Policy information acquisition and analysis system based on artificial intelligence

Abstract

The invention relates to the technical field of computers, and discloses an artificial intelligence-based policy information acquisition and analysis system, which comprises a data acquisition module for acquiring historical policy data and real-time policy data; the data processing module is used for constructing a training data set based on the historical policy data and the real-time policy data; the policy prediction module is used for constructing a policy prediction model, inputting the real-time policy feature data set into the policy prediction model and generating future policy data; the derivation industry prediction module is used for constructing a derivation industry prediction model and generating an enterprise planning scheme based on future policy data through the derivation industry prediction model; according to the invention, previous policy data and historical industry data derived during policy change are learned and analyzed through a policy prediction model and an industry derivation prediction model constructed by an artificial intelligence technology, so that targeted market strategies, resource allocation schemes and risk management measures can be formulated for enterprises; therefore, competitive advantages are kept in a rapidly changing market environment.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and more specifically, to a policy information collection and analysis system based on artificial intelligence. Background Art

[0002] The expansion of the Internet and the continuous reduction of big data storage and computing costs have made the sources of contemporary information data extensive. Big data, with its huge volume, diversity, high speed and value, makes data-driven decision-making increasingly important. Nowadays, big data processing and analysis have become an indispensable part of many industry research. In the big data of the modern information society, text data accounts for the largest proportion. The processing of text data information has a wide range of application scenarios. Trend prediction and sentiment analysis based on text data are of great value to businesses, news media, governments and other entities. At present, many cutting-edge studies are based on text big data mining to analyze social media public opinion information. This quantitative modeling and analysis method with the help of knowledge graphs and neural networks has a high accuracy in the prediction results.

[0003] However, predicting future policies based solely on historical data does not provide detailed guidance for a company's future planning. Human analysis and judgment of predicted future policies are still needed to conceive of the company's possible future industry direction planning and changes, which still has many errors affected by human subjective consciousness. Summary of the invention

[0004] The purpose of the present invention is to provide a policy information collection and analysis system based on artificial intelligence in order to solve the above problems.

[0005] The present invention provides a policy information collection and analysis method based on artificial intelligence, comprising the following steps:

[0006] Step 100: Using data collection tools to obtain historical policy data and real-time policy data from multiple source channels, and pre-processing the historical policy data and real-time policy data;

[0007] Step 200: Perform data analysis and feature extraction on the preprocessed historical policy data and real-time policy data to generate a historical policy feature data set and a real-time policy feature data set, and construct a training data set based on the historical policy feature data set and the real-time policy feature data set;

[0008] Step 300: construct a policy prediction model, train the policy prediction model using a training data set, and generate a trained policy prediction model;

[0009] The policy prediction model includes a first input layer, a first hidden layer and a fully connected layer;

[0010] Step 400: Input the real-time policy feature data set into the policy prediction model to generate future policy data;

[0011] Step 500: construct a derivative industry prediction model, and generate an enterprise planning scheme based on future policy data through the derivative industry prediction model;

[0012] The derivative industry prediction model includes a second input layer, a second hidden layer and an output layer.

[0013] As a further optimization solution of the present invention, the preprocessing of the historical policy data and the real-time policy data is specifically as follows:

[0014] Step S1: Data cleaning:

[0015] Step S101, deduplication: using a Python script to check and delete duplicate policy files;

[0016] Step S102, denoising: using regular expressions to remove advertising information or invalid information from the data;

[0017] Step S103, format conversion: converting data in different formats into a unified structured format;

[0018] Step S2: Data classification and labeling:

[0019] Step S201, classification: preliminary classification based on keywords;

[0020] Step S202, labeling: adding labels to the data;

[0021] Step S3, data integration: integrating data from different channels to form a unified data set;

[0022] Step S4, data storage: storing the preprocessed data in a database.

[0023] As a further optimization scheme of the present invention, data analysis and feature extraction are performed on the pre-processed historical policy data and real-time policy data, specifically:

[0024] 1) Text analysis. The analysis steps include:

[0025] Word segmentation: split the text into words or phrases;

[0026] Part-of-speech tagging: identifying the part of speech of each word;

[0027] Named Entity Recognition: Identifying proper nouns in text;

[0028] Topic modeling: using NLP algorithms to find recurring themes or topics in a text collection;

[0029] Sentiment analysis: determine the emotional tendency of the text;

[0030] 2) Data analysis: the analysis steps include:

[0031] Frequency analysis: statistics on the frequency of various policies;

[0032] Central tendency: Calculate the mean, median, and mode of key numerical features in policy documents;

[0033] Correlation analysis: Calculate the correlation coefficients between policy characteristics;

[0034] Co-occurrence analysis: Analyze the co-occurrence of keywords;

[0035] Cluster analysis: grouping policy documents into several categories to identify common features of similar policies;

[0036] Time series analysis: Analyze the time series data of policy releases;

[0037] 3) Feature extraction:

[0038] The extracted feature types include: numerical features, classification features, time series features, and text features;

[0039] All extracted features are serialized and standardized, and a training data set is constructed based on all processed features.

[0040] As a further optimization solution of the present invention, the calculation formula of the first hidden layer is:

[0041] a t =σ(W1·[h t-1 , x t ]+p1);

[0042] b t =σ(W2·[h t-1 , x t ]+p2);

[0043]

[0044]

[0045] d t =σ(W4·[h t-1 , x t ]+p4);

[0046] Where σ is the sigmoid function, ⊙ is the element-wise dot product, W1 is the first weight matrix, and h t-1 is the previous hidden state, x tis the current input, W2 is the second weight matrix, W3 is the third weight matrix, tanh is the hyperbolic tangent activation function, W4 is the fourth weight matrix, p1, p2, p3 and p4 are the first bias vector, the second bias vector, the third bias vector, the fourth bias vector and the fifth bias vector respectively.

[0047] As a further optimization scheme of the present invention, the calculation formula of the fully connected layer is:

[0048] e=w5d t +p5;

[0049] Among them, w5 is the weight matrix of the fully connected layer and p5 is the fifth bias vector.

[0050] As a further optimization scheme of the present invention, the process of training the policy prediction model is as follows:

[0051] Define the loss function. The calculation formula of the loss function is:

[0052]

[0053] Where n is the total number of data points, y i is the ith actual value, is the i-th predicted value;

[0054] Select Adam optimizer;

[0055] Divide the dataset into training, validation and test sets;

[0056] Train the model on the training set and monitor the performance on the validation set to adjust parameters;

[0057] Perform hyperparameter tuning;

[0058] Conduct performance evaluations;

[0059] Perform overfitting monitoring.

[0060] As a further optimization solution of the present invention, the second input layer is used to receive future policy data;

[0061] The calculation formula for the second hidden layer is:

[0062] f t =tanh(W yy f t-1 +W sy x t +P f );

[0063] Among them, f t represents the current hidden state, f t-1 represents the hidden state of the previous time step, xt represents the input of the current time step, W yy Represents the weight matrix from hidden state to hidden state, W sy Represents the weight parameter input to the hidden state, p f is the bias parameter of the second hidden layer, and tanh is the tanh function.

[0064] As a further optimization scheme of the present invention, the calculation formula of the output layer is:

[0065] g t =σ(W g f t +p g );

[0066] Among them, g t represents the output of the tth time step, W g represents the weight parameter of the output layer, p g Represents the bias parameter of the output layer.

[0067] A policy information collection and analysis system based on artificial intelligence, which is used to execute the above-mentioned policy information collection and analysis method based on artificial intelligence, comprising:

[0068] The data collection module uses data collection tools to obtain historical policy data and real-time policy data from multiple sources and pre-processes the historical policy data and real-time policy data;

[0069] The data processing module performs data analysis and feature extraction on the pre-processed historical policy data and real-time policy data, generates a historical policy feature data set and a real-time policy feature data set, and constructs a training data set based on the historical policy feature data set and the real-time policy feature data set;

[0070] The policy prediction module builds a policy prediction model, trains the policy prediction model using a training data set, and generates a trained policy prediction model;

[0071] The policy prediction model includes a first input layer, a first hidden layer and a fully connected layer;

[0072] The calculation formula for the first hidden layer is:

[0073] a t =σ(W1·[h t-1 , x t ]+p1);

[0074] b t =σ(W2·[h t-1 , x t ]+p2);

[0075]

[0076]

[0077] d t =σ(W4·[h t-1 , x t ]+p4);

[0078] Where σ is the sigmoid function, ⊙ is the element-wise dot product, W1 is the first weight matrix, and h t-1 is the previous hidden state, x t is the current input, W2 is the second weight matrix, W3 is the third weight matrix, tanh is the hyperbolic tangent activation function, W4 is the fourth weight matrix, p1, p2, p3 and p4 are the first bias vector, the second bias vector, the third bias vector, the fourth bias vector and the fifth bias vector respectively;

[0079] The calculation formula of the fully connected layer is:

[0080] e=w5d t +p5;

[0081] Among them, w5 is the weight matrix of the fully connected layer, and p5 is the fifth bias vector;

[0082] Input the real-time policy feature dataset into the policy prediction model to generate future policy data;

[0083] Derivative industry forecasting module, builds a derivative industry forecasting model, and generates enterprise planning solutions based on future policy data through the derivative industry forecasting model;

[0084] The derivative industry prediction model includes a second input layer, a second hidden layer and an output layer;

[0085] The second input layer is used to accept future policy data;

[0086] The calculation formula for the second hidden layer is:

[0087] f t =tanh(W yy f t-1 +W sy x t +p f );

[0088] Among them, f t represents the current hidden state, f t-1 represents the hidden state of the previous time step, x t represents the input of the current time step, W yy Represents the weight matrix from hidden state to hidden state, W syRepresents the weight parameter input to the hidden state, p f is the bias parameter of the second hidden layer, tanh is the tanh function;

[0089] The calculation formula of the output layer is:

[0090] g t =σ(W g f t +p g );

[0091] Among them, g t represents the output of the tth time step, W g represents the weight parameter of the output layer, p g Represents the bias parameter of the output layer.

[0092] A storage medium storing non-temporary computer-readable instructions, which, when executed by a computer, can execute the steps in the above-mentioned artificial intelligence-based policy information collection and analysis method.

[0093] The beneficial effect of the present invention is that the policy prediction model and industry derivative prediction model constructed by artificial intelligence technology can learn and analyze the past policy data and the historical industry data derived from policy changes, and can provide scientific planning basis for enterprises. Based on the output of the model, enterprises can formulate targeted market strategies, resource allocation plans and risk management measures, so as to maintain competitive advantages in a rapidly changing market environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 It is a flow chart of a policy information collection and analysis method based on artificial intelligence of the present invention;

[0095] Figure 2 It is a module block diagram of a policy information collection and analysis system based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0096] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only intended to enable those skilled in the art to better understand and implement the subject matter described herein. In addition, the features described relative to some examples may also be combined in other examples.

[0097] like Figure 1 As shown, a policy information collection and analysis method based on artificial intelligence includes the following steps:

[0098] Step 100: Using data collection tools to obtain historical policy data and real-time policy data from multiple source channels, and pre-processing the historical policy data and real-time policy data;

[0099] In one embodiment of the present invention, the data collection tool includes a web crawler and an API interface;

[0100] Web crawlers: Use crawler technology to automatically collect policy-related data from the Internet;

[0101] API interface: Use public API interfaces to obtain structured data, such as news API, social media API, etc.

[0102] Manual collection can also be adopted. Important documents and reports in specific fields can be collected manually;

[0103] Data collection requires obtaining comprehensive and accurate policy-related information from multiple channels. The main data sources include:

[0104] Official government websites: policy documents, announcements, press releases, etc. issued by the government;

[0105] News media: policy interpretations and analysis articles published by major news websites;

[0106] Social media: public discussions, expert comments, etc. on social platforms;

[0107] Academic database: academic papers, research reports, policy research documents, etc.;

[0108] Industry reports: reports and white papers issued by industry associations;

[0109] For example, suppose you need to collect information about new energy policies. The specific operations are as follows:

[0110] Official government websites: Visit the official website of the National Energy Administration and collect all documents on new energy policies released in recent years;

[0111] News media: Use crawlers to capture news reports on new energy policies from media websites such as People's Daily and Xinhua News Agency;

[0112] Social media: collect tweets and discussions related to “new energy policy” through Twitter API;

[0113] Academic databases: Search for academic papers and research reports related to "new energy policy" in academic databases such as Google Scholar and CNKI;

[0114] Industry reports: Download the latest new energy industry development report from the International Energy Agency (IEA) official website;

[0115] In one embodiment of the present invention, the historical policy data and the real-time policy data are preprocessed, specifically:

[0116] 1) Data cleaning:

[0117] Deduplication: Delete duplicate policy documents, news reports, etc. You can use Python scripts to check and delete duplicate policy documents;

[0118] Denoising: Remove irrelevant information, such as advertisements and meaningless comments. Regular expressions can be used to remove advertising information from the data.

[0119] Format conversion: convert data in different formats into a unified structured format, such as converting a policy document in PDF format to text format, using a PDF parsing library such as PyPDF2;

[0120] 2) Data classification and labeling:

[0121] Classification: preliminary classification based on keywords, such as policy type, issuing agency, time, etc. For example, new energy policies can be divided into subsidy policies, technology research and development support policies, etc.;

[0122] Annotation: Add labels to data, such as using NLP technology to extract policy themes and key influencing factors for subsequent analysis;

[0123] 3) Data Integration:

[0124] Integrate data from different channels to form a unified data set; during the integration process, attention should be paid to data deduplication and consistency processing;

[0125] 4) Data storage:

[0126] Store the preprocessed data in a database to facilitate subsequent query and analysis; commonly used databases include relational databases (such as MySQL) and NoSQL databases (such as MongoDB);

[0127] Step 200: Perform data analysis and feature extraction on the preprocessed historical policy data and real-time policy data to generate a historical policy feature data set and a real-time policy feature data set, and construct a training data set based on the historical policy feature data set and the real-time policy feature data set;

[0128] Among them, data analysis and feature extraction are performed on the pre-processed historical policy data and real-time policy data, specifically:

[0129] 1) Text analysis:

[0130] The analysis steps include:

[0131] Word segmentation: divide the text into words or phrases; this helps with subsequent annotation and classification;

[0132] Part-of-speech tagging: Identify the part of speech of each word (noun, verb, etc.); helps to filter out important information units;

[0133] Named Entity Recognition: Identify proper nouns in text, such as places, names of people, names of organizations, etc.

[0134] Topic modeling: using NLP algorithms to find recurring themes or topics in a text collection;

[0135] Sentiment analysis: determining the emotional tendency of a text, such as positive, negative, or neutral;

[0136] For example, suppose there is a series of documents about new energy policies, which can be analyzed as follows:

[0137] Word segmentation: Break down long sentences in policy documents into individual words or phrases, paying special attention to policy-related professional terms such as "renewable energy" and "carbon emissions";

[0138] Part-of-speech tagging: tagging the parts of speech of words in documents, and selecting important verbs and nouns to understand the action direction and focus of the policy;

[0139] Named Entity Recognition: Identify all specific locations, policy names, company names, etc. mentioned, providing a basis for subsequent relationship analysis and impact assessment;

[0140] Topic modeling: extract the most common topics in the document set through statistical methods. For example, you may find that "green energy subsidies" and "carbon trading market" are hot topics.

[0141] Sentiment analysis: Conduct sentiment analysis on policy documents to determine the government’s support for a new energy project. Enthusiasm may indicate further support measures.

[0142] It should be noted that text analysis can help us understand the main content, objectives and expected impact of a policy, evaluate the emotional tone of the text, and judge the positivity or negativity of the policy. At the same time, by analyzing the associations between keywords, it can reveal the implicit relationships in policy documents.

[0143] 2) Data analysis:

[0144] Descriptive Statistics:

[0145] Frequency analysis: Count the frequency of various policies and understand the release of different types of policies;

[0146] Central tendency: Calculate the mean, median, mode, etc. of key numerical features in policy documents;

[0147] Association analysis

[0148] Correlation analysis: Calculate the correlation coefficients between policy characteristics and analyze their linear relationships;

[0149] Co-occurrence analysis: Analyze the co-occurrence of keywords to reveal the implicit structure in the policy text;

[0150] Advanced Analytics

[0151] Cluster analysis: grouping policy documents into several categories to identify common features of similar policies;

[0152] Regression analysis: Use numerical and categorical features to build a regression model to predict future policy releases;

[0153] Time series analysis: Analyze the time series data of policy releases to identify seasonal and trend changes;

[0154] For example, for new energy policy data, the following data analysis can be performed:

[0155] Frequency analysis: Count the frequency of release of various new energy policies to find out which policies are most common;

[0156] Correlation analysis: Calculate the correlation between the subsidy amount and the policy implementation effect, and find out the key factors affecting the policy effect;

[0157] Cluster analysis: cluster the policies released in different time periods and analyze the emphasis and trend changes of policies in different time periods;

[0158] Regression analysis: Use historical data to build models to predict the number and type of new energy policies that may be released in the future.

[0159] Time series analysis: Analyze the time series data of policy releases to find the peak and trough periods of policy releases, as well as possible cyclical changes;

[0160] 3) Feature extraction:

[0161] The types of features extracted include:

[0162] Numerical features: numerical information related to the policy, such as the amount of funds, execution time, etc.;

[0163] Categorical features: categorical information in the text, such as policy type, implementation area, target group, etc.;

[0164] Time series characteristics: the time when policy documents are released helps to analyze the timeliness and trends of policies;

[0165] Text features: features extracted based on text content, such as keyword frequency, topic distribution, etc.

[0166] For example, continue to take the new energy policy as an example to perform feature extraction:

[0167] Numerical features: extract numerical features such as subsidy amount, tax reduction amount, number of supported projects, etc. involved in policy documents;

[0168] Classification features: Classify policy documents according to policy types (such as "subsidy policy", "tax incentive policy", etc.), and use the industries involved in the policies (such as "wind energy", "solar energy", etc.) as classification features;

[0169] Time series features: record the specific date of policy release and effective time, and analyze the time patterns and trends of policy implementation;

[0170] Text features: extract high-frequency keywords and their occurrence frequency from word segmentation and part-of-speech tagging results; extract the topic distribution of documents from topic modeling results;

[0171] All extracted features are serialized and standardized, and a training data set is constructed based on all processed features;

[0172] Data serialization: converting policy text into word embedding or other numerical forms; for example, using pre-trained word vectors (such as Word2Vec, GloVe) or custom word vectors;

[0173] Standardization: Standardize the numerical features so that their mean is 0 and their variance is 1;

[0174] Time series construction: construct data into sequences based on time order.

[0175] It should be noted that feature extraction can extract the most informative features from a large amount of data, simplify the complexity of the model, provide more predictive variables, improve the performance of the model on unknown data, and select key variables that can clearly explain policy decisions.

[0176] Step 300: construct a policy prediction model, train the policy prediction model using a training data set, and generate a trained policy prediction model;

[0177] The policy prediction model includes a first input layer, a first hidden layer and a fully connected layer;

[0178] The calculation formula for the first hidden layer is:

[0179]

[0180]

[0181]

[0182] d t =σ(W4·[h t-1 , x t ]+p4)

[0183] Where σ is the sigmoid function, ⊙ is the element-wise dot product, W1 is the first weight matrix, and h t-1 is the previous hidden state, x t is the current input, W2 is the second weight matrix, W3 is the third weight matrix, tanh is the hyperbolic tangent activation function, W4 is the fourth weight matrix, p1, p2, p3 and p4 are the first bias vector, the second bias vector, the third bias vector, the fourth bias vector and the fifth bias vector respectively;

[0184] The calculation formula of the fully connected layer is:

[0185] e=w5d t +p5

[0186] Among them, W5 is the weight matrix of the fully connected layer, and p5 is the fifth bias vector;

[0187] The process of training the policy prediction model is:

[0188] Define the loss function. The calculation formula of the loss function is:

[0189]

[0190] Where n is the total number of data points, y i is the ith actual value, is the i-th predicted value;

[0191] Select Adam optimizer;

[0192] Divide the dataset into training, validation and test sets;

[0193] Train the model on the training set and monitor the performance on the validation set to adjust parameters;

[0194] Perform hyperparameter tuning: such as learning rate, batch size, number of iterations, etc. to optimize model performance; perform performance evaluation: use the test set to evaluate the model's accuracy, recall rate, F1 score and other performance indicators; perform overfitting monitoring: prevent overfitting by introducing techniques such as Dropout and L2 regularization.

[0195] Step 400: Input the real-time policy feature data set into the policy prediction model to generate future policy data;

[0196] Step 500: construct a derivative industry prediction model, and generate an enterprise planning scheme based on future policy data through the derivative industry prediction model;

[0197] The derivative industry prediction model includes a second input layer, a second hidden layer and an output layer;

[0198] The second input layer is used to accept future policy data;

[0199] The calculation formula for the second hidden layer is:

[0200] f t =tanh(W yy f t-1 +W sy x t +p f )

[0201] Among them, f t represents the current hidden state, f t-1 represents the hidden state of the previous time step, x t represents the input of the current time step, W yy Represents the weight matrix from hidden state to hidden state, W sy Represents the weight parameter input to the hidden state, p f is the bias parameter of the second hidden layer, tanh is the tanh function;

[0202] The calculation formula of the output layer is:

[0203] g t =σ(W g f t +p g )

[0204] Among them, g t represents the output of the tth time step, W g represents the weight parameter of the output layer, p g Represents the bias parameter of the output layer.

[0205] It should be noted that the output of the derivative industry forecasting model is decoded to obtain the enterprise planning solution.

[0206] In one embodiment of the present invention, the training process of the derivative industry prediction model is as follows:

[0207] Obtain historical data related to the industry, including economic indicators, records of policy changes, technological development levels, and market demand;

[0208] Extract features useful for forecasting models from industry-related historical data, such as industry growth data before and after policy implementation;

[0209] Serialization: construct a time series dataset, where each data point includes the selected features and the corresponding time label;

[0210] Divide data: Divide the data into training set, validation set and test set;

[0211] Select the loss function and optimizer; the commonly used loss function is mean square error (MSE), and the optimizer is SGD, Adam, etc.

[0212] Model training: Train the model using the training set data and evaluate performance on the validation set to monitor overfitting and adjust model parameters.

[0213] like Figure 2 As shown, a policy information collection and analysis system based on artificial intelligence is used to execute the policy information collection and analysis method based on artificial intelligence as described above, including:

[0214] The data collection module uses data collection tools to obtain historical policy data and real-time policy data from multiple sources and pre-processes the historical policy data and real-time policy data;

[0215] The data processing module performs data analysis and feature extraction on the pre-processed historical policy data and real-time policy data, generates a historical policy feature data set and a real-time policy feature data set, and constructs a training data set based on the historical policy feature data set and the real-time policy feature data set;

[0216] The policy prediction module builds a policy prediction model, trains the policy prediction model using a training data set, and generates a trained policy prediction model;

[0217] The policy prediction model includes a first input layer, a first hidden layer and a fully connected layer;

[0218] The calculation formula for the first hidden layer is:

[0219] a t =σ(W1·[h t-1 , x t ]+p1)

[0220] b t =σ(W2·[h t-1 , x t ]+p2)

[0221]

[0222] d t =σ(W4·[h t-1 , x t ]+p4)

[0223] Where σ is the sigmoid function, ⊙ is the element-wise dot product, W1 is the first weight matrix, and h t-1 is the previous hidden state, x t is the current input, W2 is the second weight matrix, W3 is the third weight matrix, tanh is the hyperbolic tangent activation function, W4 is the fourth weight matrix, p1, p2, p3 and p4 are the first bias vector, the second bias vector, the third bias vector, the fourth bias vector and the fifth bias vector respectively;

[0224] The calculation formula of the fully connected layer is:

[0225] e=w5d t +p5

[0226] Among them, W5 is the weight matrix of the fully connected layer, and p5 is the fifth bias vector;

[0227] Input the real-time policy feature dataset into the policy prediction model to generate future policy data;

[0228] Derivative industry forecasting module, builds a derivative industry forecasting model, and generates enterprise planning solutions based on future policy data through the derivative industry forecasting model;

[0229] The derivative industry prediction model includes a second input layer, a second hidden layer and an output layer;

[0230] The second input layer is used to accept future policy data;

[0231] The calculation formula for the second hidden layer is:

[0232] f t =tanh(W yy f t-1 +W sy x t +p f )

[0233] Among them, f t represents the current hidden state, f t-1 represents the hidden state of the previous time step, x t represents the input of the current time step, W yy Represents the weight matrix from hidden state to hidden state, W sy Represents the weight parameter input to the hidden state, p f is the bias parameter of the second hidden layer, tanh is the tanh function;

[0234] The calculation formula of the output layer is:

[0235] g t =σ(W g f t +pg )

[0236] Among them, g t represents the output of the tth time step, W g represents the weight parameter of the output layer, p g Represents the bias parameter of the output layer.

[0237] A storage medium storing non-temporary computer-readable instructions, which, when executed by a computer, can execute the steps in the above-mentioned artificial intelligence-based policy information collection and analysis system.

[0238] The present embodiment is described above, but the present embodiment is not limited to the above-mentioned specific implementation manner. The above-mentioned specific implementation manner is merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, which are all within the protection of the present embodiment.

Claims

1. A policy information collection and analysis method based on artificial intelligence, characterized in that: The following steps are involved: Step 100: Using data collection tools to obtain historical policy data and real-time policy data from multiple source channels, and pre-processing the historical policy data and real-time policy data; Step 200: Perform data analysis and feature extraction on the preprocessed historical policy data and real-time policy data to generate a historical policy feature data set and a real-time policy feature data set, and construct a training data set based on the historical policy feature data set and the real-time policy feature data set; Step 300: construct a policy prediction model, train the policy prediction model using a training data set, and generate a trained policy prediction model; The policy prediction model includes a first input layer, a first hidden layer and a fully connected layer; Step 400: Input the real-time policy feature data set into the policy prediction model to generate future policy data; Step 500: construct a derivative industry prediction model, and generate an enterprise planning scheme based on future policy data through the derivative industry prediction model; The derivative industry prediction model includes a second input layer, a second hidden layer and an output layer.

2. According to the artificial intelligence-based policy information collection and analysis method of claim 1, it is characterized in that: The preprocessing of the historical policy data and the real-time policy data is specifically as follows: Step S1: Data cleaning: Step S101, deduplication: using a Python script to check and delete duplicate policy files; Step S102, denoising: using regular expressions to remove advertising information or invalid information from the data; Step S103, format conversion: converting data in different formats into a unified structured format; Step S2: Data classification and labeling: Step S201, classification: preliminary classification based on keywords; Step S202, labeling: adding labels to the data; Step S3, data integration: integrating data from different channels to form a unified data set; Step S4, data storage: storing the preprocessed data in a database.

3. According to claim 2, a policy information collection and analysis method based on artificial intelligence is characterized in that: Perform data analysis and feature extraction on the preprocessed historical policy data and real-time policy data, specifically: 1) Text analysis. The analysis steps include: Word segmentation: split the text into words or phrases; Part-of-speech tagging: identifying the part of speech of each word; Named Entity Recognition: Identifying proper nouns in text; Topic modeling: using NLP algorithms to find recurring themes or topics in a text collection; Sentiment analysis: determine the emotional tendency of the text; 2) Data analysis: the analysis steps include: Frequency analysis: statistics on the frequency of various policies; Central tendency: Calculate the mean, median, and mode of key numerical features in policy documents; Correlation analysis: Calculate the correlation coefficients between policy characteristics; Co-occurrence analysis: Analyze the co-occurrence of keywords; Cluster analysis: grouping policy documents into several categories to identify common features of similar policies; Time series analysis: Analyze the time series data of policy releases; 3) Feature extraction: The extracted feature types include: numerical features, classification features, time series features, and text features; All extracted features are serialized and standardized, and a training data set is constructed based on all processed features.

4. According to claim 3, a policy information collection and analysis method based on artificial intelligence is characterized in that: The calculation formula for the first hidden layer is: a t =σ(W1·[h t-1 ,x t ]+p1); b t =σ(W2·[h t-1 ,x t ]+p2); d t =σ(W4·[h t-1 ,x t ]+p4); Where σ is the sigmoid function, ⊙ is the element-wise dot product, W1 is the first weight matrix, and h t-1 is the previous hidden state, x t is the current input, W2 is the second weight matrix, W3 is the third weight matrix, tanh is the hyperbolic tangent activation function, W4 is the fourth weight matrix, p1, p2, p3 and p4 are the first bias vector, the second bias vector, the third bias vector, the fourth bias vector and the fifth bias vector respectively.

5. According to the artificial intelligence-based policy information collection and analysis method of claim 4, it is characterized in that: The calculation formula of the fully connected layer is: e=w5d t +p5; Among them, w5 is the weight matrix of the fully connected layer and p5 is the fifth bias vector.

6. The artificial intelligence-based policy information collection and analysis method according to claim 5 is characterized in that: The process of training the policy prediction model is: Define the loss function. The calculation formula of the loss function is: Where n is the total number of data points, y i is the ith actual value, is the i-th predicted value; Select Adam optimizer; Divide the dataset into training, validation and test sets; Train the model on the training set and monitor the performance on the validation set to adjust the parameters; Perform hyperparameter tuning; Conduct performance evaluations; Perform overfitting monitoring.

7. The artificial intelligence-based policy information collection and analysis method according to claim 6 is characterized in that: The second input layer is used to accept future policy data; The calculation formula for the second hidden layer is: f t =tanh(W yy f t-1 +W sy x t +p f ); Among them, f t represents the current hidden state, f t-1 represents the hidden state of the previous time step, x t represents the input of the current time step, W yy Represents the weight matrix from hidden state to hidden state, W sy Represents the weight parameter input to the hidden state, p f is the bias parameter of the second hidden layer, and tanh is the tanh function.

8. The artificial intelligence-based policy information collection and analysis method according to claim 7 is characterized in that: The calculation formula of the output layer is: g t =σ(W g f t +p g ); Among them, g t represents the output of the tth time step, W g represents the weight parameter of the output layer, p g Represents the bias parameter of the output layer.

9. A policy information collection and analysis system based on artificial intelligence, characterized in that: It is used to execute the policy information collection and analysis method based on artificial intelligence as described in any one of claims 1 to 8, including: The data collection module uses data collection tools to obtain historical policy data and real-time policy data from multiple sources and pre-processes the historical policy data and real-time policy data; The data processing module performs data analysis and feature extraction on the pre-processed historical policy data and real-time policy data, generates a historical policy feature data set and a real-time policy feature data set, and constructs a training data set based on the historical policy feature data set and the real-time policy feature data set; The policy prediction module builds a policy prediction model, trains the policy prediction model using a training data set, and generates a trained policy prediction model; The policy prediction model includes a first input layer, a first hidden layer and a fully connected layer; The calculation formula for the first hidden layer is: a t =σ(W1·[h t-1 ,x t ]+p1); b t =σ(W2·[h t-1 ,x t ]+p2); d t =σ(W4·[h t-1 ,x t ]+p4); Where σ is the sigmoid function, ⊙ is the element-wise dot product, W1 is the first weight matrix, and h t-1 is the previous hidden state, x t is the current input, W2 is the second weight matrix, W3 is the third weight matrix, tanh is the hyperbolic tangent activation function, W4 is the fourth weight matrix, p1, p2, p3 and p4 are the first bias vector, the second bias vector, the third bias vector, the fourth bias vector and the fifth bias vector respectively; The calculation formula of the fully connected layer is: e=w5d t +p5; Among them, w5 is the weight matrix of the fully connected layer, and p5 is the fifth bias vector; Input the real-time policy feature dataset into the policy prediction model to generate future policy data; Derivative industry forecasting module, builds a derivative industry forecasting model, and generates enterprise planning solutions based on future policy data through the derivative industry forecasting model; The derivative industry prediction model includes a second input layer, a second hidden layer and an output layer; The second input layer is used to accept future policy data; The calculation formula for the second hidden layer is: f t =tanh(W yy f t-1 +W sy x t +p f ); Among them, f t represents the current hidden state, f t-1 represents the hidden state of the previous time step, x t represents the input of the current time step, W yy Represents the weight matrix from hidden state to hidden state, W sy Represents the weight parameter input to the hidden state, p f is the bias parameter of the second hidden layer, tanh is the tanh function; The calculation formula of the output layer is: g t =σ(W g f t +p g ); Among them, g t represents the output of the tth time step, W g represents the weight parameter of the output layer, p g Represents the bias parameter of the output layer.

10. A storage medium, characterized in that: It stores non-temporary computer-readable instructions, which, when executed by a computer, can execute the steps in an artificial intelligence-based policy information collection and analysis method as described in any one of claims 1-8.

Images