Electronic government affair platform based on block chain

Abstract

The invention relates to an electronic government affair platform based on a block chain, which comprises a block chain service platform terminal, the block chain service platform terminal is connected with a user terminal through the Internet, and the user terminal comprises a personal user terminal and a government affair user terminal; the block chain service platform terminal is connected with the AI processing center through a high-speed transmission line, and the AI processing center comprises a dialect recognition module, an image recognition module, a natural language processing module and a noise analysis and processing module; the block chain service platform terminal is further connected with a block chain network through an encryption channel, security and transparency of data are ensured, a deep learning algorithm is adopted to train and optimize a dialect recognition module, a plurality of recognition models for specific dialects are formed, meanwhile, a personal native place database is accessed, and the recognition efficiency of the dialect recognition module is improved. The platform can automatically select the dialect recognition model matched with the native place of the user, realizes accurate and rapid conversion of dialects, improves the dialect recognition capability, reduces communication obstacles, and improves the user experience and satisfaction.

Description

Technical Field

[0001] The present invention relates to the field of blockchain service platforms, and specifically to an e-government platform based on blockchain. Background Art

[0002] With the rapid development of information technology, e-government platforms have become a crucial tool for governments to improve service efficiency and enhance transparency. Traditional e-government platforms primarily rely on centralized database and server architectures. While these platforms have enabled online government services to a certain extent, they still suffer from numerous shortcomings. In particular, in terms of data processing, user interaction, information security, and intelligent services, traditional solutions are no longer able to meet society's current demand for efficient, convenient, and secure government services.

[0003] According to the patent publication 202110477884.X, a blockchain-based smart e-government management platform and method thereof include a government operation platform, the government operation platform is connected to the government cloud assistant client via a transmission line, the government cloud assistant client is connected to the government cloud assistant server, the government operation platform is connected to the voice enhancement module via a wire, and the voice enhancement module is connected to the voice equipment device via a wire. This blockchain-based smart e-government management platform and method thereof, through the added frequency domain discriminator, enables the voice enhancement model to simultaneously learn the time domain characteristics and frequency domain characteristics of speech and noise, thereby improving the performance and generalization of voice enhancement; by designing the network structure of the frequency domain discriminator and the time domain discriminator to have the same structure, this can effectively prevent the model from experiencing training instability or non-convergence during training; through the structure of the voice enhancement module, three training sessions can greatly improve the robustness and accuracy of the model, in the process of implementing the present invention.

[0004] The inventors have discovered that the prior art has at least the following unresolved problems, for example:

[0005] (1) Although traditional speech recognition technology can recognize standard language speech information to a certain extent, it is often ineffective when processing dialects. The pronunciation characteristics, intonation changes, and regional vocabulary of dialects make it difficult for traditional speech recognition technology to accurately recognize and understand dialect speech. Due to low processing efficiency and difficulty in ensuring accuracy, users may encounter communication barriers when using traditional blockchain e-government platforms for multilingual communication, thereby affecting user experience and satisfaction;

[0006] (2) Traditional solutions usually do not consider the user's personalized needs and information background, and are unable to provide accurate grammatical, spelling, and semantic correction suggestions based on the user's place of origin, language habits, and other information. This limits the platform's ability to provide personalized services. Due to the lack of intelligent and personalized services, users may encounter comprehension barriers or poor communication when using traditional blockchain e-government platforms for natural language communication, thus affecting user experience and satisfaction.

[0007] (3) Due to the lack of visualization methods, researchers and developers find it difficult to intuitively understand the characteristics of noisy data and the processing effects. This makes it difficult for them to design and optimize processing algorithms and accurately evaluate the performance and effects of the algorithms. Since they cannot intuitively observe the processing process and results of noisy data, researchers and developers may need to spend a lot of time and energy on parameter adjustment and experimental verification when debugging and optimizing algorithms. This not only increases development costs and time costs, but also easily affects the final performance and stability of the algorithm. Summary of the Invention

[0008] The present invention aims to overcome the shortcomings of existing technologies and meet current needs by providing a blockchain-based e-government platform. This solution addresses the problem that, while traditional speech recognition technology can recognize standard language speech information to a certain extent, it often performs poorly when processing dialects. Factors such as the pronunciation characteristics, intonation variations, and regional vocabulary of dialects make it difficult for traditional speech recognition technology to accurately recognize and understand dialect speech. Due to low processing efficiency and difficulty ensuring accuracy, users may encounter communication barriers when using traditional blockchain e-government platforms for multilingual communication, thereby affecting user experience and satisfaction. Traditional solutions generally do not consider users' personalized needs and information context, and are unable to provide accurate grammatical, spelling, and semantic correction suggestions based on information such as their place of origin and language habits. This limits the platform's ability to provide personalized services. Due to the lack of intelligent and personalized services, users may encounter comprehension barriers or communication problems when using traditional blockchain e-government platforms for natural language communication, thereby affecting user experience and satisfaction. The lack of visualization tools makes it difficult for researchers and developers to intuitively understand the characteristics of noisy data and the effects of processing. This makes it difficult to design and optimize processing algorithms, making it difficult to accurately evaluate their performance and effectiveness. Without the ability to visually observe the processing process and results of noisy data, researchers and developers may need to spend considerable time and effort on parameter adjustments and experimental verification when debugging and optimizing algorithms. This not only increases development costs and time, but also easily leads to technical issues that affect the algorithm's ultimate performance and stability.

[0009] In order to achieve the purpose of the present invention, the technical solution adopted by the present invention is: designing a blockchain-based e-government platform, including a blockchain service platform terminal, the blockchain service platform terminal is connected to the user terminal through the Internet, and the user terminal includes a personal user terminal and a government user terminal; the blockchain service platform terminal is connected to the AI ​​processing center through a high-speed transmission line, and the AI ​​processing center includes a dialect recognition module, an image recognition module, a natural language processing module, and a noise analysis and processing module; the blockchain service platform terminal is also connected to the blockchain network through an encrypted channel to ensure the security and transparency of the data.

[0010] Preferably, the blockchain service platform terminal is connected to the personal place of origin database, and automatically selects the corresponding dialect recognition model based on the personal place of origin information of government officials and ordinary people, thereby improving the efficiency and accuracy of dialect recognition.

[0011] Preferably, the dialect recognition module adopts a deep learning algorithm, is trained and optimized for a variety of dialects, and forms multiple dialect recognition models. When receiving user voice, it automatically selects the corresponding dialect recognition model for recognition based on the user's place of origin information, thereby achieving accurate and fast conversion of dialects.

[0012] Preferably, the image recognition module adopts a convolutional neural network (CNN) algorithm to identify objects, scenes or facial information in pictures or videos uploaded by users. At the same time, the module can be linked with camera equipment to realize real-time video stream analysis, assist in lip recognition and gesture recognition, and enhance user interaction experience.

[0013] Preferably, the natural language processing module adopts the Transformer architecture and has functions such as text classification, entity recognition, sentiment analysis and automatic error correction. It can provide more accurate grammatical, spelling and semantic error correction suggestions by combining contextual information and user's place of origin when the user enters text content.

[0014] Preferably, the noise analysis and processing module analyzes the noise in the received voice signal and converts the noise into a picture form; the module further uses the big data of the noise picture and, through deep learning, performs a holistic correction process on the original sound signal to improve the accuracy and clarity of voice recognition.

[0015] Preferably, the blockchain service platform terminal is also connected to an intelligent recommendation system, which adopts a hybrid recommendation algorithm combining collaborative filtering and deep learning to accurately recommend relevant services, information or policy information to users based on their historical behavior, preferences, place of origin and contextual information, thereby improving user experience.

[0016] Preferably, the blockchain service platform terminal includes a user information management module, a service request processing module, a service evaluation feedback module and a data statistical analysis module. The user information management module is used to store and manage the user's personal information, native place information and service records. The service request processing module is used to receive and process the user's service request. The service evaluation feedback module is used to collect the user's evaluation and feedback on the service. The data statistical analysis module is used to mine and analyze the data on the platform to provide a basis for platform optimization and service improvement.

[0017] Preferably, the blockchain network is used to store historical records of all transactions, service requests and evaluation feedback on the platform to ensure the immutability and traceability of data. At the same time, through smart contract technology, it realizes the automated execution and settlement of services, thereby improving service efficiency and transparency.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. The present invention uses a deep learning algorithm to train and optimize the dialect recognition module, forming multiple recognition models for specific dialects. At the same time, by accessing the personal place of origin database, the platform can automatically select the dialect recognition model that matches the user's place of origin, achieving accurate and rapid conversion of dialects, improving dialect recognition capabilities, reducing communication barriers, and enhancing user experience and satisfaction.

[0020] 2. This invention uses a natural language processing module and a Transformer architecture to combine contextual information and user's place of origin to provide accurate grammatical, spelling and semantic correction suggestions.

[0021] 3. By converting noise into images, this invention visualizes noise characteristics. This visualization helps researchers and developers more intuitively understand the characteristics of noise, enabling more targeted design of processing algorithms. Visualization also facilitates rapid identification of noise types, facilitating subsequent automated processing. Furthermore, by converting noise into images, a large-scale noise image database can be constructed. This large dataset provides rich training material for machine learning algorithms, helping to develop more generalized and accurate noise processing models. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is an overall schematic diagram of the present invention;

[0023] Figure 2 A schematic diagram of a user terminal of the present invention;

[0024] Figure 3 This is a schematic diagram of the blockchain network connection of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0026] A blockchain-based e-government platform, see Figures 1 to 3 , including blockchain service platform terminals, which are connected to user terminals through the Internet. User terminals include personal user terminals and government user terminals; blockchain service platform terminals are connected to AI processing centers through high-speed transmission lines. AI processing centers include dialect recognition modules, image recognition modules, natural language processing modules, and noise analysis and processing modules; blockchain service platform terminals are also connected to blockchain networks through encrypted channels to ensure data security and transparency.

[0027] For details, see Figure 1 and Figure 3 The blockchain service platform terminal is connected to the personal place of origin database, and automatically selects the corresponding dialect recognition model based on the personal place of origin information of government officials and ordinary people, thereby improving the efficiency and accuracy of dialect recognition.

[0028] For details, see Figure 1 The dialect recognition module adopts a deep learning algorithm (using the LSTM (long short-term memory network) algorithm, trained for the eight major Chinese dialects (such as Mandarin, Wu, Cantonese, etc.), and the training data volume of each dialect model reaches millions of voice samples). It is trained and optimized for multiple dialects to form multiple dialect recognition models. When receiving user voice, it automatically selects the corresponding dialect recognition model according to the user's place of origin information (according to the information in the user's place of origin database, it automatically matches the optimal dialect recognition model, and the matching time is less than 0.1 second) to achieve accurate and fast dialect conversion.

[0029] For more details, see Figure 1 The image recognition module adopts convolutional neural network (CNN) (using ResNet-50 architecture to identify objects, scenes or faces in pictures and videos with an accuracy of 98%) to identify objects, scenes or face information in pictures or videos uploaded by users. At the same time, the module can be linked with camera equipment to realize real-time video stream analysis (linked with high-definition camera, the frame rate reaches 30fps, and the delay of lip recognition and gesture recognition is less than 0.2 seconds), assisting in lip recognition and gesture recognition, and improving user interaction experience.

[0030] For further information, see Figure 1The natural language processing module adopts the Transformer architecture (based on the BERT (Bidirectional Encoder Representations from Transformers) model, with text classification, entity recognition, sentiment analysis and automatic error correction functions). It has functions such as text classification, entity recognition, sentiment analysis and automatic error correction. When the user enters text content, it can combine contextual information and user's place of origin information to provide more accurate grammatical, spelling and semantic correction suggestions.

[0031] Further, see Figure 1 The noise analysis and processing module (using Mel spectrum to represent the noise in the speech signal, converting the noise into image form, each noise image contains 1024x1024 pixels) analyzes the noise in the received speech signal and converts the noise into image form; this module further utilizes the big data of noise images and performs overall correction processing on the original sound signal through deep learning (using the Wave-U-Net model to comprehensively correct the original sound signal, improving the speech clarity by about 30% and the recognition accuracy to 97%) to improve the accuracy and clarity of speech recognition.

[0032] It is worth noting that, see Figure 1 The blockchain service platform terminal is also connected to the intelligent recommendation system, which uses a hybrid recommendation algorithm that combines collaborative filtering and deep learning (combining collaborative filtering (based on the user-item rating matrix) and deep learning (based on user historical behavior and preference characteristics), the recommendation accuracy is increased to 90%). According to the user's historical behavior, preferences, place of origin information and contextual information, the system accurately recommends relevant services, information or policy information to users, improving user experience, with a response time of less than 0.5 seconds.

[0033] It is worth noting that see Figure 1 The blockchain service platform terminal includes a user information management module, a service request processing module, a service evaluation feedback module and a data statistical analysis module. The user information management module is used to store and manage users' personal information, place of origin information and service records. The service request processing module is used to receive and process users' service requests. The service evaluation feedback module is used to collect users' evaluation and feedback on the services. The data statistical analysis module is used to mine and analyze the data on the platform to provide a basis for platform optimization and service improvement.

[0034] It is worth mentioning that see Figure 1The blockchain network is used to store the historical records of all transactions, service requests and evaluation feedback on the platform to ensure the immutability and traceability of the data. At the same time, through smart contract technology, it realizes the automated execution and settlement of services, improving service efficiency and transparency.

[0035] Example 1

[0036] Dialect Recognition Module

[0037] S1. Data preparation and model training

[0038] Data collection:

[0039] Speech samples of the eight major Chinese dialects were collected from voice databases, TV programs, live webcasts and other sources.

[0040] The training data volume for each dialect exceeds 1,200,000. For example, the Cantonese dataset contains 1,300,000 speech samples, covering pronunciations from Guangzhou, Shenzhen, Hong Kong, and other places.

[0041] Data preprocessing:

[0042] The collected voice data is cleaned to remove noise and irrelevant information, such as background music and coughing sounds.

[0043] Dialect categories and text content are annotated to ensure that each voice sample has an accurate label. For example, a Cantonese voice sample is annotated as "Cantonese - Guangzhou - Daily Conversation - 'Hello, where are you going?'".

[0044] Model training:

[0045] The LSTM (Long Short-Term Memory Network) algorithm is used to train a recognition model for each dialect.

[0046] During training, the Adam optimizer was used, the learning rate was set to 0.001, and the batch size was 64.

[0047] The number of LSTM layers is set to 3, and each layer contains 256 hidden units.

[0048] The dropout technique is used to prevent overfitting, and the dropout rate is set to 0.5.

[0049] Model optimization:

[0050] After multiple iterations of training, the accuracy of the model on the validation set remained stable at over 96%.

[0051] Fine-tune the model, such as adjusting the learning rate decay strategy, adding regularization terms, etc., to further improve the recognition accuracy.

[0052] S2. User Native Place Information Matching and Model Selection

[0053] User Native Place Database:

[0054] Build a user native place database to store users' personal information, including the native place (specific to the province or region).

[0055] For example, the native place information of user A is "Guangdong Province - Guangzhou City".

[0056] Model Matching Mechanism:

[0057] When receiving the user's voice, first select the optimal model from the trained dialect recognition models according to the user's native place information. [[ID=1--7]]

[0058] For example, if the native place of user A is Guangdong Province, select the Cantonese recognition model for recognition.

[0059] Fast Matching:

[0060] Design an efficient indexing mechanism, such as a hash table or B-tree, to ensure that the model selection can be completed within 0.05 seconds.

[0061] S3. Dialect Recognition and Conversion

[0062] Speech Recognition:

[0063] Use the selected dialect recognition model to recognize the user's voice and output the text content.

[0064] For example, for a Cantonese sentence "食咗饭未?" spoken by user A, the model outputs the text "吃了饭没?" after recognition.

[0065] Dialect Conversion (Optional):

[0066] If it is necessary to convert the recognition result into Mandarin or other dialects, a dialect-to-Mandarin (or other dialects) conversion module can be further developed.

[0067] For example, convert the Cantonese sentence "食咗饭未?" into Mandarin "吃了饭没?", or into Hakka "食饭嘞冒?".

[0068] The conversion module can be implemented based on statistical machine translation or deep learning techniques, such as using the Transformer model for sequence-to-sequence translation. [[ID=--1]]

[0069] Result Output:

[0070] Output the recognized and converted text content to the subsequent processing modules of the e-government platform, such as the natural language processing module or the intelligent recommendation system.

[0071] For example, the input text "Have you eaten?" is passed to the natural language processing module for sentiment analysis or intent recognition.

[0072] S4. Performance evaluation and optimization

[0073] Recognition accuracy:

[0074] The accuracy of each dialect recognition model is evaluated through the test set to ensure that it meets the requirements of practical application.

[0075] For example, the Cantonese recognition model achieved an accuracy of 97% on the test set, and the Hakka recognition model achieved an accuracy of 95%.

[0076] User feedback:

[0077] Collect user feedback on recognition results for continuous iteration and optimization of the model.

[0078] For example, user B reports that the recognition result is incorrect, misidentifying "how good is the weather today" as "the weather is much better today". Based on this, the model parameters can be adjusted or relevant training data can be added.

[0079] Adaptive adjustments:

[0080] As the user group changes, the distribution of training data should be adjusted in a timely manner to adapt to new dialects or voice characteristics.

[0081] For example, with the development of the Guangdong-Hong Kong-Macao Greater Bay Area, the collection and training of Cantonese pronunciation in Hong Kong and Macao will be increased to improve the recognition accuracy of the model.

[0082] Example

[0083] Assume that user C's hometown is Changsha City, Hunan Province. He uses the voice interaction function of the e-government platform to say a sentence in Hunan dialect "Ni qia fan da mao?"

[0084] User native place information matching: Based on user C’s native place information, select the Xiang dialect recognition model.

[0085] Dialect recognition: The Xiang dialect recognition model recognizes user C's speech and outputs the text "Have you eaten?"

[0086] Result output: The recognition result "Have you eaten?" is passed to the subsequent processing module of the e-government platform for further processing.

[0087] Through the above embodiments, the dialect recognition module can efficiently and accurately recognize and convert the dialect content in the user's voice, providing powerful voice interaction support for the regional chain e-government platform.

[0088] Example 2

[0089] Image recognition module

[0090] S1. Image recognition module architecture

[0091] Architecture selection: The image recognition module uses a convolutional neural network (CNN) architecture, specifically ResNet-50 as the model foundation. ResNet-50 has been widely used in the image recognition field due to its powerful feature extraction capabilities and high recognition accuracy.

[0092] S2. Recognition accuracy and performance

[0093] Recognition accuracy: Across a large number of test samples, the image recognition module achieved 98% accuracy in identifying objects, scenes, or faces in images and videos. This means that out of 1,000 test images, the module correctly identified 980.

[0094] Processing Speed: Powered by a high-performance GPU, the module processes a single image in an average of 0.05 seconds. This speed ensures the module can respond to user upload requests in real time, providing a smooth interactive experience. For example, after a user uploads an image, they can see the recognition results almost immediately.

[0095] Video Stream Analysis: When coupled with a high-definition camera, the module can process video streams at 30 frames per second (fps), enabling real-time lip and gesture recognition. The latency for both lip and gesture recognition is less than 0.2 seconds, ensuring real-time and accurate recognition.

[0096] S3, real-time video streaming analysis

[0097] Hardware configuration: The HD camera uses a 1080P resolution and a frame rate of 30fps. The camera is connected to the AI ​​processing center via high-speed transmission lines such as Gigabit Ethernet to ensure real-time transmission and processing of video streams.

[0098] Identification task:

[0099] Lip recognition: This module analyzes the user's lip movements in a video stream to identify the words or phrases being spoken. For example, in the voice interaction function of the e-government platform, when a user says "submit an application," the module accurately recognizes the user's lip movements and triggers the corresponding submission action within 0.18 seconds. This greatly simplifies the process and improves efficiency.

[0100] Gesture Recognition: The module recognizes user gestures in a video stream and executes corresponding actions. For example, when a user makes a "like" gesture (such as a thumbs-up), the module recognizes it and triggers the "like" function within 0.16 seconds, storing the user's "like" record on the blockchain. This not only increases the fun of user interaction but also ensures data immutability and traceability.

[0101] Value: In one test, the image recognition module successfully identified 980 lip changes and 995 hand gestures in a 10-minute video stream analysis. Specifically:

[0102] In the lip shape recognition task, the module identified 500 lip shape changes of "submit an application", of which 493 were correctly recognized, with an accuracy rate of 98.6%.

[0103] In the gesture recognition task, the module recognized 500 “thumbs-up” gestures, of which 498 were correctly recognized, with an accuracy rate of 99.6%.

[0104] In terms of recognition delay, the average delay for lip recognition is 0.18 seconds (standard deviation is 0.02 seconds), and the average delay for gesture recognition is 0.16 seconds (standard deviation is 0.01 seconds).

[0105] During the entire test process, the module did not experience any lag or recognition failure, ensuring the stability and reliability of real-time video stream analysis.

[0106] S4. Improved user experience

[0107] The image recognition module's real-time video stream analysis significantly enhances the user experience. Users can now complete complex operations, such as submitting applications, liking posts, and commenting, without having to manually enter text or click buttons. This not only simplifies the process but also increases user satisfaction and loyalty. For example, an elderly user applying for social security benefits on the e-government platform can simply mouth the words "submit application" to complete the process, without having to worry about unfamiliar mobile phone operations.

[0108] Through the above examples, we can see that the image recognition module plays an important role in the blockchain-based e-government platform. Its high recognition accuracy, fast processing speed, and real-time video stream analysis capabilities provide users with a smooth and convenient interactive experience.

[0109] Example 3

[0110] Natural language processing module (based on Transformer architecture and BERT model)

[0111] Module architecture and model selection

[0112] The natural language processing module uses the Transformer architecture and is built on the BERT model. The BERT model uses bidirectional encoding to more comprehensively understand the context of text, thereby improving the accuracy and efficiency of text processing.

[0113] Function Implementation

[0114] Text Classification: The module can automatically classify the text input by the user, such as government affairs consultation, business handling, complaints and suggestions, etc. For example, when the user inputs "I want to handle the social security transfer procedure", the module can accurately identify and classify it into the "business handling" category.

[0115] Entity Recognition: The module can recognize key entities in the text, such as person names, place names, organization names, etc. For example, in the text "Zhang San submitted a social security transfer application to the Beijing Municipal Government", the module can accurately identify entities such as "Zhang San", "Beijing Municipal Government", and "social security transfer application".

[0116] Sentiment Analysis: The module can analyze the sentiment tendency of the text input by the user, such as positive, negative or neutral. For example, when the user inputs "This handling process is too cumbersome. I hope it can be simplified in the future", the module can identify that the sentiment tendency of the text is negative.

[0117] Automatic Error Correction: When the user inputs text content, the module can provide accurate grammar, spelling and semantic error correction suggestions in combination with context information and the user's native place information. For example, when the user inputs "I want to handle social security transfer" (there is a misspelled character "xiang"), the module can identify and suggest modifying it to "I want to handle social security transfer".

[0118] Performance Parameters and Values

[0119] Processing Speed: The processing speed of the module for the text input by the user reaches the millisecond level, ensuring that the user can obtain the processing result immediately. For example, in the test, when the module processes 1000 pieces of text input by the user, the average processing time is only 0.02 seconds per piece.

[0120] Accuracy Rate: The accuracy rate of the module in text classification, entity recognition, sentiment analysis and automatic error correction all reaches a high level. The specific values are as follows:

[0121] Text Classification Accuracy Rate: 96% (In the test, when classifying 1000 pieces of text, 960 pieces were correctly classified).

[0122] Entity Recognition Accuracy Rate: 98% (In the test, when recognizing text containing 500 entities, 490 were correctly recognized).

[0123] Sentiment Analysis Accuracy Rate: 95% (In the test, when analyzing the sentiment of 1000 pieces of text, 950 pieces were correctly judged for their sentiment tendency).

[0124] Automatic Error Correction Accuracy Rate: 97% (In the test, when correcting text containing 1000 misspelled characters, 970 were correctly corrected).

[0125] Examples and Values

[0126] Example 1: The user inputs "I want to go through the retirement procedures" (there is a typo "Item"), the module recognizes and suggests changing it to "I want to go through the retirement procedures", and the error correction accuracy rate is 100% (in the test, texts containing 100 similar typos were corrected, and all were corrected correctly).

[0127] Example 2: The user inputs "I am very satisfied with the experience of this government service", and the module performs sentiment analysis and judges it to be a positive sentiment. The sentiment analysis accuracy rate is 100% (in the test, 1,000 texts with similar sentiments were analyzed and all were correctly judged).

[0128] Example 3: The user inputs "I want to know about social security policies", and the module classifies the text into the "Government Affairs Consultation" category with a classification accuracy of 98% (in the test, 1,000 similar texts were classified, and 980 were correctly classified).

[0129] Example 4: The user inputs "I submitted a social security transfer application to the Beijing Municipal Government", and the module performs entity recognition, identifying entities such as "Beijing Municipal Government" and "Social Security Transfer Application", with an entity recognition accuracy of 99% (in the test, similar text containing 500 entities was identified, and 495 were correctly identified).

[0130] The above examples demonstrate the important role that natural language processing (NLP) modules play in blockchain-based e-government platforms. Its efficient text processing capabilities, accurate classification and recognition capabilities, and precise error correction suggestions provide users with a convenient and efficient government service experience.

[0131] Example 4

[0132] Noise analysis and processing module

[0133] S1. Module architecture and workflow

[0134] The noise analysis and processing module receives the voice signal from the user terminal and first converts the noise in the voice signal into an image using a Mel spectrogram. Each noise image contains 1024x1024 pixels.

[0135] Next, the module uses the big data of the noise image and deep learning technology (specifically the Wave-U-Net model) to perform a holistic correction process on the original sound signal.

[0136] Finally, the module outputs the corrected sound signal for recognition by the subsequent speech recognition module.

[0137] S2. Performance parameters and values

[0138] Noise conversion efficiency: The module can quickly convert the noise in the received voice signal into image form, with the conversion time being less than 0.1 second.

[0139] Image resolution: Each noise image contains 1024x1024 pixels, which can clearly show the spectral characteristics of the noise.

[0140] Deep Learning Model Performance: The Wave-U-Net model, trained on a large number of noisy images, significantly improves the original sound signal. In testing, the model improved speech clarity by approximately 30% and recognition accuracy to 97%.

[0141] S3. Values ​​and Examples

[0142] Test environment: To verify the performance of the noise analysis and processing module, we conducted experiments in a test environment that includes a user terminal (user voice input), a noise analysis and processing module, and a speech recognition module.

[0143] Test data: We prepared 1,000 speech samples with varying noise levels as test data. These samples cover common environmental noises (such as traffic noise and human voices) as well as speech content in different dialects.

[0144] Testing process: Each voice sample is first sent to the noise analysis and processing module via the user terminal. This module analyzes the noise in the voice signal and converts it into an image. The Wave-U-Net model then corrects the original voice signal. Finally, the corrected voice signal is sent to the speech recognition module for recognition.

[0145] Test results:

[0146] Before the noise analysis and processing module was used, the recognition accuracy of the speech recognition module was 85%.

[0147] After using the noise analysis and processing module, the recognition accuracy of the speech recognition module increased to 97%.

[0148] Specifically for each voice sample, for example, a voice sample containing traffic noise, before using the module, the voice recognition module could not correctly identify its content; after using the module, the voice recognition module can accurately identify the key information in the voice.

[0149] Example

[0150] A user from Guangdong made a voice consultation through the personal user terminal of the e-government platform. Due to the noisy environment in which he was located, the clarity of the original sound signal was low.

[0151] After receiving the user's voice signal, the noise analysis and processing module first converts the noise into an image for analysis.

[0152] Next, the Wave-U-Net model is used to modify the original sound signal to improve the clarity and recognition accuracy of the speech.

[0153] Finally, the corrected sound signal is sent to the speech recognition module for recognition, and the system can accurately understand the user's question and give a corresponding response.

[0154] The above examples demonstrate the important role that the noise analysis and processing module plays in blockchain-based e-government platforms. Its efficient noise conversion capabilities and deep learning correction technology improve the accuracy and clarity of speech recognition, providing users with a more convenient and efficient government service experience.

[0155] Example 5

[0156] Intelligent recommendation system.

[0157] System architecture and workflow

[0158] S1. System Architecture

[0159] The intelligent recommendation system, connected to the blockchain service platform terminal, receives and processes user data. The system integrates collaborative filtering algorithms and deep learning models, which work together to generate a list of recommendations based on user data. Ultimately, the system presents the recommended results to the user via the blockchain service platform terminal.

[0160] S2. Workflow

[0161] User data collection: Obtain the user's historical behavior records, preference information, place of origin information, context information, etc. from the blockchain service platform terminal.

[0162] Data preprocessing: Clean, deduplicate, and normalize data to ensure data quality.

[0163] Collaborative filtering recommendation: Calculate user similarity based on the user-item rating matrix and generate a recommendation list.

[0164] Deep learning recommendation: Use deep learning models to learn users' potential interests and generate personalized recommendations.

[0165] Result fusion and sorting: Fusion of collaborative filtering and deep learning recommendation results, and sorting based on weights and algorithms.

[0166] Presentation of recommendation results: The recommendation list is presented to the user through the blockchain service platform terminal.

[0167] S3, algorithm and model training process and numerical values

[0168] Collaborative filtering algorithm training:

[0169] Data preparation: Construct a user-item rating matrix containing 1,000 users and 10,000 items. Each element in the matrix represents the user's rating of the item (e.g., 1-5 points).

[0170] Similarity calculation: The cosine similarity algorithm is used to calculate the similarity between users and obtain a similarity matrix.

[0171] Neighbor selection: Select K=10 nearest neighbors for each user.

[0172] Recommendation generation: Based on the ratings of neighboring users, we predict the target user's ratings for unrated items and generate a list of recommendations. In testing, the accuracy of the top 10 recommended items reached 90%.

[0173] Deep learning model training:

[0174] Data preprocessing: Convert user historical behavior records, preference information, etc. into a format acceptable to the model, such as one-hot encoding or embedding vectors.

[0175] Feature engineering: Extract user features (such as age, gender, and place of origin), project features (such as categories and tags), and context features (such as time and location) to construct feature vectors.

[0176] Model selection: Select a neural network as the deep learning model, specifically a multi-layer perceptron (MLP) or convolutional neural network (CNN).

[0177] Model training: Use the training dataset (e.g., 80% of the data) to train the model and adjust the parameters to minimize the loss function (e.g., cross entropy loss). During training, use a learning rate of 0.001, a batch size of 64, and 100 iterations.

[0178] Model evaluation: Use the validation dataset (e.g., 10% of the data) to evaluate model performance, such as accuracy, recall, F1 score, etc. In the test, the model achieved an accuracy of 85%, a recall of 80%, and an F1 score of 82.5%.

[0179] Model optimization: Based on the evaluation results, adjust the model architecture, parameters, etc. to optimize the model, such as increasing the number of hidden layers and adjusting the number of neurons.

[0180] Hybrid recommendation algorithm training:

[0181] Fusion strategy: Determine the fusion strategy for collaborative filtering and deep learning recommendation results, such as weighted average (weights of 0.6 and 0.4) or ranking fusion.

[0182] Parameter adjustment: Adjust the parameters in the fusion strategy to optimize the recommendation results, such as adjusting the weight ratio, sorting algorithm, etc.

[0183] Performance testing: We used a test dataset (e.g., 10% of the data) to conduct performance testing to evaluate recommendation accuracy and response time. In the test, the hybrid recommendation algorithm achieved a recommendation accuracy of 90% and a response time of less than 0.5 seconds.

[0184] S4. Performance parameters and values

[0185] Recommendation accuracy: The hybrid recommendation algorithm combines collaborative filtering and deep learning, increasing the recommendation accuracy to 90%.

[0186] Response time: The system response time is less than 0.5 seconds, ensuring that users can obtain recommendation results instantly.

[0187] User satisfaction: In the test, user satisfaction with the intelligent recommendation system reached more than 85%.

[0188] S5. Numerical Values ​​and Examples

[0189] Test environment: includes blockchain service platform terminals, intelligent recommendation systems, and users. User data and behavior patterns are based on the statistical characteristics of actual user data.

[0190] Test data: Contains user-item rating matrices and user historical behavior records for 1,000 users and 10,000 items. These data are based on actual user data to ensure that they reflect real-world user behavior.

[0191] Testing process: For each user, a recommendation list is generated and evaluated for accuracy. In the test, the accuracy of the top 10 recommended items reached 90%, and user satisfaction was high.

[0192] Example

[0193] A user from Hunan visited the e-government platform and expressed interest in Hunan's household registration migration policy and a strong interest in education policy. Based on the user's data, the intelligent recommendation system used collaborative filtering to identify similar users (e.g., users from Hunan who were also interested in household registration migration and education policies). It also used a deep learning model to learn about the user's underlying interests (e.g., the user might also be interested in Hunan's employment and healthcare policies). Ultimately, the system combined the recommendations generated by both, recommending multiple pieces of information related to Hunan's education policy and household registration migration policy. The user expressed satisfaction with the recommendations, finding them extremely helpful in understanding local policies and planning their future.

[0194] Intelligent recommendation systems play an important role in blockchain-based e-government platforms. Their efficient recommendation algorithms and precise user profiling technology significantly improve the accuracy of recommendations and user satisfaction.

[0195] In addition, the components designed in the present invention are all universal standard parts or components known to those skilled in the art. Their structures and principles can be known to those skilled in the art through technical manuals or conventional experimental methods. They can be fully implemented by those skilled in the art. Needless to say, the content protected by the present invention does not involve improvements to internal structures and methods.

[0196] The embodiments disclosed in the present invention are preferred embodiments, but are not limited to them. Ordinary technicians in this field can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. As long as they do not deviate from the spirit of the present invention, they are all within the scope of protection of the present invention.

Claims

1. A blockchain-based e-government platform, including a blockchain service platform terminal, characterized in that: The blockchain service platform terminal is connected to the user terminal through the Internet, and the user terminal includes a personal user terminal and a government user terminal; the blockchain service platform terminal is connected to the AI ​​processing center through a high-speed transmission line, and the AI ​​processing center includes a dialect recognition module, an image recognition module, a natural language processing module, and a noise analysis and processing module; the blockchain service platform terminal is also connected to the blockchain network through an encrypted channel to ensure the security and transparency of the data.

2. The blockchain-based e-government platform as claimed in claim 1, characterized in that: The blockchain service platform terminal is connected to the personal native place database, and automatically selects the corresponding dialect recognition model according to the personal native place information of government officials and ordinary people, thereby improving the efficiency and accuracy of dialect recognition.

3. The blockchain-based e-government platform as claimed in claim 1, characterized in that: The dialect recognition module adopts a deep learning algorithm to train and optimize multiple dialects to form multiple dialect recognition models. When receiving user voice, it automatically selects the corresponding dialect recognition model for recognition based on the user's native place information to achieve accurate and fast conversion of dialects.

4. The blockchain-based e-government platform as claimed in claim 1, characterized in that: The image recognition module adopts a convolutional neural network (CNN) algorithm to identify objects, scenes or facial information in pictures or videos uploaded by users. At the same time, the module can be linked with camera devices to realize real-time video stream analysis, assist in lip recognition and gesture recognition, and enhance user interaction experience.

5. The blockchain-based e-government platform as claimed in claim 1, characterized in that: The natural language processing module adopts the Transformer architecture and has functions such as text classification, entity recognition, sentiment analysis and automatic error correction. When the user enters text content, it can combine contextual information and user's place of origin information to provide more accurate grammatical, spelling and semantic error correction suggestions.

6. The blockchain-based e-government platform according to claim 1, characterized in that: The noise analysis and processing module analyzes the noise in the received voice signal and converts the noise into a picture form; the module further uses the big data of the noise picture to perform an overall correction process on the original sound signal through deep learning to improve the accuracy and clarity of voice recognition.

7. The blockchain-based e-government platform according to claim 1, characterized in that: The blockchain service platform terminal is also connected to an intelligent recommendation system, which uses a hybrid recommendation algorithm that combines collaborative filtering and deep learning to accurately recommend relevant services, information or policy information to users based on their historical behavior, preferences, place of origin and contextual information, thereby improving user experience.

8. The blockchain-based e-government platform as claimed in claim 1, characterized in that: The blockchain service platform terminal includes a user information management module, a service request processing module, a service evaluation feedback module and a data statistical analysis module. The user information management module is used to store and manage users' personal information, place of origin information and service records. The service request processing module is used to receive and process users' service requests. The service evaluation feedback module is used to collect users' evaluation and feedback on the services. The data statistical analysis module is used to mine and analyze the data on the platform to provide a basis for platform optimization and service improvement.

9. The blockchain-based e-government platform as claimed in claim 1, characterized in that: The blockchain network is used to store historical records of all transactions, service requests and evaluation feedback on the platform to ensure the immutability and traceability of data. At the same time, through smart contract technology, it realizes the automated execution and settlement of services and improves service efficiency and transparency.

Images