Big data platform intelligent analysis method and system

Abstract

The invention discloses a big data platform intelligent analysis method and system, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting multi-source data, and carrying out the preprocessing; performing state detection on the multi-source data by utilizing the edge computing node; the cloud server extracts and integrates the multi-source features of the multi-source data, calculates the variation amplitude of the multi-source features, screens out consultation features, randomly generates a bit sequence by using a quantum random number generator, randomly selects fragments from the audio data, and arranges the fragments into a quantum sampling evidence set; and predicting the class of the legal scene according to the consultation features through an XGBoost model to obtain a classification result, verifying the classification result of the legal scene in combination with the sound segments of the quantum sampling evidence set, and generating an evidence association report. According to the method, the legal scenes are classified through the XGBoost model, and high-precision classification of the legal scenes is realized in combination with multi-source consultation features.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for intelligent analysis of a big data platform. Background Art

[0002] With the acceleration of the digital transformation of legal consulting services, the application of big data platforms in case analysis, user behavior modeling, and legal scenario classification has gradually become a research hotspot. Existing technologies generally adopt a multi-source data collection framework, which integrates heterogeneous data such as user geographic location, device interaction logs, and voice recordings, and combines cloud computing resources to achieve preliminary classification of legal consulting scenarios. In recent years, the introduction of edge computing technology has provided a low-latency solution for data preprocessing, and machine learning models have significantly outperformed traditional rule engines in feature fusion and scenario prediction. In addition, blockchain technology has been tried in the field of legal evidence storage due to its immutability, but existing solutions mostly rely on single-modal data and lack dynamic correlation analysis and cross-platform verification mechanisms for multi-source heterogeneous data.

[0003] Existing technologies still face multiple challenges in intelligent analysis of big data platforms, especially in the field of legal scenario classification models. The classification accuracy and generalization capabilities are limited when processing complex multi-source data. First, traditional machine learning algorithms find it difficult to effectively integrate the heterogeneous features of multi-source data, resulting in limited classification accuracy; second, traditional machine learning algorithms are sensitive to noise and outliers in the data, and the classification performance further deteriorates in the case of insufficient preprocessing; in addition, traditional machine learning algorithms have weak generalization capabilities and are difficult to adapt to diverse legal scenarios, especially when facing new types of cases, the classification effect is often poor; finally, traditional machine learning algorithms lack the ability to deeply reason about emotional data, which limits the personalization and intelligence level of legal consulting services. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent analysis method for a big data platform to solve the problems of limited classification accuracy and generalization ability when processing complex multi-source data.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides an intelligent analysis method for a big data platform, which includes collecting multi-source data and preprocessing it, wherein the multi-source data includes environmental positioning data, equipment status data and audio data; using edge computing nodes to perform status detection on the multi-source data and generate a preliminary screening report; a cloud server extracts and integrates multi-source features of the multi-source data, calculates the change amplitude of the multi-source features, and screens out consulting features, while using a quantum random number generator to randomly generate bit sequences, randomly selects fragments from the audio data, and organizes them into a quantum sampling evidence set; predicts the legal scenario category based on the consulting features through the XGBoost model to obtain a classification result, and verifies the classification result of the legal scenario in combination with the sound fragments of the quantum sampling evidence set to generate an evidence association report; constructs a bimodal reasoning model to generate legal consulting advice, combines map visualization tools and audio processing tools to generate a legal scenario report, and stores it through blockchain.

[0008] As a preferred solution of the big data platform intelligent analysis method described in the present invention, wherein: the use of edge computing nodes to perform status detection on multi-source data refers to receiving and parsing multi-source data through edge computing hardware devices, and performing validity analysis and scenario classification on each type of data in the multi-source data based on preset rules;

[0009] The rules refer to preset standards used by edge computing nodes to detect the status of multi-source data.

[0010] As a preferred solution of the big data platform intelligent analysis method described in the present invention, the method uses a quantum random number generator to randomly generate a bit sequence, randomly selects fragments from the audio data, and organizes them into a quantum sampling evidence set. The specific steps are as follows:

[0011] Using quantum random number generators to randomly generate sequences of bits by measuring quantum phenomena;

[0012] Group the bit sequence and convert it into index value;

[0013] Map the index value to the sound segment of the audio data, retain the index value that falls within the valid range, and discard the index value that exceeds the valid range;

[0014] According to the mapped index value, the corresponding sound clip is selected from the audio data;

[0015] The selected sound clips are organized into a quantum sampling evidence set.

[0016] As a preferred solution of the big data platform intelligent analysis method described in the present invention, the method of predicting the legal scenario category based on the consultation characteristics through the XGBoost model to obtain the classification result refers to constructing an XGBoost model based on a decision tree, utilizing the collaborative work and gradient optimization of multiple trees, and predicting the prediction probability of the legal scenario category based on the consultation characteristics in multi-source data. The classification result is generated through branch condition judgment, and the classification result is verified in combination with the audio clips in the quantum sampling evidence set to adjust the legal scenario matching degree.

[0017] As a preferred solution of the big data platform intelligent analysis method described in the present invention, the sound verification of the preliminary scene classification result in combination with quantum sampling evidence refers to checking whether the sound clips of the quantum sampling evidence are consistent with the preliminary scene classification results. If they are consistent, the classification confidence is enhanced; if they are contradictory, the classification confidence is reduced. By checking the consistency of the quantum sampling evidence, the legal scene prediction probability is adjusted and the final classification report is output.

[0018] As a preferred solution of the big data platform intelligent analysis method described in the present invention, the construction of a bimodal reasoning model to generate legal consulting advice refers to constructing a bimodal reasoning model based on a dual-branch neural network model, the branches process legal vectors and consulting demand vectors, the fusion layer and the reasoning layer splice and compress the legal vectors and consulting demand vectors, and output legal consulting advice.

[0019] As a preferred embodiment of the big data platform intelligent analysis method of the present invention, the legal vector refers to legal data composed of legal scenario names, predicted probabilities of legal scenarios, and consulting features in multi-source data, which is obtained by converting the legal data through one-hot encoding, normalization, numerical embedding, and a fully connected layer;

[0020] The consultation demand vector refers to the collection of user consultation feedback, processing of user consultation feedback through the BERT model and the Wav2Vec model, and then generated by weighted averaging after dimensionality reduction and normalization.

[0021] In the second aspect, the present invention provides a big data platform intelligent analysis system, including an acquisition module, a detection module, an evidence module, a classification module and an inference module; the acquisition module is used to collect and preprocess multi-source data, and the multi-source data includes environmental positioning data, equipment status data and audio data; the detection module is used to use edge computing nodes to perform status detection on multi-source data and generate a preliminary screening report; the evidence module is used to extract and integrate multi-source features of multi-source data from a cloud server, calculate the variation range of multi-source features, screen out consulting features, and at the same time use a quantum random number generator to randomly generate a bit sequence, randomly select fragments from the audio data, and organize them into a quantum sampling evidence set; the classification module is used to predict the legal scenario category according to the consulting features through the XGBoost model to obtain a classification result, and verify the classification result of the legal scenario in combination with the sound fragments of the quantum sampling evidence set to generate an evidence association report; the inference module is used to construct a bimodal reasoning model to generate legal consulting advice, generate a legal scenario report in combination with a map visualization tool and an audio processing tool, and store it through a blockchain.

[0022] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the big data platform intelligent analysis method described in the first aspect of the present invention is implemented.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the big data platform intelligent analysis method as described in the first aspect of the present invention.

[0024] The present invention achieves high-precision classification of legal scenarios by using the XGBoost model to classify legal scenarios and combining it with multi-source consultation features. The XGBoost model effectively captures nonlinear relationships in complex multi-source data through decision tree branch conditional judgment and weighted score aggregation, significantly improving classification accuracy and generalization capabilities. The initial classification results are verified using quantum sampling evidence, further enhancing the credibility of the classification results. Finally, the XGBoost model supports dynamic adjustment of legal scenario prediction probabilities, ensuring the reliability of the classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 This is a flowchart of the intelligent analysis of the legal consulting service in Example 1.

[0027] Figure 2 This is a schematic diagram of the legal consulting service big data platform in Example 1.

[0028] Figure 3 This is the data flow diagram in Example 1.

[0029] Figure 4 This is the decision logic diagram in Example 1. DETAILED DESCRIPTION

[0030] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0031] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0032] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0033] Example 1, with reference to Figures 1 to 4 , which is the first embodiment of the present invention, provides a big data platform intelligent analysis method, comprising the following steps:

[0034] S1. Collect multi-source data and perform preprocessing.

[0035] Furthermore, the user launches the legal consulting client and requests to enable the hardware function through the operating platform API;

[0036] The legal consulting client pops up a permission request window. After the user clicks "Agree," the authorization timestamp is recorded and the interface indicates that data collection has begun.

[0037] Calling GPS (supporting Beidou / GPS dual-mode, positioning accuracy ≤ 2 meters), continuously monitoring the user's location, recording the latitude and longitude and the number of satellites every 5 minutes, generating environmental positioning data, recording the user's location at the time of consultation (such as home, company, and court), and reflecting the consultation environment;

[0038] The device status sensor is activated simultaneously to record the phone screen status (screen on / off) and touch screen click count at 1Hz. This data is continuously collected until the consultation ends (the user closes the legal consultation client), generating device status data that reflects user operation behavior.

[0039] It should be noted that the device is an electronic device such as a mobile phone or tablet;

[0040] The microphone is activated and a 10-second sound (WAV) is recorded when the user actively triggers or opens the consultation interface to capture spoken questions and generate audio data;

[0041] The legal consulting client connects to the time server via the NTP protocol to calibrate the clock of the user's mobile device every 5 seconds, ensuring that the timestamp accuracy error of all collected multi-source data is ≤1 millisecond. After calibration, a unified timestamp benchmark is generated;

[0042] Save environmental positioning data as CSV files, device status data as JSON arrays, and audio data as WAV files;

[0043] The environmental positioning data, device status data, and audio data are encapsulated into multi-source data packets through Python's struct library.

[0044] S2. Use edge computing nodes to perform status detection on multi-source data and generate a preliminary screening report.

[0045] Furthermore, the legal consulting client transmits multi-source data packets to the edge computing node through a secure network;

[0046] After receiving the multi-source data packets, the edge computing node (a hardware device dedicated to processing data) stores the multi-source data packets in a temporary location (such as a folder on the hard disk) and records the time when the reception is completed;

[0047] Open the multi-source data package, read the description information of the multi-source data package (including user identification, data collection start time and data type mark), and understand what data the multi-source data package contains;

[0048] According to the index table in the multi-source data packet, the environmental positioning data, the device status data and the audio data are extracted respectively;

[0049] It should be noted that the index table is a structured metadata used to describe the content contained in the multi-source data package. It is similar to the directory of a file and lists the identification, location, type and other information of each data file or data block in the multi-source data package to facilitate parsing and extraction.

[0050] Check whether the size of each data file in the multi-source data package meets the minimum requirements (environmental positioning data is larger than 0.5KB, device status data is larger than 1KB, and audio data is larger than 20KB). If any data is missing or too small, mark it as a collection failure and record the problem details (such as missing audio file);

[0051] Review environmental location data and analyze the location distribution of users during consultations. If more than 50% of the records are in legally relevant areas (such as court coordinates), it is marked as a legal scenario, indicating that the consultation is related to a legal institution. Otherwise, it is marked as a regular scenario, indicating that the consultation took place in an everyday place (such as home or work).

[0052] Check device status data to assess the user's device activity during the consultation. If the screen click frequency is less than 5 times / minute or the screen active time is less than 30% (for example, the screen is on for less than 18 seconds in 60 seconds), it is marked as invalid operation, indicating that the user is not actively participating in the consultation. Otherwise, it is marked as normal operation (click frequency ≥ 5 times / minute or active time ≥ 30%), indicating that the user is using the electronic device normally.

[0053] Check the audio data to determine whether it contains valid consultation content. If human voice is detected (frequency range 85-2000Hz, such as stating a question), it is marked as valid audio, indicating that it contains a statement of legal issues. Otherwise, if there is silence or only noise, it is marked as invalid audio, indicating that the audio does not contain valid information.

[0054] Based on the detection results of multi-source data, label the status of each type of data and explain the reasons for the status;

[0055] The environmental positioning data status includes legal scenario or regular scenario, reflecting the scenario type of the consultation;

[0056] Device status data status includes invalid operation or normal operation, reflecting user engagement;

[0057] The audio data status includes valid audio or invalid audio, reflecting the validity of the audio content;

[0058] The status and reasons for the multi-source data are compiled into a preliminary screening report, including user identification, multi-source data processing completion time, and the status and details of the multi-source data (e.g., in legal scenarios, court coordinates account for 70% of the 12 records);

[0059] Based on the initial screening report, classify and process multi-source data;

[0060] If the environmental location data is marked as a legal scenario, it will be retained first; if it is a regular scenario, it will still be retained but marked as non-priority for subsequent user behavior statistics;

[0061] If the device status data is marked as invalid operation and more than 50% of the records (more than 30) are marked, the entire device status data will be deleted, indicating that the user has not effectively participated in the consultation; if it is normal operation or the invalid records are less than 50%, the device status data will be retained;

[0062] If the audio data is marked as invalid audio, the entire audio data is deleted, indicating that there is no valid consultation content; if it is valid audio, the valid audio data is retained;

[0063] Normalize the retained data (all or part of the multi-source data) to unify the numerical range to [0,1], and integrate it with the initial screening report into an edge processing data package;

[0064] When integrating, first write a description (including user identity and multi-source data processing time), and then put the retained multi-source data and preliminary screening report in chronological order to ensure that the content is complete and readable;

[0065] Use Python's query function to check whether the edge processing data packet size is reasonable (at least larger than 1MB). If not, record the edge processing data packet error and reprocess it.

[0066] Edge computing nodes transmit edge-processed data packets through a secure network to a cloud server (a large device dedicated to processing and storing data).

[0067] S3. The cloud server extracts and integrates the multi-source features of the multi-source data, calculates the variation range of the multi-source features, and screens out the consulting features. At the same time, it uses the quantum random number generator to randomly generate bit sequences, randomly selects fragments from the audio data, and organizes them into a quantum sampling evidence set.

[0068] Furthermore, after receiving the edge-processed data packet, the cloud server stores the edge-processed data packet in a processing area within the server (e.g., a working directory on the server's hard disk) and records the time of receipt;

[0069] Open the edge processing data packet, read the metadata of the edge processing data packet, and understand what data the edge processing data packet contains;

[0070] According to the index table in the edge processing data packet, the retained environmental positioning data, device status data and audio data files, as well as the preliminary screening report, are extracted respectively;

[0071] Review each record in the environmental positioning data and calculate the proportion of legally relevant areas (for example, court coordinates account for 60% of the records) and the number of location changes (for example, different longitudes and latitudes appear 3 times in 12 records) as environmental positioning features to reflect the type of consultation scenario;

[0072] Check each record in the device status data and calculate the screen click frequency (times / minute, for example, 6 times / minute) and the percentage of active screen time (for example, 40 seconds on in 60 seconds, accounting for 66%) as device status features to reflect user engagement;

[0073] Using Python audio processing tools (such as Librosa), the 10-second audio is divided into five segments (each 2 seconds). The mean amplitude of each segment (for example, 0.2 for the third segment) and the number of significant peaks (for example, 3) are calculated. Combined with Google Speech-to-Text, these extract human voice keyword features (such as "contract" twice and "help" once) as audio features to reflect the content of the consultation.

[0074] Check all multi-source features (environmental positioning features, device status features, and audio features) one by one. Based on the different value ranges of each feature type, normalize the values of different multi-source features so that all multi-source features fall between 0 and 1.

[0075] Specifically, for each multi-source feature, find the maximum and minimum values (for example, the maximum click frequency is 10 times / minute and the minimum is 1 time / minute), then subtract the minimum value from the current value, and then divide it by the difference between the maximum and minimum values (for example, if the current value is 6 times / minute, then (6-1) / (10-1)=0.56) to get a ratio value between 0 and 1;

[0076] The adjusted environmental positioning features, device status features, and audio features were arranged in order of their numerical values (e.g., percentage, number of occurrences) and directly merged into a multi-source feature set. The contribution of each feature to the user behavior and environmental features was compared. User behavior and environmental features refer to the comprehensive situation reflected in the multi-source data (e.g., the consultation environment is a court, user participation is high, and the audio contains statements of legal issues). The multi-source feature set was simplified and 15 consultation features were selected as valid features.

[0077] Specifically, based on the data of each feature in the multi-source feature set, all values of each feature in the multi-source feature set are extracted from multiple user samples, the variation range of each multi-source feature is calculated (i.e., the maximum value minus the minimum value of the multi-source feature), and representative multi-source features are evaluated by correlation with user behavior and environmental characteristics. Multi-source features with high representativeness are given scores and ranked to screen out the top 15 consulting features;

[0078] Assume that the cloud server has multiple user samples. For each multi-source feature (such as click frequency), take the maximum value (such as 10) and the minimum value (such as 1) of the click frequency feature to calculate the user's click frequency (10-1=9).

[0079] Based on the variation range of multi-source features (i.e., representative features), a multi-source feature score is assigned to each multi-source feature (larger amplitudes give higher scores), and then the features are sorted from largest to smallest by multi-source feature score, and the top 15 are selected;

[0080] A quantum random number generator is used to randomly select 5-second sound clips from the audio data and organize them into a set of legal evidence, which is used as a quantum sampling evidence set.

[0081] Specifically, the audio data contains 10 seconds of sound, divided into 10 small segments (each segment is 1 second, a total of 10 segments);

[0082] Assign a unique number to each 1-second sound, such as 1 to 10;

[0083] Use a quantum random number generator to generate unpredictable random numbers. The quantum random number generator produces a random sequence by measuring quantum phenomena (such as the polarization state of photons). It generates five random numbers for the audio data, ranging from 1 to 10. Unlike traditional pseudo-random number generators, the randomness of the quantum random number generator comes from physical unpredictability.

[0084] Specifically, a quantum random number generator based on photon polarization is used;

[0085] Connect the quantum random number generator to the cloud server;

[0086] Randomly generate a bit stream by measuring the polarization state of photons, generating a total of 20 bits;

[0087] Specifically, a single-photon source was used to emit photons with a polarization state of 45°, emitting 1 million photons per second;

[0088] When a photon is input into a polarization beam splitter, the photon chooses a horizontal polarization path or a vertical polarization path with a 50% probability;

[0089] Single-photon detectors are placed on the horizontal polarization path and the vertical polarization path respectively. Photons detected by the horizontal polarization path are recorded as 0, and photons detected by the vertical polarization path are recorded as 1. Each measurement generates 1 bit;

[0090] Continuously measure 20 photons to generate a 20-bit random bit stream (about 20 microseconds) and record it as a continuous bit sequence;

[0091] Check the bit stream length to ensure it reaches 20 bits. If it is less than 20 bits, continue measuring.

[0092] Divide the bit stream into 5 groups of 4 bits each (range 0-15), convert each group of bit streams into a decimal integer, and map it to the valid range of 1-10 (discard numbers > 10) to ensure that all random numbers are unique;

[0093] The quantum random number generator generates 5 unique random numbers for audio random numbers and records the generation time. If the random number is repeated, it will be regenerated;

[0094] According to the five random numbers generated, select five corresponding segments from the sound clips (for example, numbers 2, 4, 6, 8, and 10), totaling 5 seconds;

[0095] The selected 5-second sound clips were directly organized into a set, recording the source of each sound segment (e.g., the 2nd to 3rd seconds of the sound stating the contract issue), and named it the quantum sampling evidence set;

[0096] The simplified multi-source feature set and quantum sampling evidence set are directly spliced together to form a quantum sampling feature package, which is stored in the database of the cloud server.

[0097] S4. Use the XGBoost model to predict the legal scenario category based on the consultation features to obtain the classification results. Combined with the sound clips of the quantum sampling evidence set, the classification results of the legal scenario are verified to generate an evidence association report.

[0098] Furthermore, the decision tree model is selected as the basic framework of the XGBoost (Extreme Gradient Boosting) model;

[0099] The XGBoost model inputs 15 consultation features (e.g., legal region percentage 60%, click frequency 6 times / minute), and outputs the predicted probability of the legal scenario category (e.g., legal-related consultation, daily consultation);

[0100] The decision tree makes branch condition judgments based on the input consultation features. Each decision tree automatically selects consultation features by optimizing the target (e.g., minimizing the scene classification error through cross entropy loss).

[0101] For example, for the proportion of legal areas, the decision tree found that >50% is the best split point for distinguishing legal-related consultations from routine consultations; for click frequency, it was determined that >5 times / minute is highly correlated with legal-related consultations;

[0102] The branch condition judgment is as follows:

[0103] Tree 1: If the proportion of legal areas is greater than 50%, it indicates a preference for legal-related consultations and outputs a score of 1.

[0104] Tree 2: If click frequency > 5 times / minute and active time > 30%, it tends to be legal-related consultation, and the output score is 1;

[0105] Tree 3: If the legal area ratio is less than 20% and the click frequency is less than 2 times / minute, it tends to be daily consultation and the output score is 0;

[0106] The decision tree scores the currently input consultation features (1 point for legal-related consultations and 0 point for daily consultations), and each decision tree outputs a binary score;

[0107] The XGBoost model aggregates the binary scores of all decision trees by weighting them to generate a raw score (logit value), which is expressed as:

[0108]

[0109] Where logit is the original score, ln is the natural logarithm function, and P is the predicted probability, which represents the predicted probability of a certain legal scenario category;

[0110] The original score is converted into the predicted probability of the legal scenario category through the sigmoid function, and the expression is:

[0111]

[0112] Where e is the base of natural logarithms, a mathematical constant approximately equal to 2.718281828459045;

[0113] For example:

[0114] Assuming logit = 1.5, substitute into the formula to calculate,

[0115] Legal consultations include:

[0116] Daily consultation was: P = 1-0.82 = 0.18 (18%);

[0117] The legal scenario with the highest predicted probability is selected as the preliminary classification result;

[0118] The highest legal scenario is determined based on binary classification. For example, if 0.82 > 0.18, the legal scenario is determined to be legal-related consultation.

[0119] The preliminary classification results are organized into a descriptive file containing the scenario name (e.g., legal-related consultation), the predicted probability of the legal scenario (e.g., 82%), and the main contributing features (determined by the XGBoost model when calculating the predicted probability of the legal scenario category, sorting the 15 consultation features based on their contribution to the decision tree classification, and taking the top 5 (e.g., proportion of legal regions, click frequency)). This file is named the Legal Scenario Classification Report.

[0120] Extract a 5-second sound clip from the quantum sampling evidence set and check whether the quantum sampling evidence set supports the preliminary classification results, for example:

[0121] If the audio is classified as legal consultation (82%), the researchers then check whether the audio contains legal keywords (e.g., contract, litigation, 85-2000Hz). If it does, the prediction probability increases to 90%. If it contains only noise or irrelevant content, the prediction probability drops to 70%.

[0122] If the sound is classified as daily consultation (18%), check whether the sound contains daily statements (such as consultation fees, 85-255Hz) without legal keywords. If it does, the prediction probability increases to 30%. If it contains legal keywords, the prediction probability drops to 10%.

[0123] The legal scenario category refers to the classification of multi-source data into specific legal scenario categories (such as legal-related consultation or daily consultation) through the XGBoost model and quantum sampling evidence set, identifying event types and verifying the prediction accuracy of the XGBoost model;

[0124] Based on the predicted probability of the legal scenario category output by the XGBoost model, a judgment threshold is set, ranging from 0 to 5 seconds, and the consistency of the quantum sampling evidence set and the preliminary classification results of the XGBoost model is checked to ensure the accuracy and reliability of the final legal scenario classification. The judgment threshold is as follows:

[0125] Highly supportive: at least 4-5 seconds of consistent sound, predicted probability +10%-15%;

[0126] Partially supported: 1-3 second correlation, prediction probability ±5%;

[0127] Not supported: 0-1 sound conflict, predicted probability -10%-15%;

[0128] Please set the specific judgment threshold according to your actual situation. This is just for reference.

[0129] Record the relevance of each piece of quantum sampling evidence to the preliminary classification results (e.g., for contract issues in the 2nd and 3rd seconds, indicate relevance to legal-related consultations);

[0130] According to the correlation between each quantum sampling evidence and the preliminary classification result, the matching degree of the legal scenario is adjusted. If the quantum sampling evidence highly supports the classification, the matching degree of the legal scenario is increased; if the quantum sampling evidence does not support it, the matching degree of the legal scenario is reduced.

[0131] If the matching degree of the adjusted legal scenario is less than 60%, it is marked as uncertain legal scenario and a record is made that more consulting features are needed for matching;

[0132] The adjusted classification results and the association information of the quantum sampling evidence are integrated into a file, which contains the final legal scenario name, the final matching degree, the matching feature list and the description of the relevant quantum sampling evidence, and is named as the evidence association report;

[0133] Merge the legal scenario classification report and evidence correlation report into a legal scenario analysis package.

[0134] S5. Build a bimodal reasoning model to generate legal advice, combine map visualization tools and audio processing tools to generate legal scenario reports, and store them on the blockchain.

[0135] Furthermore, the cloud server reads the legal scenario analysis package and quantum sampling feature package from the database, extracting the final legal scenario name, legal scenario prediction probability, main contributing features (the top 5 consulting features among the 15 consulting features), quantum sampling evidence description, 15 consulting features, and the original quantum sampling evidence set;

[0136] The cloud server uses the final legal scenario name, legal scenario prediction probability, and main contributing features to display to the user through the user interface (application or webpage), for example: the scenario classification is legal-related consultation, the prediction probability is 90%, the main features are: legal region accounts for 60%, and the click frequency is 6 times / minute;

[0137] The cloud server prompts users to input consultation feedback through the user interface, including text input (e.g., needing lawyer help) and voice input (recording 10 seconds of voice and analyzing the user's tone), reflecting the user's consultation needs;

[0138] The cloud server records user consultation feedback, processes text feedback through the BERT (bidirectional encoder) model, and processes voice feedback through the Wav2Vec (speech pre-training model) model to generate a 64-dimensional consultation demand vector;

[0139] Specifically, the word segmenter of the BERT model converts text feedback into tokens;

[0140] The input layer converts tokens into word embeddings and adds sentence start [CLS] and sentence end [SEP] tags to form a sequence;

[0141] The Transformer encoder processes the sequence through multi-head self-attention to generate context-enhanced hidden states;

[0142] The pooling layer averages all token hidden states of the last layer of the Transformer encoder to generate 768-dimensional features;

[0143] The fully connected layer receives 768-dimensional features and outputs a 64-dimensional text vector after dimensionality reduction;

[0144] The input layer of the Wav2Vec model receives the raw waveform of 10 seconds of speech feedback and performs inference;

[0145] The convolutional layer converts 10 seconds of speech into a frame sequence;

[0146] The Transformer encoder processes frame sequences through multi-head self-attention, captures temporal relationships, and outputs context-enhanced frame features;

[0147] The pooling layer performs average pooling on all frame features output by the Transformer encoder to generate 768-dimensional features;

[0148] The fully connected layer receives 768-dimensional features and outputs a 64-dimensional speech vector after dimensionality reduction;

[0149] Normalize the 64-dimensional text vector and the 64-dimensional speech vector (e.g., Z-Score normalization) and generate the final 64-dimensional consultation demand vector through weighted averaging.

[0150] The final legal scenario name, legal scenario prediction probability, and main contributing features are encoded using one-hot encoding to generate a 64-dimensional legal vector;

[0151] Specifically, the final legal scenario name is one-hot encoded. Assuming two legal scenario categories (legal-related consultation and daily consultation), the two legal scenario categories are converted into binary vectors using one-hot encoding, and the output encoding is a 2-dimensional vector;

[0152] Normalize the predicted probability of the legal scenario to generate a single value, and encode the output as a 1-dimensional vector;

[0153] The top five consulting features of the main contributing features are numerically embedded and encoded into a 5-dimensional vector;

[0154] The final legal scenario name, legal scenario prediction probability, and main contribution feature encoding dimensions are concatenated into an 8-dimensional initial vector;

[0155] The 8-dimensional initial vector is input into the fully connected layer, and the 8-dimensional initial vector is expanded to 64 dimensions through linear transformation and ReLU activation to generate a 64-dimensional legal vector;

[0156] A dual-branch neural network model is selected as the basic framework of the bimodal reasoning model;

[0157] Define the input layer, bimodal fusion layer, inference layer, and output layer of the bimodal reasoning model;

[0158] The input layer consists of a 64-dimensional legal vector and a 64-dimensional consulting demand vector;

[0159] The legal vector and the consultation demand vector are normalized through the bimodal fusion layer and concatenated through weighting to generate a 128-dimensional fusion vector.

[0160] The inference layer receives a 128-dimensional fusion vector, maps the 128-dimensional vector to a 32-dimensional vector through a fully connected layer, and uses the ReLU activation function to generate inference features.

[0161] It should be noted that the inference layer is a functional description. Its specific implementation depends on the fully connected layer, which is the core component for performing inference.

[0162] The output layer receives 32-dimensional inference features and generates 3-dimensional linear scores through the fully connected layer. The 3-dimensional linear scores are converted into predicted probability distributions using the softmax function to generate predicted probabilities for the three suggestion categories. The suggestion category with the highest predicted probability is converted into a natural language suggestion based on the predefined suggestion template.

[0163] The recommended template is:

[0164] If the legal consultation + legal region ratio is > 50% + the demand is clear, we recommend contacting a professional lawyer;

[0165] If the daily consultation + click frequency is less than 5 times / minute + the demand is vague, the matching suggestion is to solve it yourself or provide additional information;

[0166] If the predicted probability is <80%, the matching suggestion is supplemented with more consultation details;

[0167] Example: The scenario is legal consultation (90%). The user feedback requires legal assistance, and the suggestion is to contact a professional lawyer to handle the legal issue.

[0168] Save the emotional advice as a text file, name it Legal Advice Report, and record the time of its creation;

[0169] Extract user trajectory related features (such as latitude and longitude sequence, timestamp, scene type) from quantum sampling feature packages;

[0170] Call the map visualization tool (based on the OpenStreetMap API) to draw the user's dynamic trajectory;

[0171] Specifically, the starting point and the end point are marked on the map, and the latitude and longitude points are connected to generate the user's dynamic path;

[0172] Use colors to indicate the type of legal scenario (e.g., red for legal-related consultations, green for routine consultations);

[0173] Use audio processing tools (such as Librosa) to load voice data, extract the time-amplitude sequence, draw a sound waveform, and use frequency analysis to mark the key points of the sound (such as the 2nd-3rd second contract). Then embed the sound waveform into the corresponding position of the user's dynamic trajectory on the map.

[0174] By weightedly superimposing the predicted probabilities of the suggested categories and the legal scenario classification results, the final legal scenario name and the predicted probability of the suggested categories are labeled to generate an enhanced visualization trajectory as a legal scenario visualization report and store it in the database.

[0175] Upload the hash values of the legal advice report and the legal scenario visualization report to a blockchain (e.g., Hyperledger) to generate a record of evidence, including the file hash (the two files of the legal advice report and the legal scenario visualization report), user identification, processing time, and the final legal scenario tag;

[0176] Obtain the blockchain transaction ID through the Hyperledger API and record it as a certificate;

[0177] Check whether the legal advice report complies with local laws (e.g., whether the lawyer contact is applicable in the user's region) through the legal rule library;

[0178] Based on the classification performance of the bimodal reasoning model, a credible threshold is set, and the predicted probability of the recommended category output by the bimodal reasoning model is compared with the credible threshold to verify whether the predicted probability of the recommended category reaches the credible level. The value range is [0, 1].

[0179] If the predicted probability of the suggested category is >80% (0.8-1), it is considered highly reliable;

[0180] If the predicted probability of the recommendation category is <60% (0-0.6), it is non-compliant, indicating that the confidence of the recommendation is too low, and it is marked as requiring manual review, and the problem is recorded (such as suggesting additional consulting features);

[0181] The predicted probability of the recommendation category is 60%-80% (0.6-0.8), indicating that the recommendation is of medium confidence. The legal advice report will continue to be generated but marked as requiring caution.

[0182] It should be noted that the specific trust threshold setting should be set according to actual conditions. This is only provided for reference.

[0183] The classification performance is obtained by training a bimodal reasoning model on the legal vector and the consulting demand vector, and by evaluating (cross entropy loss) the consistency between the output of the bimodal reasoning model and the annotated suggested categories;

[0184] Save the evidence credentials and compliance verification results as evidence and compliance reports and store them in the database.

[0185] This embodiment also provides a big data platform intelligent analysis system, including: an acquisition module, a detection module, an evidence module, a classification module and an inference module; the acquisition module is used to collect and preprocess multi-source data, and the multi-source data includes environmental positioning data, equipment status data and audio data; the detection module is used to use edge computing nodes to perform status detection on multi-source data and generate a preliminary screening report; the evidence module is used to extract and integrate multi-source features of multi-source data from a cloud server, calculate the change amplitude of the multi-source features, screen out consulting features, and at the same time use a quantum random number generator to randomly generate a bit sequence, randomly select fragments from the audio data, and organize them into a quantum sampling evidence set; the classification module is used to predict the legal scenario category based on the consulting features through the XGBoost model to obtain a classification result, and verify the classification result of the legal scenario in combination with the sound fragments of the quantum sampling evidence set to generate an evidence association report; the inference module is used to construct a bimodal reasoning model to generate legal consulting advice, combine map visualization tools and audio processing tools to generate a legal scenario report, and store it through blockchain.

[0186] This embodiment also provides a computer device suitable for the case of a big data platform intelligent analysis method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the big data platform intelligent analysis method proposed in the above embodiment.

[0187] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0188] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for implementing intelligent analysis of a big data platform as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

[0189] In summary, this paper achieves high-precision classification of legal scenarios by classifying them using the XGBoost model and combining it with multi-source consultation features. The XGBoost model effectively captures the nonlinear relationships in complex multi-source data through decision tree branch conditional judgment and weighted score aggregation, significantly improving classification accuracy and generalization capabilities. The initial classification results are verified using quantum sampling evidence, further enhancing the credibility of the classification results. Finally, the XGBoost model supports dynamic adjustment of legal scenario prediction probabilities, ensuring the reliability of the classification results.

[0190] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A big data platform intelligent analysis method, characterized by: include, Collecting and preprocessing multi-source data, wherein the multi-source data includes environmental positioning data, device status data, and audio data; Use edge computing nodes to perform status detection on multi-source data and generate preliminary screening reports; The cloud server extracts and integrates multi-source features from multi-source data, calculates the magnitude of change in these features, and selects consulting features. It also uses a quantum random number generator to randomly generate bit sequences, randomly selects segments from the audio data, and compiles them into a quantum sampling evidence set. The XGBoost model is used to predict the legal scenario category based on the consultation features, and the classification results are obtained. The classification results of the legal scenario are verified by combining the sound clips of the quantum sampling evidence set to generate an evidence association report. Build a bimodal reasoning model to generate legal consulting advice, combine map visualization tools and audio processing tools to generate legal scenario reports, and store them through blockchain.

2. The big data platform intelligent analysis method according to claim 1, characterized in that: The use of edge computing nodes to perform status detection on multi-source data refers to receiving and parsing multi-source data through edge computing hardware devices, and performing validity analysis and scenario classification on each type of data in the multi-source data based on preset rules; The rules refer to preset standards used by edge computing nodes to detect the status of multi-source data.

3. The big data platform intelligent analysis method according to claim 1, characterized in that: The method uses a quantum random number generator to randomly generate a bit sequence, randomly selects segments from the audio data, and organizes them into a quantum sampling evidence set. The specific steps are: Using quantum random number generators to randomly generate sequences of bits by measuring quantum phenomena; Group the bit sequence and convert it into index value; Map the index value to the sound segment of the audio data, retain the index value that falls within the valid range, and discard the index value that exceeds the valid range; According to the mapped index value, the corresponding sound clip is selected from the audio data; The selected sound clips are organized into a quantum sampling evidence set.

4. The big data platform intelligent analysis method according to claim 1, characterized in that: The method of predicting the legal scenario category based on the consultation features through the XGBoost model to obtain the classification result refers to constructing an XGBoost model based on a decision tree, utilizing the collaborative work and gradient optimization of multiple trees, predicting the prediction probability of the legal scenario category based on the consultation features in multi-source data, generating the classification result through branch condition judgment, and verifying the classification result in combination with the audio clips in the quantum sampling evidence set to adjust the legal scenario matching degree.

5. The big data platform intelligent analysis method according to claim 1, characterized in that: The sound verification of the preliminary scene classification results in combination with quantum sampling evidence refers to checking whether the sound clips of the quantum sampling evidence are consistent with the preliminary scene classification results. If they are consistent, the classification confidence is enhanced; if they are contradictory, the classification confidence is reduced. By checking the consistency of the quantum sampling evidence, the legal scene prediction probability is adjusted and the final classification report is output.

6. The big data platform intelligent analysis method according to claim 1, characterized in that: The construction of a bimodal reasoning model to generate legal consulting advice refers to constructing a bimodal reasoning model based on a dual-branch neural network model, wherein the branches process legal vectors and consulting demand vectors, and the fusion layer and the reasoning layer splice and compress the legal vectors and consulting demand vectors to output legal consulting advice.

7. The big data platform intelligent analysis method according to claim 6, characterized in that: The legal vector refers to legal data composed of legal scenario names, predicted legal scenario probabilities, and consulting features in multi-source data, which is obtained by transforming the legal data through one-hot encoding, normalization, numerical embedding, and fully connected layers; The consultation demand vector refers to the collection of user consultation feedback, processing of user consultation feedback through the BERT model and the Wav2Vec model, and then generated by weighted averaging after dimensionality reduction and normalization.

8. A big data platform intelligent analysis system, based on the big data platform intelligent analysis method according to any one of claims 1 to 7, characterized in that: Including, acquisition module, detection module, evidence module, classification module and reasoning module; The acquisition module is used to collect and pre-process multi-source data, wherein the multi-source data includes environmental positioning data, device status data and audio data; The detection module is used to use edge computing nodes to perform status detection on multi-source data and generate a preliminary screening report; The evidence module is used by the cloud server to extract and integrate multi-source features of multi-source data, calculate the variation range of the multi-source features, screen out consulting features, and use a quantum random number generator to randomly generate bit sequences, randomly select segments from the audio data, and organize them into a quantum sampling evidence set; The classification module is used to predict the legal scenario category based on the consultation features using the XGBoost model, obtain the classification results, verify the legal scenario classification results in combination with the sound clips of the quantum sampling evidence set, and generate an evidence association report; The reasoning module is used to build a bimodal reasoning model to generate legal consulting advice, combine map visualization tools and audio processing tools to generate legal scenario reports, and store them through blockchain.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the big data platform intelligent analysis method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the big data platform intelligent analysis method according to any one of claims 1 to 7 are implemented.

Images