Decision generation auxiliary method and device, equipment and storage medium

Abstract

The invention relates to the technical field of data processing, in particular to a decision generation assisting method, device and equipment and a storage medium, and the method comprises the steps: obtaining a real-time description voice packet, and obtaining voice text information according to the real-time description voice packet; performing content analysis on the voice text information based on a preset context analysis model to obtain logic text information; acquiring corresponding theoretical text information from a preset knowledge base according to the logic text information; inputting the logic text information and the theoretical text information into a preset decision model to obtain a responsibility judgment result; performing voice processing on the responsibility judgment result to obtain a voice decision result; according to the method, the decision result of the event can be quickly judged according to the recording record of the alarm event, so that the processing efficiency of the alarm event is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a decision making auxiliary method, device, equipment and storage medium. Background Art

[0002] When a police incident occurs, relevant police officers are usually required to go to the scene to handle it. Before the police officers make a judgment on the outcome of the police incident, they need to conduct a long and comprehensive analysis of the incident. Because the situation at the scene of the incident is complex and changeable, detailed investigation and analysis are required. After the police officers have collected all relevant information, they will comprehensively consider the results of responsibility determination based on laws and regulations and on-site evidence, resulting in lower overall handling efficiency of police incidents.

[0003] It can be seen that the existing technology still needs to be improved and enhanced. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a decision-making auxiliary method, device, equipment and storage medium, which can quickly determine the decision result of the event based on the recorded records of the police event to improve the processing efficiency of the police event.

[0005] The first aspect of the present invention provides a decision-making assistance method, including: obtaining a real-time description voice package, and obtaining voice text information based on the real-time description voice package; performing content analysis on the voice text information based on a preset context analysis model to obtain logical text information; obtaining corresponding theoretical text information from a preset knowledge base based on the logical text information; inputting the logical text information and the theoretical text information into a preset decision model to obtain a responsibility determination result; and performing voice processing on the responsibility determination result to obtain a voice decision result.

[0006] Optionally, in a first implementation method of the first aspect of the present invention, the real-time description voice package is obtained, and voice and text information is obtained based on the real-time description voice package, including: capturing the user's real-time voice signal in real time, and digitally converting the real-time voice signal to obtain a police recording data set; annotating the police data recording set according to the police scene to obtain a police annotated recording set; obtaining the network structure of the Deep Peak2 model, and optimizing the network structure to obtain a multi-scene speech recognition model; dividing the police annotated recording set into multiple segments of recorded audio, and denoising each segment of the recorded audio to obtain multiple segments of denoised audio; inputting the multiple segments of denoised audio into the multi-scene speech recognition model respectively to obtain multiple segments of recognized text information; splicing the multiple segments of recognized text information to obtain voice and text information.

[0007] Optionally, in a second implementation method of the first aspect of the present invention, the network structure of the Deep Peak2 model is obtained and the network structure is optimized to obtain a multi-scenario speech recognition model, including: obtaining the network structure of the DeepPeak2 model, and introducing a one-dimensional convolutional neural network and a long short-term memory network into the network structure to form a network optimization structure; introducing the Focal Loss algorithm into the weighted cross entropy loss function to obtain an optimized loss function, and introducing the optimized loss function into the network optimization structure to obtain an optimized basic model; performing time expansion and frequency conversion processing on a preset police recording training set to obtain a model training set; training the optimized basic model based on the cross-validation method and the model training set to obtain multiple training results, and optimizing each training result using a hyperparameter tuning method to obtain an optimized basic model with the best training results as the multi-scenario speech recognition model.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the content analysis of the speech and text information based on the preset context analysis model to obtain logical text information includes: obtaining police event record information from a preset police database, and marking the police event record information according to preset classification labels to obtain police classification record information; splitting the police classification record information into multiple word sets, and denoising each word set according to a preset stop word library to obtain a police word set; using the Word2Ve model to map each word in the police word set to a preset dimension. Vector space is used to obtain the alarm vector set; the Adam optimizer is introduced into the softmax algorithm to obtain the first input model; the second input model is constructed based on the SVR model; the first input model and the second input model are introduced into the preset multi-layer neural network to obtain the original context analysis model; the original context analysis model is trained based on the alarm vector set, and the classification score value of each training result is calculated using the confusion matrix, and the training model corresponding to the classification score value is used as the context analysis model; the voice and text information is input into the context analysis model to obtain logical text information.

[0009] Optionally, in a fourth implementation method of the first aspect of the present invention, the corresponding theoretical text information is obtained from a preset knowledge base based on the logical text information, including: obtaining the information collection link of the government affairs systems of each province and city, and obtaining the legal text data of the government affairs systems of each province and city through the information collection link; performing word segmentation processing on the legal text data to obtain legal word segmentation data; constructing a knowledge base based on the Elasticsearch search engine, NLP model and legal word segmentation data; performing intent analysis on the logical text information, extracting keywords from the logical text information to obtain a search term set; and inputting the search term set into the knowledge base for content retrieval to obtain theoretical text information.

[0010] Optionally, in a fifth implementation method of the first aspect of the present invention, the logical text information and the theoretical text information are input into a preset decision model to obtain a responsibility determination result, including: obtaining relevant data of historical police events, and extracting feature information related to preset police features from the relevant data of historical police events to obtain police feature information; extracting a target classification model, a target clustering model and a target prediction model from a preset model resource library according to the police feature information; forming a multi-model fusion decision model based on the target classification model, the target clustering model and the target prediction model; and inputting the logical text information and the theoretical text information into the decision model to obtain a responsibility determination result.

[0011] Optionally, in a sixth implementation method of the first aspect of the present invention, the responsibility determination result is voice-processed to obtain a voice decision result, including: collecting historical interaction data of the user, and performing content analysis on the historical interaction data to obtain personalized demand information; building a voice conversion model based on the personalized demand information and the TTS engine; and performing voice conversion on the responsibility determination result through the voice conversion model to obtain a voice decision result.

[0012] The second aspect of the present invention provides a decision-making auxiliary device, including: a recognition module, used to obtain a real-time description voice package, and obtain voice text information based on the real-time description voice package; an analysis module, used to perform content analysis on the voice text information based on a preset context analysis model to obtain logical text information; an acquisition module, used to obtain corresponding theoretical text information from a preset knowledge base based on the logical text information; a decision module, used to input the logical text information and the theoretical text information into a preset decision model to obtain a responsibility determination result; and a voice module, used to perform voice processing on the responsibility determination result to obtain a voice decision result.

[0013] The third aspect of the present invention provides a decision-making assistance device, which includes: a memory and at least one processor, wherein instructions are stored in the memory; at least one of the processors calls the instructions in the memory so that the decision-making assistance device performs each step of the decision-making assistance method described in any one of the above items.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed by a processor, the steps of any of the above-mentioned decision-making assistance methods are implemented.

[0015] In the technical solution of the present invention, by obtaining the real-time description voice package, the rapid and efficient collection of information is realized, and the timeliness and accuracy of the information are ensured. This process not only shortens the accident handling time, but also gains precious time for on-site emergency rescue; the context is analyzed by using a deep learning algorithm, which can deeply understand the implicit meaning and emotional color in the context, thereby providing a more accurate event analysis and responsibility determination basis; the corresponding theoretical text information is obtained from the preset knowledge base to provide a comprehensive theoretical basis for decision-making, including laws and regulations, case analysis and professional literature. The knowledge base stores a wealth of reference materials. These theoretical text information can provide a comprehensive reference for the system and help the system make both reasonable and legal decisions; the logical text information and theoretical text information are input into the preset decision model, which realizes intelligent and accurate responsibility determination, improves the accuracy and efficiency of decision-making, and can perform multi-dimensional analysis on complex situations, thereby obtaining the most reasonable responsibility determination result; the responsibility determination result is processed by voice, which improves the efficiency of conveying the decision result, and facilitates relevant personnel to understand the decision result in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart of a decision making assistance method provided by an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of the structure of a decision making auxiliary device provided by an embodiment of the present invention;

[0018] Figure 3 A schematic diagram of the structure of a decision making auxiliary device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention provides a decision-making auxiliary method, device, equipment and storage medium. The present invention realizes fast and efficient information collection by real-time description of voice package acquisition, and ensures the timeliness and accuracy of information. This process not only shortens the accident handling time, but also gains precious time for on-site emergency rescue; the context is analyzed by using a deep learning algorithm, which can deeply understand the implicit meaning and emotional color in the context, thereby providing a more accurate event analysis and responsibility determination basis; the corresponding theoretical text information is obtained from a preset knowledge base to provide a comprehensive theoretical basis for decision-making, including laws and regulations, case analysis and professional literature. The knowledge base stores a wealth of reference materials. These theoretical text information can provide a comprehensive reference for the system and help the system make a reasonable and legal decision; the logical text information and theoretical text information are input into the preset decision model, which realizes intelligent and accurate responsibility determination, improves the accuracy and efficiency of decision-making, and can perform multi-dimensional analysis on complex situations, thereby obtaining the most reasonable responsibility determination result; the responsibility determination result is processed by voice, which improves the efficiency of conveying the decision result, and facilitates relevant personnel to understand the decision result in time.

[0020] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0021] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the decision making auxiliary method in the embodiment of the present invention includes:

[0022] 101. Obtain a real-time description voice package, and obtain voice text information according to the real-time description voice package;

[0023] 102. Perform content analysis on the speech and text information based on a preset context analysis model to obtain logical text information;

[0024] In the implementation example of this city, voice and text information is obtained by describing voice packages in real time, so as to achieve fast and efficient information collection. This process not only ensures the timeliness of information, but also can accurately capture key details through advanced voice recognition technology, providing a solid foundation for subsequent analysis and decision-making; for example, after a traffic accident occurs, eyewitnesses at the scene can describe the accident in real time through mobile phone applications, and the system quickly converts these voice information into text for use by the analysis model. This immediacy not only shortens the accident handling time, but also buys precious time for emergency rescue at the scene. The preset context analysis model is used to perform content analysis on voice and text information, extract key information, and provide a basis for subsequent decision-making. This link can understand the implicit meaning and emotional color in the context through deep learning algorithms, so as to more accurately grasp the overall picture and details of the incident and provide a more accurate basis for responsibility determination.

[0025] 103. Obtain corresponding theoretical text information from a preset knowledge base according to the logical text information;

[0026] In this embodiment, corresponding theoretical text information is obtained from a preset knowledge base to provide a theoretical basis for decision-making. The knowledge base stores a wealth of laws and regulations, case analysis, and professional literature. These theoretical text information can provide a comprehensive reference for decision makers to ensure that the decision-making process is both in line with legal norms and professional and authoritative.

[0027] 104. Input the logical text information and the theoretical text information into a preset decision model to obtain a responsibility determination result;

[0028] 105. Perform voice processing on the responsibility determination result to obtain a voice decision result.

[0029] In this embodiment, logical text information and theoretical text information are input into the decision-making model to achieve intelligent and accurate responsibility determination. The decision-making model integrates a variety of algorithms and data, and can perform multi-dimensional analysis of complex situations to obtain the most reasonable responsibility determination results. This process not only improves the accuracy of decision-making, but also greatly shortens the decision-making time and improves the efficiency of police handling; the responsibility determination results are processed in voice to improve the efficiency of communication of decision results, so that relevant personnel can understand the decision results in a timely manner. Through natural language generation technology, complex determination results are converted into easy-to-understand voice information to ensure that the information can be quickly conveyed to on-site personnel and relevant decision makers, thereby speeding up the response speed and processing efficiency of police situations.

[0030] In the embodiment of the present invention, by obtaining the real-time description voice package, the rapid and efficient collection of information is achieved, and the timeliness and accuracy of the information are ensured. This process not only shortens the accident handling time, but also gains precious time for on-site emergency rescue; the context is analyzed by using a deep learning algorithm, which can deeply understand the implicit meaning and emotional color in the context, thereby providing a more accurate event analysis and responsibility determination basis; the corresponding theoretical text information is obtained from the preset knowledge base to provide a comprehensive theoretical basis for decision-making, including laws and regulations, case analysis and professional literature. The knowledge base stores a wealth of reference materials. These theoretical text information can provide a comprehensive reference for the system and help the system make both reasonable and legal decisions; the logical text information and theoretical text information are input into the preset decision model, which realizes intelligent and accurate responsibility determination, improves the accuracy and efficiency of decision-making, and can perform multi-dimensional analysis on complex situations, so as to obtain the most reasonable responsibility determination result; the responsibility determination result is processed by voice, which improves the efficiency of conveying the decision result, and facilitates relevant personnel to understand the decision result in time.

[0031] A second embodiment of the decision making auxiliary method in the embodiment of the present invention includes:

[0032] 201. Capture the user's real-time voice signal in real time, and perform digital conversion on the real-time voice signal to obtain a police recording data set;

[0033] 202. Annotate the police situation data recording set according to the police situation scene to obtain a police situation annotated recording set;

[0034] 203. Obtain a network structure of a Deep Peak2 model, and optimize the network structure to obtain a multi-scenario speech recognition model;

[0035] 204. Divide the police situation annotated recording set into multiple audio recordings, and perform denoising on each audio recording respectively to obtain multiple denoised audio recordings;

[0036] 205. Input the multiple noise-removed audio segments into the multi-scene speech recognition model to obtain multiple segments of recognized text information;

[0037] 206. Connect multiple segments of recognized text information to obtain voice and text information.

[0038] In this embodiment, by capturing the user's real-time voice signals and converting these signals into digital form, the system can create a detailed police recording data set. This process not only ensures the integrity of the information, but also lays the foundation for subsequent analysis and processing; then these recording data are carefully annotated according to the specific scene of the police situation, so as to obtain a more accurate police situation annotated recording set. This step makes each recording in the data set correspond to a specific police situation type, providing a clear classification basis for subsequent analysis.

[0039] The network architecture of the Deep Peak2 model is adopted and optimized to meet the needs of speech recognition in multiple scenarios. After optimization, a more flexible and powerful multi-scenario speech recognition model is built. This model can not only cope with a variety of different emergency scenarios, but also significantly improve the accuracy and robustness of speech recognition, ensuring stable and reliable recognition results in complex and changing environments.

[0040] In order to further improve the processing quality of police recording data, the system divides the police recording set into multiple audio segments and performs denoising on each audio segment. This process effectively removes background noise, making the voice signal clearer, and thus obtaining multiple denoised audio segments. These denoised audio segments provide higher quality input for the speech recognition model, helping to improve the accuracy of recognizing text information.

[0041] The system inputs multiple denoised audio recordings into the optimized multi-scenario speech recognition model to obtain multiple recognized text messages. After splicing, these text messages form complete speech and text messages. This series of steps not only improves the speed of police information processing, but also ensures the accuracy and completeness of the information, providing strong support for the rapid understanding and response of police incidents, and can more effectively respond to various police incidents and ensure public safety.

[0042] A third embodiment of the decision making auxiliary method in the embodiment of the present invention includes:

[0043] 301. Obtain a network structure of a Deep Peak2 model, and introduce a one-dimensional convolutional neural network and a long short-term memory network into the network structure to form a network optimization structure;

[0044] 302. Introducing the Focal Loss algorithm into the weighted cross entropy loss function to obtain an optimized loss function, and introducing the optimized loss function into the network optimization structure to obtain an optimized basic model;

[0045] 303. Performing time expansion and frequency conversion processing on the preset police situation recording training set to obtain a model training set;

[0046] 304. The optimized basic model is trained based on the cross-validation method and the model training set to obtain multiple training results, and each training result is optimized using the hyperparameter tuning method to obtain the optimized basic model with the best training result as the multi-scenario speech recognition model.

[0047] In this embodiment, the network architecture of the Deep Peak2 model is obtained, and based on this architecture, a one-dimensional convolutional neural network (1D-CNN) and a long short-term memory network (LSTM) are integrated to construct a more efficient network optimization structure. The integration of this structure significantly improves the model's ability to extract complex data features and enhances its efficiency in processing time series information. This is of vital importance for accurately identifying and responding to alarm events in a timely manner; and the FocalLoss algorithm is introduced to improve the weighted cross entropy loss function. This strategy innovatively alleviates the sample imbalance problem, allowing the model to focus more on a few difficult-to-identify categories during the training phase, significantly improving the accuracy of identifying key information of the alarm. For the alarm processing system, this improvement means that emergency situations can be identified more accurately, providing solid support for rapid response.

[0048] In addition, this application also implements time stretching and frequency conversion processing on the preset police recording training set, aiming to simulate a variety of police scenarios. This processing strategy significantly improves the diversity of the model training set and enhances the model's adaptability to police recordings in different environments, thereby demonstrating higher robustness and adaptability in practical applications; and based on the cross-validation method and the model training set, the optimized basic model is comprehensively trained, and each training result is carefully optimized through the hyperparameter tuning method. This process ensures that the multi-scenario speech recognition model can perform optimally in various police scenarios, enabling the multi-scenario speech recognition model to quickly and accurately identify police situations in practical applications, greatly improving the efficiency and accuracy of emergency response, and providing strong technical support for public safety.

[0049] A fourth embodiment of the decision making auxiliary method in the embodiment of the present invention includes:

[0050] 401. Obtaining police event record information from a preset police event database, and marking the police event record information according to preset classification labels to obtain police event classification record information;

[0051] 402. Split the police situation classification record information into multiple word sets, and perform denoising on each word set according to a preset stop word library to obtain a police situation word set;

[0052] 403. Using the Word2Ve model, each word in the warning word set is mapped into a vector space of a preset dimension to obtain a warning vector set;

[0053] 404. Introduce the Adam optimizer into the softmax algorithm to obtain the first input model;

[0054] 405. Build the second input model based on the SVR model;

[0055] 406. Introduce the first input model and the second input model into a preset multi-layer neural network to obtain the original context analysis model;

[0056] 407. Train the original context analysis model based on the police situation vector set, calculate the classification score value of each training result using the confusion matrix, and use the training model corresponding to the classification score value as the context analysis model;

[0057] 408. Input the speech text information into the context analysis model to obtain the logical text information.

[0058] In this embodiment, by obtaining the police situation event record information from a preset police situation database and marking the police situation event record information according to the preset classification labels, the rapid identification and classification of police situations can be realized, thereby improving the speed and accuracy of police situation response. This intelligent processing method not only reduces the workload of police officers but also ensures the efficiency and accuracy of police situation handling; splitting the police situation classification record information into multiple word sets and denoising each word set according to the preset stop word library helps to purify the data and extract more valuable information, such as 'de' (的), 'shi' (是), etc. Although these words appear frequently in the text, they are of little significance for retrieval, and these meaningless words are removed according to the preset stop word library. This process ensures the purity and relevance of the police situation word set, providing a solid foundation for subsequent analysis and processing.

[0059] Using the Word2Vec model to map each word in the police word set to a vector space of preset dimensions, text information can be converted into numerical information, which is convenient for mathematical operations and analysis. This vectorization processing not only improves the efficiency of data processing, but also enhances the analyzability of police information; introducing the Adam optimizer into the softmax algorithm can optimize the model training process, improve the convergence speed and prediction accuracy of the model, and this optimization method makes the first input model more accurate and efficient, providing strong support for subsequent decision-making; building a second input model based on the SVR model can conduct a deeper analysis and prediction of police data. The introduction of the SVR model enables the model to handle more complex nonlinear relationships, thereby improving the ability of police prediction and risk assessment; introducing the first input model and the second input model into the preset multi-layer neural network can construct a more complex and powerful context analysis original model. This multi-model fusion method can combine the advantages of different models and improve the overall analysis and prediction capabilities.

[0060] The original context analysis model is trained based on the alarm vector set, and the classification score of each training result is calculated using the confusion matrix to ensure the training quality of the model. In this way, the training model with the highest classification score can be selected as the final context analysis model, ensuring the accuracy and reliability of the model.

[0061] By inputting speech and text information into the context analysis model, speech information can be converted into logical text information for further analysis and processing. This conversion not only improves the efficiency of information processing, but also enhances the operability and application scope of information.

[0062] A fifth embodiment of the decision making auxiliary method in the embodiment of the present invention includes:

[0063] 501. Obtain the information collection link of the government affairs systems of various provinces and cities, and obtain the text data of regulations of the government affairs systems of various provinces and cities through the information collection link;

[0064] 502. Perform word segmentation processing on the legal text data to obtain legal word segmentation data;

[0065] In this embodiment, by integrating the information collection links of the provincial and municipal government systems, comprehensive coverage and grasp of the latest dynamics of the text data of regulations in various regions are achieved. This comprehensive information collection strategy ensures the integrity of the decision-making basis and can be updated in real time, providing cutting-edge legal support for police incident handling; word segmentation is performed on the legal text data. This step not only improves the efficiency of data processing, but also enhances the accuracy of subsequent analysis. Word segmentation ensures that each word in the legal text can be accurately identified and used, providing more detailed and accurate data support for police incident decision-making.

[0066] 503. Build a knowledge base based on Elasticsearch search engine, NLP model and regulatory word segmentation data;

[0067] 504. Perform intent analysis on the logical text information and extract keywords from the logical text information to obtain a search word set;

[0068] 505. Input the search term set into the knowledge base to perform content search to obtain theoretical text information.

[0069] In this embodiment, the knowledge base constructed based on the Elasticsearch search engine, the natural language processing (NLP) model and the regulatory word segmentation data significantly improves the intelligent level of police incident processing. The establishment of the knowledge base enables relevant laws and cases to be quickly retrieved and matched, provides efficient knowledge support for police incident decision-making, and effectively shortens the decision-making time; the intent analysis of logical text information helps to clarify the retrieval target and improve retrieval efficiency and accuracy. By deeply analyzing the inherent logic of text information, keywords related to police incidents can be more accurately extracted to provide more accurate information support for decision-making. The search word set is input into the knowledge base for content retrieval. This operation can quickly obtain theoretical text information related to police incidents. Using the powerful retrieval function of the knowledge base, the most relevant laws and cases can be quickly matched, providing a strong theoretical basis for police incident decision-making and ensuring the scientificity and accuracy of the decision.

[0070] A sixth embodiment of the decision making auxiliary method in the embodiment of the present invention includes:

[0071] 601. Obtaining historical police event related data, and extracting feature information related to preset police event features from the historical police event related data to obtain police event feature information;

[0072] 602. Extracting a target classification model, a target clustering model, and a target prediction model from a preset model resource library according to the alarm feature information;

[0073] 603. Forming a multi-model fusion decision model based on the target classification model, the target clustering model and the target prediction model;

[0074] 604. Input the logical text information and the theoretical text information into the decision model to obtain the responsibility determination result.

[0075] In this embodiment, by collecting relevant data of historical police events, conducting in-depth mining, and extracting characteristic information associated with preset police characteristics, a more accurate description of police characteristics can be obtained. For example, the analysis of historical data can reveal the periodicity or regional characteristics of certain crime patterns, thereby strengthening patrol and monitoring measures in a targeted manner; based on these police characteristic information, the system can accurately select the classification model, clustering model and prediction model that best matches the current police situation from the preset model resource library. The selection of these models is based on the in-depth analysis and understanding of historical data, which ensures the applicability and accuracy of the models. Specifically, the classification model can quickly identify the type of police situation, and the clustering model can classify similar police events. For example, if the system needs to classify police events, classification algorithms such as support vector machines (SVM) and random forests can be selected; if the system needs to predict police events, prediction models such as linear regression and neural networks can be selected.

[0076] Based on these selected target models, a multi-model fusion decision model is constructed, which can integrate the strengths of each model and provide more comprehensive and accurate decision support. The construction of this fusion model not only enhances the scientific nature of decision-making, but also improves the ability to deal with complex police events. For example, when dealing with mass incidents, the fusion model can comprehensively consider multiple factors such as the scale, nature, and scope of influence of the incident, and propose a more reasonable response strategy; finally, the logical text information and theoretical text information are input into the decision model, which can obtain a more accurate responsibility determination result. These text information usually covers laws and regulations, policy orientations, and historical case analysis, providing a solid theoretical basis and logical support for the decision model. For example, when dealing with traffic accidents, the decision model will comprehensively consider the relevant legal provisions for accident responsibility determination, the actual situation at the accident site, and the results of historical case handling, so as to make a more fair and reasonable responsibility determination.

[0077] Through the above series of steps, not only the classification accuracy of police incidents is improved, but also similar police incidents are effectively summarized through cluster analysis, realizing the rational allocation of resources. At the same time, using the prediction model to predict future police trends helps us prepare in advance and improve our ability to respond to emergencies. The application of the multi-model fusion decision model makes the decision results more scientific and accurate, promotes the optimization of the police incident handling process, and improves the efficiency of police work.

[0078] A seventh embodiment of the decision making auxiliary method in the embodiment of the present invention includes:

[0079] 701. Collect historical interaction data of users and perform content analysis on the historical interaction data to obtain personalized demand information;

[0080] 702. Construct a speech conversion model according to the personalized demand information and the TTS engine;

[0081] 703. Perform voice conversion on the responsibility determination result through a voice conversion model to obtain a voice decision result.

[0082] In this embodiment, the system collects the user's historical interaction data and conducts in-depth content analysis on these data, so that the system can accurately mine the user's personalized demand information. This process helps us to understand the user's behavior patterns and preferences more deeply, so as to provide users with a more accurate and considerate service experience. For example, when analyzing the user's interaction records in different scenarios, we can predict the user's possible needs and prepare in advance to provide timely help when the user needs it; then combine these personalized demand information and TTS (Text-to-Speech) engine technology, we can build a highly customized voice conversion model. This model can not only imitate the user's language habits, but also add emotional colors in the speech synthesis process, making the voice output more natural and expressive. For example, when providing navigation services to users, the voice conversion model can adjust the speed and tone according to the user's preferences, and even add appropriate modal particles to make the navigation information sound more friendly and easy to understand; through this voice conversion model, the responsibility determination result can be converted into a voice decision result. This process not only improves the efficiency of information transmission, but also makes the decision result easier to be accepted by users. In an emergency, the system can quickly convert complex responsibility determination results into concise and clear voice information, so that users can immediately understand the situation and respond accordingly, which undoubtedly greatly improves the response speed and accuracy of decision-making. In this way, we not only improve the user experience, but also enhance the practicality and reliability of the system.

[0083] The above describes the decision-making auxiliary method in the embodiment of the present invention. The following describes the decision-making auxiliary device in the embodiment of the present invention. Figure 2 , an embodiment of the decision making auxiliary device in the embodiment of the present invention includes:

[0084] The recognition module 801 is used to obtain a real-time description voice package and obtain voice text information according to the real-time description voice package;

[0085] The analysis module 802 is used to perform content analysis on the speech and text information based on a preset context analysis model to obtain logical text information;

[0086] The acquisition module 803 is used to acquire corresponding theoretical text information from a preset knowledge base according to the logical text information;

[0087] The decision module 804 is used to input the logical text information and the theoretical text information into a preset decision model to obtain a responsibility determination result;

[0088] The voice module 805 is used to perform voice processing on the responsibility determination result to obtain a voice decision result.

[0089] In this embodiment, the recognition module 801 achieves fast and efficient information collection by real-time description of the acquisition of voice packets, ensuring the timeliness and accuracy of the information. This process not only shortens the accident handling time, but also gains precious time for on-site emergency rescue; the analysis module 802 uses a deep learning algorithm to analyze the context, which can deeply understand the implicit meaning and emotional color in the context, thereby providing a more accurate event analysis and responsibility determination basis; the acquisition module 803 obtains the corresponding theoretical text information from the preset knowledge base to provide a comprehensive theoretical basis for decision-making, including laws and regulations, case analysis and professional literature. The knowledge base stores a wealth of reference materials. These theoretical text information can provide a comprehensive reference for the system and help the system make both reasonable and legal decisions; the decision module 804 inputs the logical text information and theoretical text information into the preset decision model, realizes intelligent and accurate responsibility determination, improves the accuracy and efficiency of decision-making, and can perform multi-dimensional analysis on complex situations, thereby obtaining the most reasonable responsibility determination result; the voice module 805 performs voice processing on the responsibility determination result, improves the efficiency of conveying the decision result, and facilitates relevant personnel to understand the decision result in a timely manner.

[0090] above Figure 2 The decision making auxiliary device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The decision making auxiliary device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0091] Figure 3: is a schematic diagram of the structure of a decision-making auxiliary device provided by an embodiment of the present invention. The decision-making auxiliary device 900 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 910 (for example, one or more processors) and a memory 920, and one or more storage media 930 (for example, one or more mass storage devices) storing application programs 933 or data 932. Among them, the memory 920 and the storage medium 930 can be short-term storage or permanent storage. The program stored in the storage medium 930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the decision-making auxiliary device 900. Furthermore, the processor 910 can be configured to communicate with the storage medium 930, and execute a series of instruction operations in the storage medium 930 on the decision-making auxiliary device 900 to implement the steps of the decision-making auxiliary method provided by the above-mentioned method embodiments.

[0092] The decision making auxiliary device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input and output interfaces 960, and / or one or more operating systems 931, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the decision-making assistance device shown does not constitute a limitation on the decision-making assistance device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0093] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the decision-making auxiliary method.

[0094] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device or unit described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0095] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0096] Finally, it should be noted that the above description is only a preferred example of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A decision making auxiliary method, characterized in that: include: Acquire a real-time description voice package, and acquire voice text information according to the real-time description voice package; Conduct content analysis on speech and text information based on a preset context analysis model to obtain logical text information; Acquire corresponding theoretical text information from a preset knowledge base according to the logical text information; Inputting logical text information and theoretical text information into a preset decision model to obtain a responsibility determination result; The responsibility determination result is processed into voice to obtain a voice decision result.

2. The decision making assistance method according to claim 1, characterized in that: The step of obtaining a real-time description voice package and obtaining voice text information according to the real-time description voice package includes: Capture the user's real-time voice signal in real time and convert it into digital form to obtain a police recording data set; The police situation data recording set is annotated according to the police situation scene to obtain a police situation annotated recording set; Obtain the network structure of the Deep Peak2 model and optimize the network structure to obtain a multi-scenario speech recognition model; The police situation annotated recording set is divided into multiple recording audio segments, and each recording audio segment is subjected to denoising processing to obtain multiple denoised audio segments; Inputting multiple noise-removed audio segments into a multi-scene speech recognition model to obtain multiple segments of recognized text information; Multiple segments of recognized text information are concatenated to obtain voice text information.

3. The decision making assistance method according to claim 2, characterized in that: The method of obtaining the network structure of the Deep Peak2 model and optimizing the network structure to obtain a multi-scenario speech recognition model includes: Obtain the network structure of the Deep Peak2 model, and introduce a one-dimensional convolutional neural network and a long short-term memory network into the network structure to form a network optimization structure; The Focal Loss algorithm is introduced into the weighted cross entropy loss function to obtain the optimized loss function, and the optimized loss function is introduced into the network optimization structure to obtain the optimized basic model; Perform time expansion and frequency conversion on the preset police recording training set to obtain a model training set; The optimized basic model is trained based on the cross-validation method and the model training set to obtain multiple training results, and each training result is optimized using the hyperparameter tuning method to obtain the optimized basic model with the best training results as the multi-scenario speech recognition model.

4. The decision making assistance method according to claim 1, characterized in that: The content analysis of the speech and text information based on the preset context analysis model to obtain logical text information includes: Acquire police event record information from a preset police event database, and mark the police event record information according to preset classification tags to obtain police classification record information; The police classification record information is split into multiple word sets, and each word set is denoised according to a preset stop word library to obtain a police word set; The Word2Ve model is used to map each word in the warning word set into a vector space of a preset dimension to obtain a warning vector set; Introduce the Adam optimizer into the softmax algorithm to obtain the first input model; The second input model is formed based on the SVR model; The first input model and the second input model are introduced into a preset multi-layer neural network to obtain an original context analysis model; The original context analysis model is trained based on the alarm vector set, and the classification score value of each training result is calculated using the confusion matrix, and the training model corresponding to the classification score value is used as the context analysis model; Input the speech and text information into the context analysis model to obtain logical text information.

5. The decision making assistance method according to claim 1, characterized in that: The step of obtaining corresponding theoretical text information from a preset knowledge base according to the logical text information includes: Obtain information collection links from provincial and municipal government systems, and obtain regulatory text data from provincial and municipal government systems through information collection links; Perform word segmentation processing on the regulatory text data to obtain regulatory word segmentation data; Build a knowledge base based on Elasticsearch search engine, NLP model and regulatory word segmentation data; Perform intent analysis on the logical text information and extract keywords from the logical text information to obtain a search term set; The search term set is input into the knowledge base for content retrieval to obtain theoretical text information.

6. The decision making assistance method according to claim 1, characterized in that: The step of inputting the logical text information and the theoretical text information into a preset decision model to obtain a responsibility determination result includes: Acquire historical police event related data, and extract characteristic information related to preset police event characteristics from the historical police event related data to obtain police event characteristic information; Extract target classification model, target clustering model and target prediction model from the preset model resource library according to the characteristic information of the alarm situation; A decision model based on multi-model fusion is formed based on target classification model, target clustering model and target prediction model; The logical text information and the theoretical text information are input into the decision-making model to obtain the responsibility determination result.

7. The decision making assistance method according to claim 1, characterized in that: The voice processing of the responsibility determination result to obtain the voice decision result includes: Collect users' historical interaction data and perform content analysis on the historical interaction data to obtain personalized demand information; Build a speech conversion model based on personalized demand information and TTS engine; The responsibility determination result is converted into voice through the voice conversion model to obtain the voice decision result.

8. A decision making auxiliary device, characterized in that: include: A recognition module, used to obtain a real-time description voice package and obtain voice text information according to the real-time description voice package; An analysis module, used to perform content analysis on speech and text information based on a preset context analysis model to obtain logical text information; An acquisition module is used to acquire corresponding theoretical text information from a preset knowledge base according to the logical text information; A decision module is used to input the logical text information and the theoretical text information into a preset decision model to obtain a responsibility determination result; The voice module is used to perform voice processing on the responsibility determination result to obtain a voice decision result.

9. A decision making auxiliary device, characterized in that: The decision-making auxiliary device comprises: a memory and at least one processor, wherein instructions are stored in the memory; At least one of the processors calls the instructions in the memory to enable the decision-making assistance device to perform each step of the decision-making assistance method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the steps of the decision making assistance method according to any one of claims 1 to 7 are implemented.

Images