Public policy engagement evaluation method and system based on lda and vector space model

Abstract

The application provides a public policy participation evaluation method and system based on an LDA and a vector space model, and the method comprises the following steps: obtaining target public policy documents and target dynamic data; adopting a TF-IDF algorithm to perform word segmentation processing on the target public policy documents and the target dynamic data respectively, so as to obtain a document feature word segmentation set and a data feature word segmentation set; extracting semantic information hidden in the target public policy documents and the target dynamic data respectively through an LDA model, so as to obtain a document-data matrix; constructing a vector space model based on the document feature word segmentation set and the data feature word segmentation set, calculating the similarity between the target public policy documents and the target dynamic data, so as to obtain a similarity matrix; and performing linear weighting processing on the document-data matrix and the similarity matrix. The application can effectively evaluate the citizen participation in public policy, and also solves the problems of data sparseness and semantic loss, and improves the accuracy and effectiveness of the evaluation.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for evaluating public policy participation based on LDA and vector space models. Background Technology

[0002] Public policy evaluation refers to the comprehensive judgment and assessment made by an evaluation body based on certain standards and procedures, using scientific evaluation tools and methods, regarding the necessity, feasibility, cost-benefit analysis, and implementation effects of a policy. Its purpose is to reduce decision-making risks, improve policy implementation efficiency, adjust and improve unreasonable or imperfect policy initiation and implementation processes in a timely manner, and establish theoretical basis and experience comparison for the formulation of new policies. Scientific evaluation procedures and methods are the technical guarantee for carrying out public policy evaluation work. Citizen participation is indispensable in the process of public policy evaluation. Citizen participation in the system and implementation of public policies can enhance citizens' understanding and awareness of public policies, national policies, and social development, and can also provide guarantees for the implementation of public policies. Therefore, evaluating the degree of citizen participation in public policy is particularly important.

[0003] Currently, using traditional text clustering methods from data mining to detect topics in online text remains the mainstream approach, and many related market products have emerged. However, traditional text clustering algorithms do not perform ideally in terms of data dimensionality reduction and semantic representation, thus affecting the accuracy and effectiveness of the evaluation results. Summary of the Invention

[0004] In view of this, this application proposes a public policy participation evaluation method and system based on LDA and vector space model, which effectively solves the problems of data sparsity and semantic loss in traditional text clustering methods, and can effectively evaluate citizen participation in public policies.

[0005] The technical solution of this invention is implemented as follows:

[0006] This invention proposes a public policy participation assessment method based on LDA and vector space models, the method comprising:

[0007] S1, obtain the target public policy document, and at the same time obtain the dynamic data within a preset period after the release of the target public policy document, as the target dynamic data;

[0008] S2, the TF-IDF algorithm is used to segment the target public policy document and the target dynamic data respectively to obtain the document feature segmentation set and the data feature segmentation set;

[0009] S3. Establish an LDA model and extract the semantic information implied in the target public policy documents and target dynamic data through the LDA model to obtain the document-data matrix.

[0010] S4. Based on the document feature word segmentation set and the data feature word segmentation set, construct a vector space model, calculate the similarity between the target public policy document and the target dynamic data, and obtain the similarity matrix;

[0011] S5 performs linear weighting on the document-data matrix and the similarity matrix to obtain a combined matrix that can map citizens' attention to public policies within a preset period.

[0012] Based on the above technical solutions, preferably, step S2 specifically includes:

[0013] The TF-IDF algorithm is an improved feature weighting algorithm, and its calculation formula is as follows:

[0014]

[0015] Where W(t,d) represents the weight of t in d, tf(t,d) is the frequency of t in d, d is a document in the text corpus C, t is a word in document d, and C = {c1,c2,…,c...} m}, c i ∈C, m is the number of categories in the text corpus, G represents the total number of documents in the text corpus, n t W(w) represents the number of texts t contained in the text corpus. k ,c i ) indicates the word w k Regarding category c i Category weights, w k c represents the k-th word. i This represents the i-th type of article;

[0016] The target public policy document is segmented using the TF-IDF algorithm, resulting in an M×N matrix, which is the document feature segmentation set V. M×N M represents the number of texts in the target public policy document, and N represents the number of words in each text.

[0017] The target dynamic data is segmented using the TF-IDF algorithm, resulting in a P×Q matrix, which is the data feature segmentation set V. P×Q P represents the number of texts in the target dynamic data, and Q represents the number of words in each text.

[0018] Based on the above technical solutions, preferably, step S3 specifically includes:

[0019] A first LDA model is established, and the target public policy documents are used as the corpus for training the first LDA model.

[0020] A second LDA model is established, and the target dynamic data is used as the corpus of the second LDA model for training.

[0021] The first LDA model is used to extract the hidden semantic information of the target public policy document, obtain the first word distribution matrix, and then calculate the first topic distribution matrix. The product of the first word distribution matrix and the first topic distribution matrix is ​​used as the document distribution matrix.

[0022] The second LDA model is used to extract the implicit semantic information in the target dynamic data, obtain the second word distribution matrix, and then calculate the second topic distribution matrix. The product of the second word distribution matrix and the second topic distribution matrix is ​​used as the data distribution matrix.

[0023] Multiplying the document distribution matrix by the data distribution matrix yields the document-data matrix.

[0024] Based on the above technical solutions, the preferred steps for establishing an LDA model specifically include:

[0025] S301, Randomly initialize the LDA model. For each word w in each document in the corpus, randomly assign a topic number z. The total number of topics is K.

[0026] S302, rescan the corpus, for each word w, resample its topic according to the Gibbs sampling formula, and update the corpus;

[0027] S303, Repeat the resampling process of S302 until the value of the Gibbs sampling formula converges;

[0028] S304, statistically analyze the distribution matrix of words under a topic in the corpus, and then calculate the topic distribution matrix of the corresponding document.

[0029] Based on the above technical solutions, preferably, step S304 specifically includes:

[0030] The probability distribution of words under a given topic is:

[0031]

[0032] Among them, W k For the k-th word in the document, T j If the word distribution matrix represents the j-th topic in the corpus, then it is as follows:

[0033]

[0034] The probability distribution of topics in a document is as follows:

[0035]

[0036] Among them, T j For the j-th topic in the corpus, D i For the i-th document in the corpus, the topic distribution matrix is ​​represented as:

[0037]

[0038] Based on the above technical solutions, the preferred method of multiplying the document distribution matrix and the data distribution matrix to obtain the document-data matrix specifically includes:

[0039] The product of the first word distribution matrix and the first topic distribution matrix is ​​taken as the document distribution matrix, and the expression of the document distribution matrix is:

[0040] Z′=X′·Y′

[0041] Where Z′ represents the document distribution matrix, X′ represents the first word distribution matrix, and Y′ represents the first topic distribution matrix;

[0042] The product of the second word distribution matrix and the second topic distribution matrix is ​​taken as the data distribution matrix, and the expression of the data distribution matrix is:

[0043] Z″=X″·Y″

[0044] Where Z″ represents the document distribution matrix, X″ represents the first word distribution matrix, and Y″ represents the first topic distribution matrix;

[0045] The document distribution matrix is ​​multiplied by the data distribution matrix to obtain the document-data matrix. The expression for the document-data weights is as follows:

[0046] R = Z′·Z″

[0047] Where R represents the document-data matrix, Z′ represents the document distribution matrix, and Z″ represents the document distribution matrix.

[0048] Based on the above technical solutions, preferably, step S4 specifically includes:

[0049] Calculate the similarity between the document feature word segmentation set and the data feature word segmentation set to obtain the similarity matrix, the expression of which is:

[0050]

[0051] Among them, s mp For document feature word segmentation set V M×N The m-th row and the data feature word segmentation set V P×QThe cosine similarity of the p-th row, m = 1, 2, ..., M, p = 1, 2, ..., P.

[0052] Based on the above technical solutions, preferably, step S5 specifically includes:

[0053] The document-data matrix and similarity matrix are linearly weighted to obtain the combination matrix Q, which is calculated using the following formula:

[0054] Q = θ × R + (1 - θ) × E

[0055] Where θ is the linear combination coefficient, R is the document-data matrix, and E is the similarity matrix.

[0056] Based on the above technical solutions, preferably, step S4 further includes:

[0057] Several evaluation thresholds are set, and the data in the combination matrix is ​​compared with the evaluation thresholds to determine the categories of public policy topics that citizens tend to discuss within a preset period, and the percentage of each category is calculated.

[0058] This invention also proposes a public policy participation evaluation system based on LDA and vector space models, the system comprising:

[0059] The data acquisition module acquires the target public policy document and also acquires dynamic data within a preset period after the target public policy document is published, which serves as the target dynamic data.

[0060] The data processing module uses the TF-IDF algorithm to perform word segmentation on the target public policy documents and target dynamic data respectively, to obtain a set of word segmentation for document features and a set of word segmentation for data features;

[0061] The LDA model building module builds an LDA model and extracts the semantic information implied in the target public policy documents and target dynamic data to obtain a document-data matrix.

[0062] The vector space model building module constructs a vector space model based on the document feature word segmentation set and the data feature word segmentation set, calculates the similarity between the target public policy document and the target dynamic data, and obtains the similarity matrix;

[0063] The analysis module performs linear weighting on the document-data matrix and the similarity matrix to obtain a combined matrix that can map the analysis of citizens' attention to and inclination towards public policies within a preset period.

[0064] The present invention provides a public policy participation assessment method and system based on LDA and vector space model, which has the following advantages compared with the prior art:

[0065] (1) The evaluation of citizen participation in public policy based on LDA model and vector space model can discover the semantic information hidden in big data. Vector space model can reduce the dimensionality of data when there is a lot of data and high dimensionality, thus avoiding the problem of data sparsity.

[0066] (2) By analyzing the target public policy documents and various feedback data from citizens on the target public policy documents within the preset stage, we can obtain a comprehensive understanding of the attention paid by citizens to each topic in the target public policy documents and realize the intelligent evaluation results of public policy participation.

[0067] (3) The document-data matrix obtained by the LDA model and the similarity matrix obtained by the vector space model are linearly weighted, which comprehensively considers the hidden semantic information in the target public policy documents and the impact of citizens' attention to each topic, thus improving the accuracy and effectiveness of the evaluation results. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 This is a flowchart illustrating a public policy participation assessment method based on LDA and vector space model according to the present invention.

[0070] Figure 2 This is a schematic diagram illustrating the process of establishing an LDA model in a public policy participation assessment method based on LDA and vector space model according to the present invention. Detailed Implementation

[0071] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0072] like Figure 1 As shown in the figure, this embodiment provides a method for evaluating public policy participation based on LDA and vector space models. The method includes:

[0073] S1, obtain the target public policy document, and at the same time obtain the dynamic data within a preset period after the target public policy document is published, as the target dynamic data.

[0074] It is important to understand that this embodiment primarily uses web crawlers to collect various public policy information published by official bodies on the internet. The target public policy documents can be multimedia objects; the target dynamic data includes various feedback information and data from the media and netizens. For example, regarding the "Several Opinions on Financial Services for the Development of Agriculture, Rural Areas and Farmers" issued in 2014, various feedback information and data from the media and netizens were collected within five years after the policy was released in 2014.

[0075] S2, the TF-IDF algorithm is used to perform word segmentation on the target public policy document and the target dynamic data respectively to obtain the document feature word segmentation set and the data feature word segmentation set.

[0076] The TF-IDF algorithm is an improved feature weighting algorithm, and its calculation formula is as follows:

[0077]

[0078] Where W(t,d) represents the weight of t in d, tf(t,d) is the frequency of t in d, d is a document in the text corpus C, t is a word in document d, and C = {c1,c2,…,c...} m}, c i ∈C, m is the number of categories in the text corpus, G represents the total number of documents in the text corpus, n t W(w) represents the number of texts t contained in the text corpus. k ,c i ) indicates the word w k Regarding category c i Category weights, w k c represents the k-th word. i This represents the i-th type of article.

[0079] The target public policy document is segmented using the TF-IDF algorithm, resulting in an M×N matrix, which is the document feature segmentation set V. M×N M represents the number of texts in the target public policy document, N represents the number of words in each text, and V represents the document feature word segmentation set. M×N Each line in the table represents the feature word segmentation weights of a text within the target public policy;

[0080] The target dynamic data is segmented using the TF-IDF algorithm, resulting in a P×Q matrix, which is the data feature segmentation set V. P×Q P represents the number of texts in the target dynamic data, Q represents the number of words in each text, and V represents the data feature word segmentation set. P×Q Each line represents the feature word segmentation weight of a text in the target dynamic data.

[0081] Before using the TF-IDF algorithm to segment the target public policy documents and target dynamic data, preprocessing operations are required. Preprocessing includes removing useless symbols, stemming, and removing stop words.

[0082] TF-IDF is a commonly used weighting technique for information retrieval and data mining. TF stands for Term Frequency and IDF stands for Inverse Document Frequency. In this embodiment, considering the phenomenon that the same words appear frequently in articles of a certain category but rarely in articles of other categories, or appear frequently in all categories, we improve upon the traditional TF-IDF by multiplying the numerator and denominator by the category weights.

[0083] S3. Establish an LDA model and extract the semantic information implied in the target public policy documents and target dynamic data through the LDA model to obtain the document-data matrix.

[0084] A first LDA model is established, and the target public policy documents are used as the corpus for training the first LDA model.

[0085] A second LDA model is established, and the target dynamic data is used as the corpus of the second LDA model for training.

[0086] The first LDA model is used to extract the hidden semantic information of the target public policy document, obtain the first word distribution matrix, and then calculate the first topic distribution matrix. The product of the first word distribution matrix and the first topic distribution matrix is ​​used as the document distribution matrix.

[0087] The second LDA model is used to extract the implicit semantic information in the target dynamic data, obtain the second word distribution matrix, and then calculate the second topic distribution matrix. The product of the second word distribution matrix and the second topic distribution matrix is ​​used as the data distribution matrix.

[0088] Multiplying the document distribution matrix by the data distribution matrix yields the document-data matrix.

[0089] like Figure 2 As shown, the specific steps for establishing an LDA model include:

[0090] S301, randomly initialize the LDA model, and randomly assign a topic number z to each word w in each document in the corpus, with a total of K topics.

[0091] S302, rescan the corpus, for each word w, resample its topic according to the Gibbs sampling formula, and update the corpus.

[0092] It should be understood that the Gipps sampling formula is existing technology and will not be elaborated here.

[0093] S303, repeat the resampling process of S302 until the value of the Gibbs sampling formula converges.

[0094] It is important to understand that when the value of the Gipps sampling formula gradually stabilizes, the resampling process stops and proceeds to the next step.

[0095] S304, statistically analyze the distribution matrix of words under a topic in the corpus, and then calculate the topic distribution matrix of the corresponding document.

[0096] The probability distribution of words under a given topic is:

[0097]

[0098] Among them, W k For the k-th word in the document, T j If the word distribution matrix represents the j-th topic in the corpus, then it is as follows:

[0099]

[0100] The probability distribution of topics in a document is as follows:

[0101]

[0102] Among them, T j For the j-th topic in the corpus, D i For the i-th document in the corpus, the topic distribution matrix is ​​represented as:

[0103]

[0104] The product of the first word distribution matrix and the first topic distribution matrix is ​​taken as the document distribution matrix, and the expression of the document distribution matrix is:

[0105] Z′=X′·Y′

[0106] Where Z′ represents the document distribution matrix, X′ represents the first word distribution matrix, and Y′ represents the first topic distribution matrix;

[0107] The product of the second word distribution matrix and the second topic distribution matrix is ​​taken as the data distribution matrix, and the expression of the data distribution matrix is:

[0108] Z″=X″·Y″

[0109] Where Z″ represents the document distribution matrix, X″ represents the first word distribution matrix, and Y″ represents the first topic distribution matrix;

[0110] The document distribution matrix is ​​multiplied by the data distribution matrix to obtain the document-data matrix. The expression for the document-data weights is as follows:

[0111] R = Z′·Z″

[0112] Where R represents the document-data matrix, Z′ represents the document distribution matrix, and Z″ represents the document distribution matrix.

[0113] It is important to understand that multiplying the word distribution matrix and the topic distribution matrix yields the weight distribution of words in a sentence. Although the weight distribution already reflects the importance of words in a sentence, in order to assess citizens' participation in public policy, it is necessary to consider media and netizens' comments and hot topics. Therefore, multiplying the document distribution matrix and the data distribution matrix, the implicit semantic information of the target public policy document can be analyzed through the document-data matrix.

[0114] S4. Based on the document feature word segmentation set and the data feature word segmentation set, a vector space model is constructed to calculate the similarity between the target public policy document and the target dynamic data, and a similarity matrix is ​​obtained.

[0115] Calculate the similarity between the document feature word segmentation set and the data feature word segmentation set to obtain the similarity matrix, the expression of which is:

[0116]

[0117] Among them, s mp For document feature word segmentation set V M×N The m-th row and the data feature word segmentation set V P×Q The cosine similarity of the p-th row, m = 1, 2, ..., M, p = 1, 2, ..., P.

[0118] It is important to understand that s mp This represents the similarity between a document in the target public policy and a document in the target dynamic data. This method obtains the similarity between the target public policy document and the target dynamic data, thus better reflecting citizens' attention to and inclination towards topics in public policy. The cosine similarity algorithm is existing technology and will not be elaborated upon here.

[0119] S5 performs linear weighting on the document-data matrix and the similarity matrix to obtain a combined matrix that can map citizens' attention to public policies within a preset period.

[0120] The document-data matrix and similarity matrix are linearly weighted to obtain the combination matrix Q, which is calculated using the following formula:

[0121] Q = θ × R + (1 - θ) × E

[0122] Where θ is the linear combination coefficient, R is the document-data matrix, and E is the similarity matrix.

[0123] It is important to understand that the document-data matrix reflects the semantic information of the target public policy document, while the similarity matrix reflects citizens' attention to and inclination towards topics in public policy. Combining the document-data matrix and the similarity matrix can effectively assess citizen participation in public policy, improving the accuracy and effectiveness of the assessment.

[0124] Several evaluation thresholds are set, and the data in the combination matrix is ​​compared with the evaluation thresholds to determine the categories of public policy topics that citizens tend to discuss within a preset period, and the percentage of each category is calculated.

[0125] It is important to understand that in this embodiment, the "Several Opinions on Financial Services for the Development of Agriculture, Rural Areas and Farmers" issued in 2014 is used as the target public policy document. Various feedback information and data from the media and netizens within five years of the policy's release in 2014 are used as the target dynamic data. The LDA model yields the following topic distribution: improving the agricultural financial environment, perfecting the supporting service system, strictly controlling the outflow of agricultural credit, increasing financial innovation, and others. Therefore, four evaluation thresholds are set: when a data point in the combination matrix is ​​greater than or equal to the first evaluation threshold, its topic is "improving the agricultural financial environment"; when a data point in the combination matrix is ​​less than the first evaluation threshold but greater than or equal to the second evaluation threshold, its topic is "perfecting the supporting service system"; when a data point in the combination matrix is ​​less than the second evaluation threshold but greater than or equal to the third evaluation threshold, its topic is "strictly controlling the outflow of agricultural credit"; when a data point in the combination matrix is ​​less than the third evaluation threshold but greater than or equal to the fourth evaluation threshold, its topic is "increasing financial innovation"; and when a data point in the combination matrix is ​​less than the fourth evaluation threshold, its topic is "other". The analysis of the percentage of each topic in the total data revealed the following percentages of citizens' preferences for public policy topics within the pre-defined phase: improving the agricultural financial environment (33.6%), improving the supporting service system (28%), strictly controlling the outflow of agricultural credit (21.5%), increasing financial innovation (12.2%), and other (4.7%).

[0126] This embodiment also provides a public policy participation assessment system based on LDA and vector space models, the system comprising:

[0127] The data acquisition module acquires the target public policy document and also acquires dynamic data within a preset period after the target public policy document is published, which serves as the target dynamic data.

[0128] The data processing module uses the TF-IDF algorithm to perform word segmentation on the target public policy documents and target dynamic data respectively, to obtain a set of word segmentation for document features and a set of word segmentation for data features;

[0129] The LDA model building module builds an LDA model and extracts the semantic information implied in the target public policy documents and target dynamic data to obtain a document-data matrix.

[0130] The vector space model building module constructs a vector space model based on the document feature word segmentation set and the data feature word segmentation set, calculates the similarity between the target public policy document and the target dynamic data, and obtains the similarity matrix;

[0131] The analysis module performs linear weighting on the document-data matrix and the similarity matrix to obtain a combined matrix that can map the analysis of citizens' attention to and inclination towards public policies within a preset period.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A public policy engagement evaluation method based on LDA and vector space model, characterized in that, The method comprises: S1, obtaining a target public policy document, and obtaining dynamic data in a preset stage after the target public policy document is published as target dynamic data; S2, performing word segmentation processing on the target public policy document and the target dynamic data respectively by using a TF-IDF algorithm to obtain a document feature word segmentation set and a data feature word segmentation set; S3, establishing an LDA model, extracting semantic information hidden in the target public policy document and the target dynamic data respectively by using the LDA model, and obtaining a document-data matrix; step S3 specifically comprises: establishing a first LDA model, taking the target public policy document as a corpus of the first LDA model, and training the first LDA model; establishing a second LDA model, taking the target dynamic data as a corpus of the second LDA model, and training the second LDA model; the first LDA model is used to extract hidden semantic information of the target public policy document to obtain a first word distribution matrix, and then a first topic distribution matrix is calculated, and a product of the first word distribution matrix and the first topic distribution matrix is taken as a document distribution matrix; the second LDA model is used to extract semantic information hidden in the target dynamic data to obtain a second word distribution matrix, and then a second topic distribution matrix is calculated, and a product of the second word distribution matrix and the second topic distribution matrix is taken as a data distribution matrix; the document distribution matrix and the data distribution matrix are multiplied to obtain the document-data matrix; S4, constructing a vector space model based on the document feature word segmentation set and the data feature word segmentation set, calculating a similarity between the target public policy document and the target dynamic data, and obtaining a similarity matrix; S5, performing linear weighting processing on the document-data matrix and the similarity matrix to obtain a combination matrix that can map a public policy attention degree of citizens in the preset stage.

2. The public policy engagement evaluation method based on LDA and vector space model according to claim 1, wherein, Step S2 specifically comprises: The TF-IDF algorithm is an improved feature weight algorithm, and its calculation formula is: Wherein, W(t, d) represents the weight of t in d, tf(t, d) is the frequency of t in d, d is a document in the text corpus C, t is a word in the document d, C = {c1, c2, …, c m}, c i ∈ C, m is the number of categories of the text corpus, G represents the total number of documents in the text corpus, n t is the number of texts containing t in the text corpus, W(w k ,c i ) represents the weight of the word w k in the category c i , w k represents the kth word, and c i represents the ith article. The TF-IDF algorithm is used for word segmentation processing on the target public policy document, to obtain an M*N matrix, that is, a document feature word segmentation set V M×N , where M represents the number of Chinese texts of the target public policy document, and N represents the number of words in each text. Adopt TF-IDF algorithm to the target dynamic data and carry out word segmentation processing, obtain a P × Q matrix, namely data feature word segmentation set V P×Q , P indicates the number of target dynamic data Chinese text, and Q indicates the number of words in each text.

3. The public policy engagement evaluation method based on LDA and vector space model according to claim 2, wherein, The step of establishing the LDA model specifically comprises: S301, randomly initializing the LDA model, randomly assigning a topic number z to each word w in each document in the corpus, and the total number of topics is K; S302, rescan the corpus, and resample the topic of each word w according to the Gips sampling formula, and update in the corpus; S303, repeat the resampling process of S302 until the value of the Gips sampling formula converges; S304, statistics the distribution matrix of the words under the topic in the corpus, and then calculate the corresponding document topic distribution matrix.

4. The public policy engagement evaluation method based on LDA and vector space model according to claim 3, wherein, Step S304 specifically comprises: The distribution probability of a word under a topic is: where W k is the kth word in the document, T j is the jth topic in the corpus, and the word distribution matrix is represented as: The distribution probability of a topic in a document is: where T j is the jth topic in the corpus, D i is the ith document in the corpus, and the topic distribution matrix is given by:

5. The public policy engagement evaluation method based on LDA and vector space model according to claim 4, wherein, The document-data matrix is obtained by multiplying the document distribution matrix and the data distribution matrix, specifically comprising: The product of the first word distribution matrix and the first topic distribution matrix is taken as the document distribution matrix, and the expression of the document distribution matrix is: Z′=X′·Y′ Wherein, Z' represents the document distribution matrix, X' represents the first word distribution matrix, and Y' represents the first topic distribution matrix; The product of the second word distribution matrix and the second topic distribution matrix is taken as the data distribution matrix, and the expression of the data distribution matrix is: Z"=X"Y" wherein, Z" represents a document distribution matrix, X" represents a first word distribution matrix, and Y" represents a first topic distribution matrix; a document-data matrix is obtained by multiplying the document distribution matrix and the data distribution matrix, and an expression of the document-data weight is: R=Z'Z" wherein, R represents a document-data matrix, Z' represents a document distribution matrix, and Z" represents a document distribution matrix.

6. The public policy engagement evaluation method based on LDA and vector space model according to claim 5, wherein, Step S4 specifically comprises: a similarity matrix is obtained by calculating the similarity between the document feature word set and the data feature word set, and an expression of the similarity matrix is: wherein s mp is the cosine similarity between the mth row of the document feature term set V M×N and the pth row of the data feature term set V P×Q , m = 1, 2, …, M, p = 1, 2, …, P.

7. The public policy engagement evaluation method based on LDA and vector space model according to claim 6, wherein, Step S5 specifically comprises: a combination matrix Q is obtained by linearly weighting the document-data matrix and the similarity matrix, and a calculation formula of the combination matrix Q is: Q=0XR+(1-0)XE wherein, 0 is a linear combination coefficient, R is a document-data matrix, and E is a similarity matrix.

8. The public policy engagement evaluation method based on LDA and vector space model according to claim 1, wherein, Step S4 further comprises: a plurality of evaluation thresholds are set, the data in the combination matrix is compared with the plurality of evaluation thresholds, the topic tendency category of the citizens in the preset stage is determined, and the proportion of each category is counted.

9. A public policy engagement evaluation system based on LDA and vector space model to implement the evaluation method of any one of claims 1 to 8, characterized in that, The system comprises: a data acquisition module, which acquires a target public policy document and simultaneously acquires dynamic data in a preset stage after the target public policy document is published as target dynamic data; a data processing module, which performs word segmentation processing on the target public policy document and the target dynamic data by using a TF-IDF algorithm to obtain a document feature word set and a data feature word set; an LDA model establishing module, which establishes an LDA model, extracts semantic information implied in the target public policy document and the target dynamic data by using the LDA model, and obtains a document-data matrix; a vector space model establishing module, which constructs a vector space model based on the document feature word set and the data feature word set, calculates the similarity between the target public policy document and the target dynamic data, and obtains a similarity matrix; an analysis module, which linearly weights the document-data matrix and the similarity matrix to obtain a combination matrix that can map the attention of the citizens in the preset stage to the public policy.

Images