Tobacco wisdom common method big data platform based on deep learning

Abstract

The invention provides a tobacco wisdom general law big data platform based on deep learning, and relates to the technical field of data processing, and the platform comprises a data acquisition module, a feature extraction module, an intelligent question and answer module, a law article push module and a class case push module. The data acquisition module is used for acquiring tobacco legal consultation data of a user; the feature extraction module is used for carrying out word segmentation processing on the tobacco legal consultation data and extracting text features of all segmented words; the intelligent question-answering module is used for performing intelligent reply according to the text features; the law article pushing module is used for pushing related law articles based on the deep learning network; and the similar case pushing module is used for pushing similar cases based on the deep learning network. According to the method, real-time interaction and personalized feedback are provided, the efficiency and accuracy of legal consultation and general law are improved, the user is helped to better understand the law, and legal and compliant construction of the tobacco industry is promoted.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a tobacco intelligent legal education big data platform based on deep learning. Background Art

[0002] The tobacco industry is a highly regulated industry involving a large number of complex laws and regulations. Through the smart legal education platform, industry practitioners can more easily obtain tobacco-related legal knowledge, enhance legal awareness, reduce the occurrence of illegal activities, and effectively improve the legal compliance level of the entire industry.

[0003] The popularization of tobacco laws faces many difficulties: the dimensions of the popularization are not broad enough, the content of the popularization is not deep enough, the methods of the popularization are not interactive enough, and the benefits of the popularization are not long-term.

[0004] At present, the main means of popularizing tobacco laws are public accounts and offline publicity, which rely on people's active attention or participation, resulting in the information only covering a part of the population, and failing to reach those who do not actively participate. In addition, public account articles are limited by the number of words, and the time of lectures is also limited, making it difficult to introduce complex legal concepts in depth and in detail, resulting in insufficient information transmission, and people may not be able to fully understand the complex legal content. At the same time, public accounts and offline publicity are usually one-way forms of information transmission, lacking real-time interaction and personalized feedback. Summary of the invention

[0005] In order to solve the existing problems that the public accounts and offline publicity methods of popularizing the law need to rely on people's active attention or participation, resulting in that the information can only cover a part of the population, but cannot reach those who do not actively participate, the public account articles are limited by the number of words, and the lecture time is also limited, it is difficult to introduce complex legal concepts in depth and in detail, resulting in insufficient information transmission, and people may not be able to fully understand the complex legal content. At the same time, public accounts and lectures are usually one-way information transmission forms, lacking real-time interaction and personalized feedback. The present invention provides a tobacco smart legal education big data platform based on deep learning.

[0006] The technical solution provided by the embodiment of the present invention is as follows:

[0007] The embodiment of the present invention provides a tobacco intelligent law popularization big data platform based on deep learning, including: a data acquisition module, a feature extraction module, an intelligent question and answer module, a law article push module and a similar case push module;

[0008] The data acquisition module is used to acquire tobacco legal consultation data of the user;

[0009] The feature extraction module is used to perform word segmentation processing on the tobacco legal consultation data and extract text features of each word segmentation;

[0010] The intelligent question-answering module is used to make intelligent replies based on the text features;

[0011] The legal article push module is used to push relevant legal articles based on a deep learning network;

[0012] The similar case push module is used to push similar cases based on a deep learning network.

[0013] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0014] In the present invention, users can learn tobacco legal knowledge anytime and anywhere through the tobacco intelligent legal education big data platform based on deep learning. Intelligent question and answer, relevant legal articles push and similar case push can be carried out according to the user's tobacco legal consultation data, which provides real-time interaction and personalized feedback, improves the efficiency and accuracy of legal consultation and legal education, and helps users better understand the law, thus promoting the rule of law and compliance construction in the tobacco industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in 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 creative work.

[0016] Figure 1 A schematic diagram of the structure of a tobacco intelligent legal education big data platform based on deep learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0018] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0019] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0020] Reference Manual Attached Figure 1 , showing a structural schematic diagram of a tobacco intelligent legal education big data platform based on deep learning provided in an embodiment of the present invention.

[0021] The embodiment of the present invention provides a tobacco intelligent legal education big data platform 20 based on deep learning, including: a data acquisition module 201, a feature extraction module 202, an intelligent question and answer module 203, a law push module 204 and a similar case push module 205.

[0022] The data acquisition module 201 is used to acquire the tobacco legal consultation data of the user.

[0023] The feature extraction module 202 is used to perform word segmentation processing on the tobacco legal consultation data and extract text features of each word segmentation.

[0024] The intelligent question and answer module 203 is used to make intelligent replies based on text features.

[0025] The legal article push module 204 is used to push relevant legal articles based on a deep learning network.

[0026] The similar case push module 205 is used to push similar cases based on the deep learning network.

[0027] In a possible implementation, the feature extraction module 202 is specifically used for:

[0028] Extract the part-of-speech features of each word:

[0029]

[0030] Among them, Pa i Represents the lexical features of the i-th participle, n represents noun, v represents verb, adj represents adjective, and otherwise represents other parts of speech.

[0031] It should be noted that words of different parts of speech have different importance in legal consultation texts. Descriptive nouns have the highest importance in legal consultation texts. The reason is that nouns usually represent key entities, such as legal clauses, subjects, objects, etc., and are therefore given a higher weight. Verbs mainly represent behaviors and actions in legal sentences, but compared to nouns, they have a weaker descriptive effect on legal entities, so their weight is slightly lower. Adjectives are usually used to describe the attributes of nouns or other entities, and their importance in legal texts is relatively low.

[0032] Extract the word length features of each word segmentation:

[0033]

[0034] Among them, L i represents the word length feature of the i-th word, l i Indicates the length of the i-th word, l max Indicates the maximum word length.

[0035] It should be noted that the word length feature measures the relative size of the word segmentation length in the entire text. By calculating the ratio between the word segmentation length and the maximum length, we can judge the "relative importance" of the word in the entire text. Generally, longer words may carry more information, so longer word segmentations may be assigned higher weights in the analysis.

[0036] Extract the position features of each word segmentation:

[0037]

[0038] Among them, Po i represents the position feature of the i-th word segmentation. "first sentence" means in the first sentence, "lastsentence" means in the last sentence, and "other sentence" means in other positions.

[0039] It should be noted that the position feature is used to evaluate the impact of the position of the word segmentation in the sentence on its importance. Generally, in legal consultations or documents, words appearing in the first sentence often represent the theme or key content, so they have higher weights. The last sentence may be the summary or conclusion part and also has a certain degree of importance. The importance of the position in the middle sentences is relatively low, so the weight is the lowest.

[0040] Extract the word frequency features of each word segmentation:

[0041]

[0042] Among them, TF i represents the word frequency of the i-th word segmentation, a i represents the number of times the i-th word segmentation appears in the tobacco legal consultation data, and N represents the total number of words in the tobacco legal consultation data.

[0043] It should be noted that the word frequency feature measures the frequency of a certain word segmentation in the current text. The word frequency reflects the importance of the word in the text. The more frequently a word appears, the more important it usually is in the text. However, overly frequent words (such as "de", "shi") may belong to stop words and need to be further combined with other features (such as inverse consultation frequency) to judge their actual importance.

[0044] Extract the inverse consultation frequency of each word segmentation:

[0045]

[0046] Among them, IDF i represents the inverse consultation frequency of the i-th word segmentation, b represents the total number of tobacco legal consultation historical data, b i represents the total number of tobacco legal consultation historical data containing the i-th word segmentation.

[0047] It should be noted that the reverse consultation frequency measures the rarity of a word segment appearing in the entire historical dataset. The higher the IDF, the less frequently the word segment appears in the historical consultation data, and the more discriminative and informative it is. Common words (such as "de", "he") appear in many texts, with a lower IDF and relatively less importance; while less frequently appearing words have higher discrimination, greater information content, and a higher IDF.

[0048] Based on the part-of-speech characteristics, word length characteristics, position characteristics, word frequency characteristics, and reverse consultation frequency of each word segment, determine the integrated text characteristics of each word segment:

[0049] R i =(Pa i +L i +Po i +TF i )×IDF i

[0050] Wherein, R i represents the integrated text characteristic of the i-th word segment.

[0051] In the present invention, by integrating features such as part-of-speech, word length, position, word frequency, and reverse consultation frequency, weights can be assigned to word segments in multiple dimensions, enabling the system to better understand the text semantics and improve the accuracy of information extraction and intelligent question answering. This comprehensive feature extraction method not only enhances the adaptability and precision of the model, but also effectively solves problems such as high-frequency word interference and insufficient information content, ultimately providing more accurate services for legal consultation and legal push.

[0052] In a possible implementation manner, the intelligent question answering module 203 adopts the ChatGPT model, Tongyi model, or Wenxin Yiyan model.

[0053] In the present invention, these models (such as ChatGPT, Tongyi model, Wenxin Yiyan) are deep learning language models trained with a large amount of data and have powerful natural language understanding capabilities. By adopting these models, the platform can effectively process users' tobacco legal consultations, accurately understand the legal questions raised by users, generate more semantically compliant answers, and improve the accuracy and reliability of intelligent question answering.

[0054] In one possible implementation, the deep learning network is specifically a multi-layer stacked gated loop unit, which includes a first branch and a second branch. The first branch is connected to the law pushing module 204, and the deep correlation between the tobacco legal consulting data and each law is calculated through the first branch. The second branch is connected to the similar case pushing module 205, and the deep similarity between the tobacco legal consulting data and each historical case is calculated through the second branch.

[0055] Among them, the gated recurrent unit (GRU) is a variant of a recurrent neural network (RNN) for processing sequence data. It is mainly used to solve the gradient vanishing problem of traditional RNN in long sequence processing, while simplifying the network structure. The core idea of ​​GRU is to control the transmission and forgetting of information through a gating mechanism, so as to effectively capture the contextual relationship in a long sequence. The gating mechanism in the GRU network includes two main gates: the update gate and the reset gate. Through the control of these gates, GRU can dynamically decide which information should be retained and which should be discarded, thereby avoiding the gradient vanishing problem and maintaining sensitivity to long sequences.

[0056] It should be noted that the multi-layer stacked GRU can gradually extract deeper features at multiple levels, thereby effectively capturing the potential relationship between user consultation data and legal provisions and historical cases. The first branch focuses on analyzing the relevance of consultation data to legal provisions, and the second branch focuses on the similarity with historical cases. This architecture with clear division of labor can extract more precise features in each field and improve the accuracy and relevance of push results.

[0057] In a possible implementation, the multi-layer stacked gated recurrent unit is specifically used for:

[0058] Input word embedding vectors for the tobacco legal advisory data.

[0059] Optionally, word embedding vectors are extracted via Word2Vec, GloVe, and / or FastText.

[0060] Input the word embedding vector into the first layer of gated recurrent unit to obtain the hidden state of the first layer of gated recurrent unit:

[0061]

[0062] Among them, z t represents the output vector of the update gate at time t, σ() represents the activation function, and W uz represents the weight matrix between the input layer and the update gate, x t represents the input data at time t (weighted word embedding vector sequence), W hzrepresents the self-connection weight matrix between the update gate at time t and time t-1, represents the hidden state of the first layer of gated recurrent unit at time t-1, b z represents the bias term of the update gate, r t represents the output vector of the reset gate at time t, W ur represents the weight matrix between the input layer and the reset gate, W hr represents the self-connection weight matrix between the reset gate at time t and time t-1, b r represents the bias term of the reset gate, c t represents the output vector of the candidate layer at time t, W uc Represents the weight matrix between the input layer and the candidate layer, W hc represents the self-connection weight matrix between the candidate layer at time t and time t-1, b c represents the bias term of the candidate layer, ⊙ represents the element-wise product operation, Represents the hidden state of the first layer of gated recurrent unit at time t.

[0063] Input the hidden state of the i-1th layer gated recurrent unit into the i-th layer gated recurrent unit to obtain the hidden state of the i-th layer gated recurrent unit:

[0064]

[0065] in, represents the hidden state of the i-1th layer gated recurrent unit at time t-1, represents the hidden state of the i-th layer gated recurrent unit at time t-1, represents the hidden state of the i-th layer gated recurrent unit at time t, i ≥ 2.

[0066] It should be noted that the multi-layer stacked GRU structure allows each layer to abstract and represent the data at different levels. The bottom GRU captures lower-level time series features, such as short-term dependencies between words; and as the number of layers increases, the upper GRU can capture higher-level features, such as the overall semantic structure of the entire consulting question. This layer-by-layer abstraction process enables the model to have a deeper understanding of the text and helps to identify complex legal consulting intentions.

[0067] The hidden states in the candidate layers of the gated recurrent units in each layer are fused through the fully connected layer to extract the comprehensive hidden state of the tobacco legal consultation data:

[0068]

[0069] Among them, H represents the comprehensive latent state that represents the current consultation intention, β i Represents the fusion coefficient of the i-th layer gated recurrent unit.

[0070] It should be noted that by weighted fusion of hidden states of different layers through the fully connected layer, the semantic features of different levels captured by each layer can be effectively extracted, and the information at different levels can be combined to form a more global semantic expression, thereby more accurately representing the intention of the current legal consultation.

[0071] The first branch calculates the deep correlation between tobacco legal consultation data and various laws:

[0072]

[0073] Among them, η u represents the deep correlation between tobacco legal consultation data and the u-th legal article, H u represents the comprehensive hidden state of the u-th law, and || || represents the modular operation.

[0074] It should be noted that by calculating the cosine similarity between the comprehensive hidden state of tobacco legal consultation data and the comprehensive hidden state of each legal article, the similarity of the two vectors in high-dimensional space can be quantified. Since the comprehensive hidden state is a deep semantic representation extracted by multi-layer stacked gated recurrent units (GRU), they can capture the complex semantic information in the legal consultation text and the legal article text. By calculating the cosine similarity of these two deep semantic representations, the system can identify the deep semantic association between user consultation and legal articles, rather than just a superficial literal match.

[0075] The second branch calculates the deep similarity between tobacco legal consultation data and various historical cases:

[0076]

[0077] Among them, γ j represents the deep similarity between tobacco legal consultation data and the jth historical case, H j Represents the comprehensive hidden state of the jth historical case.

[0078] It should be noted that by calculating the cosine similarity between the comprehensive hidden state of tobacco legal consultation data and the comprehensive hidden state of each historical case, the similarity of the two vectors in high-dimensional space can be quantified. Since the comprehensive hidden state is a deep semantic representation extracted through multi-layer stacked gated recurrent units (GRU), they are able to capture the complex semantic information in the legal consultation text and the historical case text. By calculating the cosine similarity of these two deep semantic representations, the system can identify the deep semantic association between user consultation and historical cases, rather than just a literal match on the surface.

[0079] In a possible implementation, the legal article push module 204 specifically includes:

[0080] According to the text features of each word, calculate the weight of each word:

[0081]

[0082] Among them, w i Represents the weight of the i-th word, R i Represents the fused text features of the i-th word segment.

[0083] It should be noted that by calculating the fused text features of each segmented word and weighting them, the system can give higher weights to important segmented words (such as key legal terms or nouns). This makes key words more influential in similarity calculations, thereby ensuring that the legal provisions can be pushed closer to the user's legal issues and needs.

[0084] Get the number of times each word appears in each legal item.

[0085] According to the number of times each word appears in the legal article and the weight of each word, the word correlation between tobacco legal consultation data and each legal article is calculated:

[0086]

[0087] Among them, ρ u represents the correlation between tobacco legal consultation data and the u-th legal article, c ui It indicates the number of times the i-th word appears in the u-th legal item, and n indicates the total number of word segments.

[0088] It should be noted that by calculating the number of times each legal article hits a segmentation word and combining the segmentation weight, the system can effectively distinguish between relevant and irrelevant legal articles. Those legal articles that contain more important segmentations and have a higher hit frequency will have a higher relevance; while those legal articles that contain fewer important segmentations or only hit non-keywords will have a lower relevance. This approach can greatly reduce the push of irrelevant legal articles and improve user experience.

[0089] The deep correlation between tobacco legal consultation data and various legal provisions is obtained from the first branch of the multi-layer stacked gated recurrent unit.

[0090] According to the segmentation relevance and deep relevance between tobacco legal consultation data and various laws, the comprehensive relevance is determined:

[0091] ε u =η u ρ u

[0092] Among them, ε u Represents the comprehensive correlation between tobacco legal consultation data and the jth legal article.

[0093] It should be noted that the word segmentation relevance reflects the lexical matching between the consulting text and the legal provisions, while the deep relevance reflects the matching at the context and semantic level. By combining these two relevances, the fusion of shallow and deep information is achieved, avoiding the shortcomings of relying solely on lexical matching while ignoring semantics, and also avoiding the neglect of certain key words in the deep semantic matching process.

[0094] Legal provisions are pushed based on the comprehensive correlation between tobacco legal consultation data and various legal provisions.

[0095] Optionally, according to the comprehensive relevance between tobacco legal consultation data and various laws, laws are pushed. Specifically, various laws are sorted in order of comprehensive relevance from high to low, and the top 10 laws with high relevance are selected for push.

[0096] In the present invention, this legal article push mechanism builds a system that comprehensively measures the relevance between the consulting text and the legal article through the combination of word segmentation relevance and deep relevance, ensuring the accuracy and efficiency of the push results. The introduction of word segmentation weights enables better recognition of key words, and deep semantic analysis enhances the system's ability to understand the context, ultimately improving the user experience and enhancing the accuracy and flexibility of the system.

[0097] In a possible implementation, the similar case push module 205 is specifically used for:

[0098] According to the text features of each word, calculate the weight of each word:

[0099]

[0100] Among them, w i Represents the weight of the i-th word, R i Represents the fused text features of the i-th word segment.

[0101] According to the weight of each segmentation, the structural similarity between the tobacco legal consultation data and each historical case is calculated:

[0102]

[0103] Among them, δ j represents the structural similarity between the tobacco legal consultation data and the jth historical case, C 0 Represents the word set of tobacco legal consultation data, C j Represents the word segmentation set of the j-th historical case data.

[0104] It should be noted that calculating the structural similarity through the intersection and union of the segmentation sets can intuitively measure the similarity between the consulting text and the historical case at the lexical level. The intersection represents the segmentations shared by the two texts, while the union represents all the segmentations in the two texts.

[0105] The attribute characteristics of tobacco legal consultation data are extracted, including: case type, crime type, amount involved, time, location, number of participants, behavior and behavior form.

[0106] According to the attribute characteristics of tobacco legal consultation data, the attribute similarity between tobacco legal consultation data and each historical case is calculated:

[0107]

[0108] Among them, σ j represents the attribute similarity between tobacco legal consultation data and the jth historical case, s 0k represents the kth attribute feature value of tobacco legal consultation data, s jk represents the kth attribute feature value of the jth historical case data, λ k It represents the weight coefficient of the kth attribute feature, and K represents the total number of attribute features.

[0109] It should be noted that this method measures the similarity between legal consultation and historical cases through multiple attribute features (such as case type, crime type, amount involved, time, location, etc.). Each attribute feature reflects a different dimension of the case, so a more comprehensive and accurate similarity assessment can be obtained by combining the similarities of multiple dimensions. This refined comparison method ensures that the pushed cases are not only similar in language structure, but also more relevant in specific case features.

[0110] The deep similarity between tobacco legal consultation data and various historical cases is obtained from the second branch of the multi-layer stacked gated recurrent unit.

[0111] Based on the structural similarity and attribute similarity between the tobacco legal consultation data and each historical case, the comprehensive similarity is determined:

[0112] τ j =δ j σ j γ j

[0113] Among them, τ j Represents the comprehensive similarity between tobacco legal consultation data and the jth historical case.

[0114] It should be noted that the comprehensive similarity formula combines structural similarity (matching at the vocabulary level), attribute similarity (matching of case features) and deep semantic similarity (matching at the semantic level), and can comprehensively evaluate the similarity between tobacco legal consultation data and historical cases from multiple dimensions. A single dimension may ignore some important information, but by combining information at different levels, the system can more accurately push cases that are highly relevant to user consultations and reduce errors.

[0115] Historical cases are pushed based on the comprehensive similarity between tobacco legal consulting data and various historical cases.

[0116] Optionally, historical cases are pushed based on the comprehensive similarity between tobacco legal consulting data and each historical case. Specifically, each historical case is sorted in order from high to low according to the comprehensive similarity, and the top 10 historical cases with high comprehensive similarity are selected for push.

[0117] In the present invention, the system evaluates the similarity between tobacco legal consultation data and historical cases through three dimensions: structural similarity, attribute similarity, and deep similarity, ensuring the comprehensiveness and accuracy of similarity calculation. Structural similarity measures lexical similarity at the word segmentation level, attribute similarity compares details through specific case features, and deep similarity captures the semantic connection between consultation texts and historical cases through multi-layer stacked gated recurrent units. This multi-dimensional analysis method ensures that the system can accurately evaluate case similarity from different angles, thereby improving the accuracy of push notifications.

[0118] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0119] In the present invention, users can learn tobacco legal knowledge anytime and anywhere through the tobacco intelligent legal education big data platform based on deep learning. Intelligent question and answer, relevant legal articles push and similar case push can be carried out according to the user's tobacco legal consultation data, which provides real-time interaction and personalized feedback, improves the efficiency and accuracy of legal consultation and legal education, and helps users better understand the law, thus promoting the rule of law and compliance construction in the tobacco industry.

[0120] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0121] There are a few points to note:

[0122] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0123] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0124] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0125] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A tobacco intelligent legal education big data platform based on deep learning, characterized by: include: Data acquisition module, feature extraction module, intelligent question and answer module, legal article push module and similar case push module; The data acquisition module is used to acquire tobacco legal consultation data of the user; The feature extraction module is used to perform word segmentation processing on the tobacco legal consultation data and extract text features of each word segmentation; The intelligent question-answering module is used to make intelligent replies based on the text features; The legal article push module is used to push relevant legal articles based on a deep learning network; The similar case push module is used to push similar cases based on a deep learning network.

2. The tobacco intelligent legal education big data platform based on deep learning according to claim 1 is characterized in that: The intelligent question-answering module adopts the ChatGPT model, the Tongyi model or the Wenxinyiyan model.

3. The tobacco intelligent legal education big data platform based on deep learning according to claim 2 is characterized in that: The deep learning network is specifically a multi-layer stacked gated loop unit, which includes a first branch and a second branch, and the first branch is connected to the legal article push module.

4. The tobacco intelligent legal education big data platform based on deep learning according to claim 3 is characterized in that: The deep correlation between the tobacco legal consulting data and each legal provision is calculated through the first branch, and the second branch is connected to the similar case push module, and the deep similarity between the tobacco legal consulting data and each historical case is calculated through the second branch.

5. The tobacco intelligent legal education big data platform based on deep learning according to claim 4 is characterized in that: The multi-layer stacked gated cycle unit is specifically used for: Input the word embedding vector of the tobacco legal consultation data; Inputting the word embedding vector into the first-layer gated recurrent unit to obtain the hidden state of the first-layer gated recurrent unit; Inputting the hidden state of the i-1th gated recurrent unit into the i-th gated recurrent unit to obtain the hidden state of the i-th gated recurrent unit; The hidden states in the candidate layers of the gated recurrent units of each layer are fused through a fully connected layer to extract the comprehensive hidden state of the tobacco legal consulting data; Calculate the deep correlation between the tobacco legal consultation data and various laws through the first branch; The deep similarity between the tobacco legal consulting data and each historical case is calculated through the second branch.

6. The tobacco intelligent legal education big data platform based on deep learning according to claim 5 is characterized in that: The word embedding vector is extracted through Word2Vec, GloVe and / or FastText.

7. The tobacco intelligent legal education big data platform based on deep learning according to claim 5 is characterized in that: The legal article push module specifically includes: Calculate the weight of each participle according to its text features; Get the number of times each word appears in each legal clause; Calculate the relevance of the tobacco legal consultation data to each legal article based on the number of times each participle appears in the legal article and the weight of each participle; Obtaining the deep correlation between the tobacco legal consultation data and each legal provision from the first branch of the multi-layer stacked gated loop unit; Determine the comprehensive relevance according to the segmentation relevance and deep relevance between the tobacco legal consultation data and various legal provisions; According to the comprehensive relevance between the tobacco legal consultation data and various laws, laws are pushed.

8. The tobacco intelligent legal education big data platform based on deep learning according to claim 7 is characterized in that: The legal provisions are pushed according to the comprehensive relevance between the tobacco legal consultation data and various legal provisions, specifically: Sort each legal provision in descending order according to the comprehensive relevance, and select the top 10 legal provisions with the highest relevance for push.

9. The tobacco intelligent legal education big data platform based on deep learning according to claim 5 is characterized in that: The similar case push module is specifically used for: Calculate the weight of each participle according to its text features; Calculating the structural similarity between the tobacco legal consultation data and each historical case according to the weight of each segmentation word; Extracting attribute features of the tobacco legal consultation data, wherein the attribute features include: case type, crime type, amount involved, time, location, number of participants, behavior, and behavior form; Calculating the attribute similarity between the tobacco legal consulting data and each historical case according to the attribute characteristics of the tobacco legal consulting data; Obtaining the deep similarity between the tobacco legal consulting data and each historical case from the second branch of the multi-layer stacked gated recurrent unit; The comprehensive similarity is determined based on the structural similarity and attribute similarity between the tobacco legal consulting data and various historical cases.

10. The tobacco intelligent legal education big data platform based on deep learning according to claim 9 is characterized in that: According to the comprehensive similarity between the tobacco legal consultation data and various historical cases, historical cases are pushed, specifically: Sort each historical case in descending order according to the comprehensive similarity, and select the top 10 historical cases with the highest comprehensive similarity to push.

Images