Intelligence analysis methods, devices, media, and electronic equipment

Abstract

The present invention proposes an intelligence analysis method, apparatus, medium, and electronic device. The method comprises: obtaining intelligence source information as an event extraction dataset, constructing question-and-answer training samples, and generating a to-be-learned course for question-and-answer model training in a question-and-answer system; inputting the current learning state value of the question-and-answer model into a course learning progress selection strategy network to dynamically select a learning progress and output a target question-and-answer training sample to be learned; inputting the target question-and-answer training sample to be learned into the question-and-answer model for training and updating the learning state value of the question-and-answer model; and using the trained question-and-answer model to extract the target intelligence information from the intelligence source information. This method, based on the question-and-answer system's analysis of the intelligence source information, dynamically selects different learning progresses at different learning stages based on the system's own learning state during the learning process, thereby improving the accuracy of intelligence analysis.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to an intelligence analysis method, device, medium, and electronic equipment. Background Art

[0002] Intelligence analysis mainly involves using machine learning tools with intelligent or non-intelligent algorithms to conduct statistical analysis on samples of acquired intelligence source information, and through screening and judgment, to find valuable event information from big data as target intelligence information, and further transform it into actionable decision-making and command capabilities.

[0003] Event analysis based on question-answering systems is an information processing method that has emerged in recent years. It is suitable for intelligence analysis and processing. It transforms events into a series of questions, with each event role being predefined as a question sentence. Answers are then generated based on the input text. The responses to the questions for event-related roles are then integrated to complete the event extraction task. For example, the MQAEE algorithm views event extraction as a multi-round question-answering process, transforming event extraction into three types of problems: trigger word recognition, trigger word classification, and event element extraction. It uses the BERT architecture for encoding, taking a sentence as the first sentence and the original text as the second sentence as input, and outputting the extracted event information. The EEasMRC algorithm first performs conventional event trigger word extraction and then generates event element question templates in an unsupervised manner based on the event type. It uses the BERT architecture for the question-answering model. However, most of these question-answering systems tend to focus on how to generate event template associations. They simply transfer advanced model architectures or pre-trained models from question-answering systems to the question-answering model framework, without effectively training models tailored to the unique characteristics of intelligence information extraction, particularly improving model learning strategies.

[0004] The present invention considers providing a question-answering learning method based on a policy network to dynamically adjust the course learning progress, which is suitable for the analysis of intelligence events. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention proposes an intelligence analysis method, device, medium, and electronic device. Through the question-answering system constructed by the present invention, different learning progress can be dynamically selected at different learning stages according to the learning status of the system, and effective training can be carried out on intelligence information, thereby improving the accuracy of intelligence analysis.

[0006] In order to achieve the above object, the present invention provides an intelligence analysis method.

[0007] The cloud cluster constructs a question-answering system to analyze the intelligence source information collected by the perception end to obtain target intelligence information; and automatically pushes the target intelligence information to different authorized user ends according to the subscription / distribution relationship established by the user;

[0008] The constructed question-answering system analyzes the intelligence source information collected by the sensing end to obtain target intelligence information, including:

[0009] The step of generating courses to be learned: obtaining the intelligence source information as an event extraction data set, constructing question-answering training samples, and generating courses to be learned for question-answering model training in the question-answering system;

[0010] Course learning step: inputting the learning state value of the current question-answering model into the course learning progress selection strategy network to dynamically select the learning progress and output the target question-answering training sample to be learned;

[0011] Question-answering model training step: inputting the target question-answering training sample to be learned into the question-answering model for training, and updating the learning state value of the question-answering model;

[0012] The trained question-answering model is used to extract the target intelligence information from the intelligence source information.

[0013] Optionally, the step of generating the courses to be studied further includes:

[0014] Converting the event extraction data set into a plurality of predefined questions, each of the questions corresponding to an event parameter role to be extracted;

[0015] According to the division of different granularities from extracting one parameter to multiple parameters, and supplementing the remaining parameter information with the corresponding question during each parameter extraction, each event extraction sample is constructed, and multiple samples constitute the question-answering training sample;

[0016] Determining the complexity of each of the samples based on the number of questions that need to be answered and the frequency of occurrence of the event parameter roles to be extracted;

[0017] The question-answering training samples are sorted according to the complexity of each of the samples to generate the courses to be learned.

[0018] Optionally, the question-answering training samples are sorted from low to high according to the complexity of each sample to generate the course to be learned.

[0019] Optionally, multiple features during the training process are used to characterize the learning state value, and the features include the current number of iteration steps, the sliding average of the loss function value during training, the loss function value of the current sample, and the distribution margin value and validation set score output by the question-answering model.

[0020] Optionally, the course learning steps further include:

[0021] At each iteration step of the question-answering model, obtaining the learning state value of the current question-answering model as the current state of the policy network;

[0022] Dynamically select different learning progress according to the learning state value;

[0023] The question-answering training samples are sampled according to the selected learning progress to obtain the target question-answering training samples to be learned.

[0024] Optionally, dynamically selecting different learning progresses according to the learning state value includes:

[0025] Each decision action of the policy network is defined to select a different learning progress;

[0026] According to the learning state value, a decision action is sampled using the policy network, and the selected learning progress is determined by the sampled decision action.

[0027] Optionally, the method further includes:

[0028] The policy network is updated according to the updated learning state value of the question-answering model.

[0029] Optionally, updating the policy network according to the updated learning state value of the question-answering model includes:

[0030] Acquire the learning state value of the question-answering model in real time, generate the decision action according to the learning state value, and select the learning progress of the next iteration step of the question-answering model training;

[0031] Obtaining the reward value returned by the question-answering model;

[0032] According to the reward value returned by the policy network when executing a certain decision action, the reward value is maximized using an optimization algorithm to output the optimal policy network.

[0033] Optionally, obtaining the reward value returned by the question-answering model includes:

[0034] After each verification set score is returned after the question-answering model is trained, the difference between two consecutive verification set scores is used as the reward value.

[0035] Another aspect of the present invention provides an intelligence analysis device, comprising:

[0036] To-be-learned course generation module: used to obtain the intelligence source information as an event extraction data set, construct question-answering training samples, and generate to-be-learned courses for question-answering model training in the question-answering system;

[0037] Course learning module: used to input the learning state value of the current question-answering model into the course learning progress selection strategy network to dynamically select the learning progress and output the target question-answering training samples to be learned;

[0038] Question-answering model training module: inputs the target question-answering training sample to be learned into the question-answering model for training, and updates the learning state value of the question-answering model;

[0039] The trained question-answering model is used to extract the target intelligence information from the intelligence source information.

[0040] Another aspect of the present invention provides a storage medium for storing a computer program for executing the above-mentioned intelligence analysis method.

[0041] Another aspect of the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned intelligence analysis method when executing the computer program.

[0042] From the above scheme, it can be seen that the advantages of the present invention are:

[0043] The method provided by the present invention is based on the analysis of intelligence source information collected by the perception end by the question-answering system to obtain target intelligence information. Specifically, the method obtains the intelligence source information as an event extraction data set, constructs question-answering training samples, and generates a course to be learned for question-answering model training in the question-answering system; then, the learning state value of the current question-answering model is input into the course learning progress selection strategy network to dynamically select the learning progress and output the target question-answering training samples to be learned; the target question-answering training samples to be learned are input into the question-answering model for training; and the trained question-answering model is used to extract the target intelligence information from the intelligence source information. The question-answering system constructed by this method can simulate the way humans learn event information from simple question-answering to complex question-answering, so that different learning progresses are dynamically selected at different learning stages according to their own learning state during the learning process. It can effectively train intelligence information, improve the learning efficiency of the model and the quality of intelligence event extraction, and improve the accuracy of intelligence analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a framework diagram of the intelligence analysis system of the present invention;

[0045] Figure 2 A schematic diagram of the flow of the intelligence analysis method provided in Example 1 of the present invention;

[0046] Figure 3 is a specific flow chart of step S1;

[0047] Figure 4 is a specific flow chart of step S2;

[0048] Figure 5 A schematic diagram of the intelligence analysis method provided in Example 2 of the present invention;

[0049] Figure 6 is a specific flow chart of step S4;

[0050] Figure 7 This is a framework diagram of the intelligence analysis device of the present invention;

[0051] Figure 8 A schematic diagram of the structure of an electronic device;

[0052] Figure 9 A schematic diagram of the hardware structure of an electronic device;

[0053] in:

[0054] 101-sensing end;

[0055] 102-Image intelligence analysis server;

[0056] 103-Message intelligence analysis server;

[0057] 104-Open source intelligence analysis server;

[0058] 105-Electromagnetic intelligence analysis server;

[0059] 106-Intelligence data server;

[0060] 107-distributed storage node;

[0061] 108-Cloud cluster end;

[0062] 109-Authorized user terminal;

[0063] 400-Intelligence Analysis Device;

[0064] 401-Course generation module to be learned;

[0065] 402-Course Learning Module;

[0066] 403-Question-answering model training module;

[0067] 404-Policy Network Update Module;

[0068] 500-electronic equipment;

[0069] 501-processor;

[0070] 502-Memory;

[0071] 600-Electronic equipment;

[0072] 601-RF unit;

[0073] 602-network module;

[0074] 603-audio output unit;

[0075] 604-input unit;

[0076] 6041-graphics processor;

[0077] 6042-Microphone;

[0078] 605-Sensor;

[0079] 606-display unit;

[0080] 6061-display panel;

[0081] 607-user input unit;

[0082] 6071-touch panel;

[0083] 6072-Other input devices;

[0084] 608-interface unit;

[0085] 609-Memory;

[0086] 610-processor. DETAILED DESCRIPTION

[0087] In order to make the above features and effects of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings.

[0088] The intelligent intelligence analysis system is a human-machine hybrid big data in-depth analysis application platform for intelligent intelligence analysis. By correlating and integrating multiple heterogeneous big data, it realizes human-machine collaborative target search, correlation analysis, spatiotemporal analysis, and intelligent reporting. It can be widely used in the intelligent analysis and mining of scientific and technological intelligence, industry intelligence, etc. Figure 11 is a system architecture diagram of the intelligence analysis system provided in an embodiment of the present application. The system adopts a service-oriented distributed architecture and is mainly composed of a sensing end 101, an image intelligence analysis server 102, a message intelligence analysis server 103, an open source intelligence analysis server 104, an electromagnetic intelligence analysis server 105, an intelligence data server 106, a distributed storage node 107, a cloud cluster end 108, and an authorized user end 109. The image intelligence analysis server 102, the message intelligence analysis server 103, the open source intelligence analysis server 104, the electromagnetic intelligence analysis server 105, and the intelligence data server 106 are simultaneously connected to the sensing end 101 and the distributed storage node 107, and the distributed storage node 107 is connected to the cloud cluster end 108 to achieve information exchange.

[0089] The system's sensor terminal 101 dynamically accesses intelligence data from various sources and types, including early warning radar, reconnaissance satellites, messages, graphics, images, video, and electromagnetic signals. The system automatically distributes the intelligence sources collected by the sensor terminal 101 to the image intelligence analysis server 102, message intelligence analysis server 103, open source intelligence analysis server 104, electromagnetic intelligence analysis server 105, and intelligence data server 106, based on intelligence type and source, for data analysis. Relying on a service registration mechanism, the system implements unified classification and management, forming a software service resource directory and enabling intelligence analysis functionality to be shared across all nodes within the system. Furthermore, the system utilizes distributed operational management technology. The cloud cluster terminal 108, through distributed storage nodes 107, centrally manages intelligence analysis tasks, computing power, and data resources. This allows for coordinated scheduling based on intelligence support requirements, enabling collaborative completion of intelligence analysis tasks and generating targeted intelligence products. These targeted intelligence products are then automatically pushed to authorized user terminals 109, based on user-defined subscription / distribution relationships.

[0090] Based on the above-mentioned intelligence analysis system, an embodiment of the present invention proposes an intelligence analysis method based on dynamic course question-answering learning. The method can simulate the learning event information mode of humans transitioning from simple questions and answers to complex questions and answers, so that the question-answering model used in intelligence analysis can dynamically select different learning progress at different learning stages according to its own learning status during the learning process, so as to better train the model, improve the learning efficiency of the model and the quality of event extraction, and improve the accuracy of intelligence analysis.

[0091] Specifically, refer to Figure 2 As shown, Figure 2 A schematic diagram of the flow of the intelligence analysis method provided in Example 1 of the present invention is shown.

[0092] An intelligence analysis method, wherein a cloud cluster constructs a question-answering system to analyze the intelligence source information collected by the sensing end to obtain target intelligence information; and automatically pushes the target intelligence information to different authorized user ends according to the subscription / distribution relationship established by the user;

[0093] The analysis of the intelligence source information collected by the sensing end to obtain target intelligence information includes:

[0094] S1. Obtain the intelligence source information collected by the perception end as the event extraction data set, construct question-answering training samples, and generate the courses to be learned for question-answering model training in the question-answering system.

[0095] In some embodiments, as Figure 3 As shown, Figure 3 The specific flow chart of step S1 is shown, which specifically includes:

[0096] S11. Convert the event extraction data set into a plurality of predefined questions, each of which corresponds to an event parameter role to be extracted.

[0097] In specific implementations, the intelligence source information collected by the perception end is generally intelligence data from multiple sources and multiple types, such as text information from open source intelligence sources such as newspapers and websites. The event extraction data set can be converted into pre-defined questions based on the trigger words, trigger word categories, and roles set according to the intelligence type and source to construct a question-answering model.

[0098] S12. Extract one parameter to multiple parameters with different granularities, and add the remaining parameter information to the corresponding question each time the parameter is extracted, to construct each event extraction sample, and multiple samples constitute the question-answering training sample.

[0099] In the specific implementation, this embodiment constructs the event extraction data set by first learning one parameter extraction, then learning multiple parameter extractions, and finally learning the complete parameter extraction. Taking into account the dependency between event parameters, when constructing each event extraction sample, the remaining parameter information in the sample is supplemented to the problem, providing richer information for model learning.

[0100] S13. Determine the complexity of each of the examples based on the number of questions that need to be answered and the frequency of occurrence of the event parameter roles to be extracted.

[0101] Specifically, when defining sample complexity, a sample is decomposed into samples ranging from those that require the extraction of only one event parameter to those that require the extraction of multiple event parameters. The sample complexity is determined based on the number of questions that need to be answered and the frequency of occurrence of the event parameter roles to be extracted. The more diverse the number of question-and-answer questions, the more complex the sample; the less frequent the occurrence of the parameter roles to be extracted, the more complex the sample. After decomposing the sample, the parameter information that does not need to be extracted in each sample is supplemented into the question sentence to enrich the question sentence. In this way, gradual learning can be achieved from extracting only one event parameter to extracting multiple event parameters, and ultimately learning how to extract all parameter information.

[0102] S14. Sort the question-answering training samples according to the complexity of each of the samples to generate the courses to be learned.

[0103] In a specific implementation, in this embodiment, the question-answering training samples are sorted from low to high according to the complexity of each sample to generate the course to be learned. Of course, the method of sorting the question-answering training samples according to the complexity of each sample is not limited to this method, and other related methods are also within the scope of protection of this application, and this application does not make specific limitations on this.

[0104] In this embodiment, when constructing the course problems to be learned, the interdependence between event parameter information is taken into consideration and the relevant parameter information is supplemented. At the same time, the complexity of the sample is determined based on the amount of supplementary parameter information and the frequency of occurrence of the parameter roles to be extracted, thereby simulating the human learning process to construct learning examples from simple to complex.

[0105] S2. Input the learning state value of the current question-answering model into the course learning progress selection strategy network to dynamically select the learning progress and output the target question-answering training sample to be learned.

[0106] In a specific implementation, multiple features in the training process are used to characterize the learning state value, including the current number of iteration steps, the sliding average of the loss function value during training, the loss function value of the current sample, and the distribution margin value output by the question-answering model, the verification set score, etc.

[0107] In some embodiments, as Figure 4 As shown, Figure 4 The specific flow chart of step S2 is shown, which specifically includes:

[0108] S21. At each iteration step of the question-answering model, obtaining a learning state value of the current question-answering model as a current state of the policy network;

[0109] S22. Dynamically select different learning progress according to the learning state value;

[0110] In a specific implementation, dynamically selecting different learning progresses according to the learning state value includes:

[0111] Each decision action of the policy network is defined to select a different learning progress;

[0112] According to the learning state value, a decision action is sampled using the policy network, and the selected learning progress is determined by the sampled decision action.

[0113] In the specific implementation, three learning mechanisms can be selected for learning progress: forward, backward, and unchanged. These respectively imitate the human learning mechanism of leapfrogging, progress reversal, and review. The learning progress can be dynamically selected, and the flexible learning method avoids the conventional rigid and inefficient learning from beginning to end.

[0114] S23. Sampling the question-answering training samples according to the selected learning progress to obtain the target question-answering training samples.

[0115] In this embodiment, during the course learning process, a batch of data to be learned in the next iterative step can be dynamically selected according to the learning status of the current question-answering model, which is in line with the process of human language learning, can improve the learning efficiency of the model and the effect of event extraction, and improve the accuracy of intelligence analysis.

[0116] S3. Input the target question-answering training sample to be learned into the question-answering model for training, and update the learning state value of the question-answering model.

[0117] The trained question-answering model is used to extract the target intelligence information from the intelligence source information.

[0118] The question-answering model training is completed through iterative training based on the target question-answering training samples selected for course learning. In each iterative cycle, a batch of target question-answering training samples selected by the course progress selection strategy network are input to update the model parameters, output the target intelligence information, and update the learning status value of the question-answering model, including changes in the loss function value, model distribution, etc., and the model effect is regularly evaluated on the validation set.

[0119] In this embodiment, a cloud cluster constructs a question-and-answer system to analyze the intelligence source information collected by the sensing end to obtain target intelligence information. This target intelligence information is then automatically pushed to different authorized user ends according to the subscription / distribution relationships established by the user. Specifically, the intelligence source information is used as an event extraction dataset to construct question-and-answer training samples, generating courses to be learned in the question-and-answer system for training the question-and-answer model. The current learning state of the question-and-answer model is then input into the course learning progress selection strategy network to dynamically select a learning progress and output target question-and-answer training samples to be learned. These target question-and-answer training samples to be learned are then re-input into the question-and-answer model for training. The trained question-and-answer model is then used to extract the target intelligence information from the intelligence source information. The question-and-answer system constructed by this method can simulate the human learning event information transition from simple to complex question-and-answer. During the learning process, different learning progresses are dynamically selected at different learning stages based on the learning state. This allows for effective training of intelligence information, improving the model's learning efficiency, the quality of intelligence event extraction, and the accuracy of intelligence analysis.

[0120] refer to Figure 5 As shown, Figure 5 A schematic diagram of the flow of the intelligence analysis method provided in the second embodiment of the present invention is shown.

[0121] An intelligence analysis method comprising:

[0122] S1. Obtain intelligence source information as an event extraction dataset, construct question-answering training samples, and generate courses to be learned for question-answering model training in the question-answering system.

[0123] In some embodiments, as Figure 3 As shown, Figure 3 The specific flow chart of step S1 is shown, which specifically includes:

[0124] S11. Convert the event extraction data set into a plurality of predefined questions, each of which corresponds to an event parameter role to be extracted.

[0125] In a specific implementation, the event extraction data set can be converted into predefined questions through set trigger words, trigger word categories, and roles.

[0126] S12. Extract one parameter to multiple parameters with different granularities, and add the remaining parameter information to the corresponding question each time the parameter is extracted, to construct each event extraction sample, and multiple samples constitute the question-answering training sample.

[0127] In the specific implementation, this embodiment constructs the event extraction data set by first learning one parameter extraction, then learning multiple parameter extractions, and finally learning the complete parameter extraction. Taking into account the dependency between event parameters, when constructing each event extraction sample, the remaining parameter information in the sample is supplemented to the problem, providing richer information for model learning.

[0128] S13. Determine the complexity of each of the examples based on the number of questions that need to be answered and the frequency of occurrence of the event parameter roles to be extracted.

[0129] In this embodiment, the sample complexity is determined based on the number of questions to be answered and the frequency of occurrence of the event parameter role to be extracted. The more diverse the number of question-and-answer questions, the more complex the sample; the lower the frequency of occurrence of the parameter role to be extracted, the more complex the sample.

[0130] S14. Sort the question-answering training samples according to the complexity of each of the samples to generate the courses to be learned.

[0131] In a specific implementation, in this embodiment, the question-answering training samples are sorted from low to high according to the complexity of each sample to generate the course to be learned. Of course, the method of sorting the question-answering training samples according to the complexity of each sample is not limited to this method, and other related methods are also within the scope of protection of this application, and this application does not make specific limitations on this.

[0132] In this embodiment, when constructing the course problems to be learned, the interdependence between event parameter information is taken into consideration and the relevant parameter information is supplemented. At the same time, the complexity of the sample is determined based on the amount of supplementary parameter information and the frequency of occurrence of the parameter roles to be extracted, thereby simulating the human learning process to construct learning examples from simple to complex.

[0133] S2. Input the learning state value of the current question-answering model into the course learning progress selection strategy network to dynamically select the learning progress and output the target question-answering training sample to be learned.

[0134] In a specific implementation, multiple features in the training process are used to characterize the learning state value, including the current number of iteration steps, the sliding average of the loss function value during training, the loss function value of the current sample, and the distribution margin value and validation set score output by the question-answering model.

[0135] In some embodiments, as Figure 4 As shown, Figure 4 The specific flow chart of step S2 is shown, which specifically includes:

[0136] S21. At each iteration step of the question-answering model, obtaining a learning state value of the current question-answering model as a current state of the policy network;

[0137] S22. Dynamically select different learning progress according to the learning state value;

[0138] In a specific implementation, dynamically selecting different learning progresses according to the learning state value includes:

[0139] Each decision action of the policy network is defined to select a different learning progress;

[0140] According to the learning state value, a decision action is sampled using the policy network, and the selected learning progress is determined by the sampled decision action.

[0141] In the specific implementation, three learning mechanisms can be selected for learning progress: forward, backward, and unchanged. These respectively imitate the human learning mechanism of leapfrogging, progress reversal, and review. The learning progress can be dynamically selected, and the flexible learning method avoids the conventional rigid and inefficient learning from beginning to end.

[0142] S23. Sample the question-answering training samples according to the selected learning progress to obtain the target question-answering training samples to be learned, which are used for training the next iteration step of the question-answering model.

[0143] In this embodiment, during the course learning process, a batch of data to be learned in the next iterative step can be dynamically selected according to the learning status of the current question-answering model, which is in line with the process of human language learning, can improve the learning efficiency of the model and the effect of event extraction, and improve the accuracy of intelligence analysis.

[0144] S3. Input the target question-answering training sample to be learned into the question-answering model for training, and update the learning state value of the question-answering model.

[0145] The trained question-answering model is used to extract the target intelligence information from the intelligence source information.

[0146] In the specific implementation, each iteration step of the question-answering model is completed by iteratively training the target question-answering training samples selected based on the course learning in step S2. Each iteration cycle inputs a batch of target question-answering training samples selected by the course progress selection policy network, updates the model parameters, and obtains a trained question-answering model. The trained question-answering model is then used to extract the target intelligence information from the intelligence source information. At the same time, the current learning state is returned for course learning in step S2 and updating the policy network in step S4.

[0147] S4. Update the policy network according to the updated learning state value of the question-answering model.

[0148] In some embodiments, as Figure 6 As shown, Figure 6 The specific flow chart of step S4 is shown, which specifically includes:

[0149] The updating of the policy network according to the updated learning state value of the question-answering model includes:

[0150] S41. Acquire the learning state value of the question-answering model in real time, generate the decision action according to the learning state value, and select the learning progress of the next iteration step of the question-answering model training;

[0151] S42. Obtain the reward value returned by the question-answering model;

[0152] In a specific implementation, each time the question-answering model is trained, a validation set score is returned, and the difference between two consecutive validation set scores can be used as the reward value.

[0153] S43. According to the reward value returned by executing a certain decision action by the policy network, the reward value is maximized by using an optimization algorithm, and the optimal policy network is output.

[0154] In this embodiment, a policy network for selecting a course learning schedule can be constructed based on the reinforcement learning paradigm. The environment state is the learning state of the question-answering model, the decision actions are forward, backward, or unchanged learning schedule, and the reward value is the difference between two adjacent validation set scores of the question-answering model. The goal of policy network learning is to maximize the reward value, and the parameters of the policy network space are optimized using methods such as gradient descent to gradually learn an optimal course schedule selection network. This embodiment builds on the previous embodiment by iteratively updating the selected policy network based on the reward value returned by the question-answering generation model, enabling the model to effectively select training samples to be learned and accelerating model training.

[0155] To sum up, the intelligence analysis method provided by the present invention is based on the analysis of intelligence source information collected by the perception end by the question-answering system, and can simulate the learning event information method of humans transitioning from simple questions and answers to complex questions and answers, so that different learning progress can be dynamically selected at different learning stages according to their own learning status during the learning process. It can effectively train intelligence information, improve the learning efficiency of the model and the quality of intelligence event extraction, and improve the accuracy of intelligence analysis.

[0156] The above embodiments of the present invention can be applied to terminal devices with information analysis functions such as event extraction. The terminal devices can include personal terminals and host terminals, etc., and the embodiments of the present invention are not limited to this. The terminal can support operating systems such as Windows, Android, iOS, and Windows Phone.

[0157] Reference Figure 7 , Figure 7 An intelligence analysis device 400 is shown, which can be applied to a personal terminal and a host terminal device, and can be realized by Figure 1-Figure 5 The intelligence analysis method shown in the figure, the setting device provided in the embodiment of the present application can implement each process of the above-mentioned intelligence analysis method, including at least a to-be-learned course generation module 401, a course learning module 402, a question-answering model training module 403, and a policy network update module 404, namely, specifically:

[0158] An intelligence analysis device 400 includes:

[0159] To-be-learned course generation module 401: used to obtain intelligence source information as an event extraction data set, construct question-answering training samples, and generate to-be-learned courses for question-answering model training in the question-answering system;

[0160] Course learning module 402: used to input the learning state value of the current question-answering model into the course learning progress selection strategy network to dynamically select the learning progress and output the target question-answering training samples to be learned;

[0161] Question-answering model training module 403: used to input the target question-answering training sample to be learned into the question-answering model for training, and update the learning state value of the question-answering model;

[0162] The trained question-answering model is used to extract the target intelligence information from the intelligence source information.

[0163] Optionally, the to-be-learned course generation module 401 is further configured to:

[0164] Converting the event extraction data set into a plurality of predefined questions, each of the questions corresponding to an event parameter role to be extracted;

[0165] According to the division of different granularities from extracting one parameter to multiple parameters, and supplementing the remaining parameter information with the corresponding question during each parameter extraction, each event extraction sample is constructed, and multiple samples constitute the question-answering training sample;

[0166] Determining the complexity of each of the samples based on the number of questions that need to be answered and the frequency of occurrence of the event parameter roles to be extracted;

[0167] The question-answering training samples are sorted according to the complexity of each of the samples to generate the courses to be learned.

[0168] Optionally, the question-answering training samples are sorted from low to high according to the complexity of each sample to generate the course to be learned.

[0169] Optionally, multiple features during the training process are used to characterize the learning state value, and the features include the current number of iteration steps, the sliding average of the loss function value during training, the loss function value of the current sample, and the distribution margin value and validation set score output by the question-answering model.

[0170] Optionally, the course learning module 402 is further configured to:

[0171] At each iteration step of the question-answering model, obtaining the learning state value of the current question-answering model as the current state of the policy network;

[0172] Dynamically select different learning progress according to the learning state value;

[0173] The question-answering training samples are sampled according to the selected learning progress to obtain the target question-answering training samples to be learned.

[0174] Optionally, dynamically selecting different learning progresses according to the learning state value includes:

[0175] Each decision action of the policy network is defined to select a different learning progress;

[0176] According to the learning state value, a decision action is sampled using the policy network, and the selected learning progress is determined by the sampled decision action.

[0177] Optionally, the apparatus further includes a policy network update module 404, configured to:

[0178] The policy network is updated according to the updated learning state value of the question-answering model.

[0179] Optionally, the policy network update module 404 is further configured to:

[0180] Acquire the learning state value of the question-answering model in real time, generate the decision action according to the learning state value, and select the learning progress of the next iteration step of the question-answering model training;

[0181] Obtaining the reward value returned by the question-answering model;

[0182] According to the reward value returned by the policy network when executing a certain decision action, the reward value is maximized using an optimization algorithm to output the optimal policy network.

[0183] Optionally, obtaining the reward value returned by the question-answering model includes:

[0184] After each verification set score is returned after the question-answering model is trained, the difference between two consecutive verification set scores is used as the reward value.

[0185] Therefore, according to the intelligence analysis device 400 of the embodiment of the present application, based on the analysis of the intelligence source information collected by the perception end by the question-answering system, it can simulate the learning event information method of humans transitioning from simple questions and answers to complex questions and answers, so that in the learning process, different learning progress can be dynamically selected at different learning stages according to their own learning status, and effective training can be carried out on intelligence information, thereby improving the learning efficiency of the model and the quality of intelligence event extraction, and improving the accuracy of intelligence analysis.

[0186] It should be understood that the descriptions of the intelligence analysis method are also applicable to the intelligence analysis device 400 according to the embodiment of the present application. To avoid repetition, they will not be described in detail.

[0187] In addition, it should be understood that in the intelligence analysis device 300 according to the embodiment of the present application, the division of the above-mentioned functional modules is only used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the intelligence analysis device 300 can be divided into functional modules different from the modules illustrated above to complete all or part of the functions described above.

[0188] Figure 8 It is a structural diagram of an electronic device provided in an embodiment of the present application.

[0189] like Figure 8 As shown in , an embodiment of the present application also provides an electronic device 500, including a processor 501, a memory 502, and a program or instruction stored in the memory 502 and executable on the processor 501. When the program or instruction is executed by the processor 501, the steps of the above-mentioned intelligence analysis method are implemented and the same technical effect can be achieved.

[0190] Therefore, according to the electronic device 500 of the embodiment of the present application, based on the analysis of the intelligence source information collected by the perception end by the question-answering system, it can simulate the learning event information method of humans transitioning from simple questions and answers to complex questions and answers, so that in the learning process, different learning progress can be dynamically selected at different learning stages according to its own learning status, and effective training can be carried out on intelligence information, thereby improving the learning efficiency of the model and the quality of intelligence event extraction, and improving the accuracy of intelligence analysis.

[0191] To avoid repetition, other technical effects of the electronic device 500 according to the embodiment of the present application will not be described in detail here.

[0192] It should be noted that the electronic devices in the embodiments of the present application may include mobile electronic devices and non-mobile electronic devices.

[0193] Figure 6 This is a schematic diagram of the specific hardware structure of the electronic device provided in the embodiment of the present application.

[0194] Reference Figure 6 The electronic device 600 includes but is not limited to: a radio frequency unit 601, a network module 602, an audio output unit 603, an input unit 604, a sensor 605, a display unit 606, a user input unit 607, an interface unit 608, a memory 609, and a processor 610.

[0195] It should be understood that in the embodiments of the present application, the RF unit 601 may be used to receive and transmit signals during information transmission or calls. Specifically, it receives downlink data from the base station and transmits it to the processor 610 for processing; in addition, it transmits uplink data to the base station. Typically, the RF unit 601 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier, a duplexer, and the like. Furthermore, the RF unit 601 may communicate with the network and other devices via a wireless communication system.

[0196] The electronic device 600 provides users with wireless broadband Internet access through the network module 602, such as helping users to send and receive emails, browse web pages, and access streaming media.

[0197] The audio output unit 603 can convert audio data received by the RF unit 601 or the network module 602 or stored in the memory 609 into an audio signal and output it as sound. In addition, the audio output unit 603 can also provide audio output related to specific functions performed by the electronic device 600 (for example, call signal reception sound, message reception sound, etc.). The audio output unit 603 includes a speaker, a buzzer, a receiver, etc.

[0198] The input unit 604 is used to receive audio or video signals. It should be understood that in the embodiment of the present application, the input unit 604 may include a graphics processing unit (GPU) 6041 and a microphone 6042. The graphics processing unit 6041 processes image data of a still picture or video obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode.

[0199] The electronic device 600 also includes at least one sensor 605, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor includes an ambient light sensor and a proximity sensor, wherein the ambient light sensor can adjust the brightness of the display panel 6061 according to the brightness of the ambient light, and the proximity sensor can turn off the display panel 6061 and / or the backlight when the electronic device 600 is moved to the ear. As a type of motion sensor, the accelerometer sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used to identify the posture of the electronic device (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; the sensor 605 can also include a fingerprint sensor, a pressure sensor, an iris sensor, a molecular sensor, a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, etc., which will not be repeated here.

[0200] The display unit 606 is used to display information input by the user or information provided to the user. The display unit 606 may include a display panel 6061, which may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), or the like.

[0201] The user input unit 607 can be used to receive input digital or character information, and to generate key signal inputs related to the user settings and function control of the electronic device. Specifically, the user input unit 607 includes a touch panel 6071 and other input devices 6072. The touch panel 6071, also known as a touch screen, can collect user touch operations on or near it (such as operations performed by the user using any suitable object or accessory such as a finger, stylus, etc. on or near the touch panel 6071). The touch panel 6071 may include two parts: a touch detection device and a touch controller. Other input devices 6072 may include, but are not limited to, a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, and an operating stick, which will not be repeated here. The interface unit 608 is an interface for connecting an external device to the electronic device 600. For example, the external device may include a wired or wireless headset port, an external power supply (or battery charger) port, a wired or wireless data port, a memory card port, a port for connecting a device with an identification module, an audio input / output (I / O) port, a video I / O port, a headphone port, and the like. The interface unit 608 may be used to receive input (eg, data information, power, etc.) from an external device and transmit the received input to one or more elements within the electronic device 600 or may be used to transmit data between the electronic device 600 and an external device.

[0202] Memory 609 can be used to store software programs and various data. Memory 609 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function or an image playback function); the data storage area may store data generated based on the use of the mobile phone (such as audio data, a phone book, etc.). Furthermore, memory 609 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0203] The processor 610 is the control center of the electronic device. It connects the various components of the electronic device using various interfaces and lines. By running or executing software programs and / or modules stored in the memory 609 and accessing data stored in the memory 609, it performs various functions of the electronic device and processes data, thereby providing overall monitoring of the electronic device. The processor 610 may include one or more processing units. Preferably, the processor 610 may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interface, and application programs, and the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into the processor 610. Those skilled in the art will understand that the electronic device 600 may also include a power supply (such as a battery) to power the various components. The power supply may be logically connected to the processor 610 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. Figure 7 The electronic device structure shown in the figure does not constitute a limitation of the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or arrange the components differently, which will not be repeated here. In the embodiments of the present application, the electronic device includes but is not limited to mobile phones, tablet computers, laptop computers, PDAs, vehicle-mounted terminals, wearable devices (such as bracelets, glasses), and pedometers.

[0204] Specifically, the processor 610 is at least used to obtain the intelligence source information collected by the perception end as an event extraction data set, construct a question and answer training sample, and generate a course to be learned for question and answer model training in the question and answer system; input the learning state value of the current question and answer model into the course learning progress selection strategy network to dynamically select the learning progress, and output the target question and answer training sample to be learned; input the target question and answer training sample to be learned into the question and answer model for training, and update the learning state value of the question and answer model; the trained question and answer model is used to extract the target intelligence information from the intelligence source information.

[0205] Therefore, according to the electronic device 600 of the embodiment of the present application, based on the analysis of the intelligence source information collected by the perception end by the question-answering system, it can simulate the learning event information method of humans transitioning from simple questions and answers to complex questions and answers, so that in the learning process, different learning progress can be dynamically selected at different learning stages according to its own learning status, and effective training can be carried out on intelligence information, thereby improving the learning efficiency of the model and the quality of intelligence event extraction, and improving the accuracy of intelligence analysis.

[0206] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above-mentioned intelligence analysis method for intelligence analysis are implemented and the same technical effect can be achieved.

[0207] Therefore, according to the readable storage medium of the embodiment of the present application, based on the analysis of the intelligence source information collected by the perception end by the question-answering system, it is possible to simulate the learning event information method of humans transitioning from simple questions and answers to complex questions and answers, so that in the learning process, different learning progress can be dynamically selected at different learning stages according to one's own learning status, and effective training can be carried out on intelligence information, thereby improving the learning efficiency of the model and the quality of intelligence event extraction, and improving the accuracy of intelligence analysis.

[0208] To avoid repetition, other technical effects of the readable storage medium according to the embodiments of the present application are not described here.

[0209] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0210] An embodiment of the present application also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the steps of the above-mentioned intelligence analysis method for intelligence analysis, and can achieve the same technical effect.

[0211] Therefore, according to the chip of the embodiment of the present application, based on the analysis of the intelligence source information collected by the perception end by the question-answering system, it can simulate the learning event information method of humans transitioning from simple questions and answers to complex questions and answers, so that in the learning process, different learning progress can be dynamically selected at different learning stages according to their own learning status, and effective training can be carried out on intelligence information, thereby improving the learning efficiency of the model and the quality of intelligence event extraction, and improving the accuracy of intelligence analysis.

[0212] To avoid repetition, other technical effects of the chip according to the embodiments of the present application will not be described here.

[0213] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0214] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be applied, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0215] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0216] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. An intelligence analysis method, characterized in that: The cloud cluster constructs a question-answering system to analyze the intelligence source information collected by the perception end to obtain target intelligence information; and automatically pushes the target intelligence information to different authorized user ends according to the subscription / distribution relationship established by the user; The constructed question-answering system analyzes the intelligence source information collected by the sensing end to obtain target intelligence information, including: The step of generating courses to be learned is to obtain the intelligence source information as an event extraction data set, construct question-answering training samples, and generate courses to be learned for question-answering model training in the question-answering system, including: Converting the event extraction data set into a plurality of predefined questions, each of the questions corresponding to an event parameter role to be extracted; According to the division of different granularities from extracting one parameter to multiple parameters, and supplementing the remaining parameter information with the corresponding question during each parameter extraction, each event extraction sample is constructed, and multiple samples constitute the question-answering training sample; Determining the complexity of each of the samples based on the number of questions that need to be answered and the frequency of occurrence of the event parameter roles to be extracted; Sorting the question-answering training samples according to the complexity of each of the samples to generate the course to be learned; Course learning step: Input the learning state value of the current question-answering model into the course learning progress selection strategy network to dynamically select the learning progress and output the target question-answering training samples to be learned, including: At each iteration step of the question-answering model, obtaining the learning state value of the current question-answering model as the current state of the policy network; Each decision action of the policy network is defined to select a different learning progress; According to the learning state value, using the policy network to sample a decision action, and determining the selected learning progress through the sampled decision action; Sampling the question-answering training samples according to the selected learning progress to obtain the target question-answering training samples to be learned; Question-answering model training step: inputting the target question-answering training sample to be learned into the question-answering model for training, and updating the learning state value of the question-answering model; The trained question-answering model is used to extract the target intelligence information from the intelligence source information.

2. The method according to claim 1, characterized in that The question-answering training samples are sorted from low to high according to the complexity of each sample to generate the course to be learned.

3. The method according to claim 1, characterized in that The learning state value is characterized by using multiple features during the training process, including the current number of iterations, the sliding average of the loss function value during training, the loss function value of the current sample, and the distribution margin value and validation set score output by the question-answering model.

4. The method according to claim 1, wherein Also includes: The policy network is updated according to the updated learning state value of the question-answering model.

5. The method according to claim 4, characterized in that The updating of the policy network according to the updated learning state value of the question-answering model includes: Acquire the learning state value of the question-answering model in real time, generate the decision action according to the learning state value, and select the learning progress of the next iteration step of the question-answering model training; Obtaining the reward value returned by the question-answering model; According to the reward value returned by the policy network when executing a certain decision action, the reward value is maximized using an optimization algorithm to output the optimal policy network.

6. The method according to claim 5, characterized in that The obtaining of the reward value returned by the question-answering model includes: After each verification set score is returned after the question-answering model is trained, the difference between two consecutive verification set scores is used as the reward value.

7. An intelligence analysis device, characterized in that: include: The module for generating courses to be learned is used to obtain intelligence source information as an event extraction dataset, construct question-answering training samples, and generate courses to be learned for question-answering model training in the question-answering system. It includes: Converting the event extraction data set into a plurality of predefined questions, each of the questions corresponding to an event parameter role to be extracted; According to the division of different granularities from extracting one parameter to multiple parameters, and supplementing the remaining parameter information with the corresponding question during each parameter extraction, each event extraction sample is constructed, and multiple samples constitute the question-answering training sample; Determining the complexity of each of the samples based on the number of questions that need to be answered and the frequency of occurrence of the event parameter roles to be extracted; Sorting the question-answering training samples according to the complexity of each of the samples to generate the course to be learned; Course learning module: used to input the learning state value of the current question-answering model into the course learning progress selection strategy network to dynamically select the learning progress and output the target question-answering training samples to be learned, including: At each iteration step of the question-answering model, obtaining the learning state value of the current question-answering model as the current state of the policy network; Each decision action of the policy network is defined to select a different learning progress; According to the learning state value, using the policy network to sample a decision action, and determining the selected learning progress through the sampled decision action; Sampling the question-answering training samples according to the selected learning progress to obtain the target question-answering training samples to be learned; Question-answering model training module: inputs the target question-answering training sample to be learned into the question-answering model for training, and updates the learning state value of the question-answering model; The trained question-answering model is used to extract target intelligence information from the intelligence source information.

8. A storage medium, characterized in that: Used to store a computer program for executing the intelligence analysis method according to any one of claims 1 to 6.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the intelligence analysis method according to any one of claims 1 to 6 is implemented.

Images