the structured information—the probability of events and the severity of events extracted by the machine learning metric extractor and the deep learning metric extractor—is summarized and input into... The event profile obtained is: "On January 1, 2030, a natural disaster occurred in the IT industry, affecting enterprises. The predicted probability of the event is 0.6, with a reference value of 0.8. The predicted severity of the event is 0.3, with a reference value of 0.3."
[0116] This event profiling generation method addresses the shortcomings of existing structured data extraction methods, such as named entity recognition, which suffer from high training costs, low versatility, and insufficient accuracy in handling diverse data. It employs a generative model for structured information extraction, combining a designed extraction template text to obtain the necessary information. Furthermore, to ensure the generative model possesses domain-specific knowledge, a bias corrector is used to supplement the extracted structured information, resulting in highly accurate and complete structured data. This eliminates the need for fine-tuning the generative model, and the bias corrector predicts and supplements missing information from the original text.
[0117] By extracting structured information, a machine learning indicator extractor is used to obtain a highly reliable predictive indicator. For structured data, traditional machine learning methods often produce more accurate predictions than deep learning, resulting in a more reliable indicator. However, traditional machine learning methods cannot directly process unstructured data and may miss some key information. Therefore, the original event text, extracted structured information, and indicator values extracted by traditional machine learning are combined as input to a deep learning model to obtain a comprehensive indicator value for reference. Finally, the information obtained from each step is summarized to generate an event profile, improving the accuracy, reliability, and richness of the indicators in the event profile.
[0118] The above describes an event profile generation method provided by the embodiments of this application. The following describes the apparatus for performing the above event profile generation method.
[0119] Please see Figure 5 , Figure 5 This is a schematic diagram of an event profile generation device provided in an embodiment of this application. Figure 5 As shown, the event profile generation device includes:
[0120] The structural data extraction module 501 is used to extract structural data from the event description text based on the generative model to obtain the first structural data.
[0121] The structural data correction module 502 is used to extract structural data from the event description text based on the correction device, and to fuse the extraction result with the first structural data to obtain the second structural data.
[0122] The first indicator extraction module 503 is used to input the second structure data into the machine learning model to extract event indicators and obtain the first event indicator data.
[0123] The second indicator extraction module 504 is used to input the event description text, the second structure data, and the first event indicator data into a deep learning model to extract event indicators, thereby obtaining the second event indicator data; and...
[0124] The event profile generation module 505 is used to merge the first event indicator data and the second event indicator data and combine them with the second structural data to obtain an event profile of the event description text.
[0125] In one possible implementation, the process by which the structure data extraction module 501 extracts structure data from the event description text based on the generative model to obtain the first structure data includes:
[0126] The extraction template and event description text are input into the generative model, which then extracts the structural data from the event description text according to the structural data items contained in the extraction template, thus obtaining the first structural data.
[0127] In one possible implementation, the bias corrector includes: a first neural network model and a logic analyzer. The structure data correction module 502 extracts structure data from the event description text based on the bias corrector, and fuses the extracted results with the first structure data to obtain second structure data. This process includes:
[0128] The first structural data and the event description text are input into the first neural network model to extract structural data, resulting in the third structural data.
[0129] The third structure data, the first structure data, and the event description text are input into the logic analyzer. The third structure data and the first structure data are then merged to obtain the second structure data.
[0130] In one possible implementation, the structure data correction module 502 inputs the third structure data, the first structure data, and the event description text into the logic analyzer, and merges the third structure data and the first structure data to obtain the second structure data, including:
[0131] The logic analyzer determines whether each structure data value in the first structure data is in the event description text, and determines that the first target structure data value does not exist in the event description text;
[0132] Replace the first target structure data value with the second target structure data value in the third structure data and add it to the first structure data. The second target structure data value and the first target structure data value correspond to the same structure data item.
[0133] In one possible implementation, the training process of the bias corrector in the structured data correction module 502 includes:
[0134] The fourth structural data and event description text output by the generated model are input into the bias corrector to obtain the fifth structural data;
[0135] The first loss value between the fifth structural data and the original structural data is determined based on the first loss function. After processing the first loss value based on the gradient descent algorithm, the parameters of the bias corrector are updated. The original structural data is the structural data corresponding to the pre-set event description text.
[0136] In one possible implementation, the training process of the machine learning model in the first indicator extraction module 503 includes:
[0137] The sixth structure data and event description text output by the corrector are input into the machine learning model to obtain the third event index data.
[0138] The second loss value between the third event index data and the original event index data is determined based on the second loss function. After processing the second loss value based on the gradient descent algorithm, the parameters of the machine learning model are updated. The original event index data is the event index data corresponding to the pre-set event description text.
[0139] In one possible implementation, the training process of the deep learning model in the second metric extraction module 504 includes:
[0140] The seventh structural data output by the corrector, the event description text, and the fourth event index data output by the machine learning model are input into the deep learning model to obtain the fifth event index data.
[0141] The third loss value between the fifth event index data and the original event index data is determined based on the third loss function. After processing the third loss value based on the gradient descent algorithm, the parameters of the deep learning model are updated. The original event index data is the event index data corresponding to the pre-set event description text.
[0142] This application also provides an electronic device in its embodiments. (See reference...) Figure 6 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this application. The electronic device in the embodiments of this application may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 6 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.
[0143] like Figure 6 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. When the electronic device is powered on, the RAM 603 also stores various programs and data required for the operation of the electronic device. The processing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.
[0144] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, memory cards, hard drives, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.
[0145] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the event profile generation methods provided in this application.
[0146] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the event profile generation methods provided in this application.
[0147] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.
[0148] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.
[0149] In the above embodiments, the implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product.
[0150] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).
Claims
1. A method for generating event profiles, characterized in that, include: Based on the generative model, structural data is extracted from the event description text to obtain the first structural data; The structural data of the event description text is extracted based on the correction device, and the extraction result is fused with the first structural data to obtain the second structural data; The second structured data is input into a machine learning model to extract event metrics, thereby obtaining the first event metric data. The event description text, the second structural data, and the first event indicator data are input into a deep learning model to extract the event indicators and obtain the second event indicator data. The first event indicator data and the second event indicator data are fused together and combined with the second structure data to obtain the event profile of the event description text.
2. The event profile generation method according to claim 1, characterized in that, The first structured data is obtained by extracting structural data from the event description text based on the generative model, including: The extraction template and the event description text are input into the generation model, so that the generation model extracts the structural data from the event description text according to the structural data items contained in the extraction template, and obtains the first structural data.
3. The event profile generation method according to claim 1, characterized in that, The correction device includes a first neural network model and a logic analyzer. The correction device is used to extract structural data from the event description text, and the extracted results are fused with the first structural data to obtain second structural data, including: The first structural data and the event description text are input into the first neural network model to extract the structural data, thereby obtaining the third structural data. The third structural data, the first structural data, and the event description text are input into the logic analyzer, and the third structural data and the first structural data are fused to obtain the second structural data.
4. The event profile generation method according to claim 3, characterized in that, The step of inputting the third structural data, the first structural data, and the event description text into the logic analyzer, and fusing the third structural data and the first structural data to obtain the second structural data includes: The logic analyzer determines whether each structural data value in the first structural data is in the event description text, and determines that the first target structural data value does not exist in the event description text; The second target structure data value in the third structure data replaces the first target structure data value and is added to the first structure data. The second target structure data value and the first target structure data value correspond to the same structure data item.
5. The event profile generation method according to any one of claims 1 to 4, characterized in that, The training process of the correction device includes: The fourth structural data output by the generative model and the event description text are input into the correction device to obtain the fifth structural data; The first loss value between the fifth structural data and the original structural data is determined based on the first loss function. After processing the first loss value based on the gradient descent algorithm, the parameters of the bias corrector are updated. The original structural data is the structural data corresponding to the pre-set event description text.
6. The event profile generation method according to any one of claims 1 to 4, characterized in that, The training process of the machine learning model includes: The sixth structural data output by the corrector and the event description text are input into the machine learning model to obtain the third event index data; The second loss value between the third event indicator data and the original event indicator data is determined based on the second loss function. After processing the second loss value based on the gradient descent algorithm, the parameters of the machine learning model are updated. The original event indicator data is the event indicator data corresponding to the pre-set event description text.
7. The event profile generation method according to any one of claims 1 to 4, characterized in that, The training process of the deep learning model includes: The seventh structural data output by the corrector, the event description text, and the fourth event index data output by the machine learning model are input into the deep learning model to obtain the fifth event index data. The third loss value between the fifth event index data and the original event index data is determined based on the third loss function. After processing the third loss value based on the gradient descent algorithm, the parameters of the deep learning model are updated. The original event index data is the event index data corresponding to the pre-set event description text.
8. An event profile generation device, characterized in that, include: The structural data extraction module is used to extract structural data from the event description text based on the generative model to obtain the first structural data. The structural data correction module is used to extract the structural data from the event description text based on the correction device, and fuse the extraction result with the first structural data to obtain the second structural data; The first indicator extraction module is used to input the second structured data into the machine learning model to extract event indicators and obtain the first event indicator data. The second indicator extraction module is used to input the event description text, the second structural data and the first event indicator data into a deep learning model to extract the event indicator and obtain the second event indicator data. as well as, The event profile generation module is used to fuse the first event indicator data and the second event indicator data and combine them with the second structure data to obtain the event profile of the event description text.
9. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program to enable the electronic device to implement the event profile generation method as described in any one of claims 1 to 7.
10. A computer storage medium, characterized in that, The storage medium carries one or more computer programs that, when executed by an electronic device, enable the electronic device to implement the event profile generation method as described in any one of claims 1 to 7.