An intelligence clue analysis method and system based on artificial intelligence

Abstract

The present invention belongs to the technical field of intelligence clue analysis, and discloses an intelligence clue analysis method and system based on artificial intelligence. The method comprises the following steps: constructing an artificial intelligence model; collecting real-time target intelligence files, and performing intelligence processing to obtain corresponding real-time intelligence keyword groups; performing clue retrieval on the real-time intelligence keyword groups to obtain corresponding real-time related clue files; performing clue processing on several real-time related clue files; performing clue analysis on the real-time intelligence keyword groups and the real-time clue relationship network to obtain corresponding real-time clue analysis results; generating corresponding real-time intelligence clue reports based on the real-time clue analysis results and the real-time clue relationship network; and performing retrieval adjustment on the preset real-time clue retrieval scheme based on the real-time clue analysis results. The present invention solves the problems of low intelligence, high cost investment, low efficiency, poor data processing capability and poor accuracy in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligence clue analysis, and specifically relates to an intelligence clue analysis method and system based on artificial intelligence. Background Art

[0002] Intelligence clue analysis is a systematic information processing method that aims to extract valuable clues from a large amount of data and information, and then reveal the underlying patterns, relationships or potential threats. Intelligence clue analysis has a wide range of applications in national security, counter-terrorism, criminal investigation, commercial competition and other fields. Through effective intelligence clue analysis, risks can be warned in advance and strong support can be provided for decision-making.

[0003] The existing intelligence clue analysis methods have the following defects:

[0004] 1) Low intelligence level, often relying on manpower, requiring a lot of manpower and material resources for manual intelligence retrieval and clue analysis, resulting in low efficiency;

[0005] 2) Poor data processing capabilities make it impossible to accurately identify and analyze intelligence-related clues from massive amounts of data and documents, resulting in poor accuracy. Summary of the invention

[0006] In order to solve the problems of low intelligence, high cost, low efficiency, poor data processing capability and poor accuracy in the prior art, the present invention aims to provide an intelligence clue analysis method and system based on artificial intelligence.

[0007] The technical solution adopted by the present invention is:

[0008] An intelligence clue analysis method based on artificial intelligence includes the following steps:

[0009] Use artificial intelligence algorithms to build intelligence processing models, clue processing models, clue analysis models, and retrieval adjustment models;

[0010] Collect real-time target intelligence files, and perform intelligence processing on the target intelligence files according to the intelligence processing model to obtain the corresponding real-time intelligence keyword groups;

[0011] According to the preset real-time clue retrieval scheme, the real-time intelligence keyword group is searched for clues to obtain a number of corresponding real-time related clue files;

[0012] Using the clue processing model, a number of real-time related clue files are processed to obtain the corresponding real-time clue relationship network;

[0013] Use the clue analysis model to analyze the real-time intelligence keyword groups and the real-time clue relationship network to obtain the corresponding real-time clue analysis results;

[0014] If the real-time clue analysis result does not meet the requirements, then proceed to the next step, otherwise, generate a corresponding real-time intelligence clue report based on the real-time clue analysis result and the real-time clue relationship network, and end the method;

[0015] According to the real-time clue analysis results, the preset real-time clue retrieval plan is adjusted using the retrieval adjustment model to obtain the corresponding adjusted real-time clue retrieval plan, and the clue retrieval is performed again.

[0016] Furthermore, using artificial intelligence algorithms, constructing intelligence processing models, clue processing models, clue analysis models, and retrieval adjustment models, including the following steps:

[0017] Collecting a number of historical target intelligence files, and preprocessing them to obtain a number of corresponding preprocessed historical target intelligence files;

[0018] Based on several pre-processed historical target intelligence files, a natural language processing and deep learning fusion algorithm is used to build an intelligence processing model and generate several corresponding historical intelligence keyword groups;

[0019] Collect historical related clue files of several data sources, and set corresponding data format conversion strategies for historical related clue files of each data source;

[0020] According to the data format conversion strategy, data format conversion is performed on several heterogeneous historical clue files to obtain several corresponding homogeneous standardized historical clue files;

[0021] Preprocessing a number of standardized history-related clue files to obtain a number of corresponding preprocessed standardized history-related clue files;

[0022] Based on several pre-processed and standardized historical clue files, a clue processing model is constructed using a fusion algorithm of natural language processing and deep learning, and several corresponding historical clue relationship networks are generated;

[0023] Based on several historical clue relationship networks and several historical intelligence keyword groups, a clue analysis model is constructed using a deep learning algorithm, and several corresponding historical clue analysis results are generated;

[0024] According to the analysis results of several historical clues, a retrieval adjustment model is constructed using a reinforcement learning algorithm, and several corresponding historical retrieval adjustment strategy generation experiences are generated.

[0025] Furthermore, the intelligence processing model is based on the BERT-LSTM-CRF algorithm;

[0026] The clue processing model is built based on the TF-IDF-CI-GCN algorithm;

[0027] The clue analysis model is built based on the LSTM-GCN-Attention-RF algorithm;

[0028] The retrieval adjustment model is built based on the DQN algorithm.

[0029] Furthermore, real-time target intelligence files are collected, and according to the intelligence processing model, intelligence processing is performed on the target intelligence files to obtain corresponding real-time intelligence keyword groups, including the following steps:

[0030] Collect real-time target intelligence files, perform preprocessing, obtain corresponding real-time target intelligence text sequences, and input the real-time target intelligence text sequences into the intelligence processing model;

[0031] Perform vector representation on the real-time target intelligence text sequence to obtain the corresponding real-time target intelligence text word vector;

[0032] Extract semantic features from the word vectors of the real-time target intelligence text to obtain the corresponding semantic features of the real-time target intelligence text;

[0033] According to the semantic features of the real-time target intelligence text, the real-time target intelligence text sequence is annotated with intelligence keywords to obtain the initial real-time intelligence keyword group;

[0034] Using the preset corpus, the initial real-time intelligence keyword group is modified and expanded to obtain the final real-time intelligence keyword group.

[0035] Further, the initial real-time intelligence keyword group is modified and expanded using a preset corpus to obtain a final real-time intelligence keyword group, including the following steps:

[0036] The initial real-time intelligence keyword group is sequenced to obtain the initial real-time intelligence keyword sequence.

[0037] Obtaining semantic similarity between real-time intelligence keywords in the initial real-time intelligence keyword sequence and corpus keywords in a preset corpus, and using corpus keywords whose semantic similarity exceeds a threshold as correction keywords;

[0038] According to the entity relationship between the revised keyword and other corpus keywords in the preset corpus, a number of extended keywords are obtained;

[0039] A number of modified keywords and a number of extended keywords are added to the initial real-time intelligence keyword sequence to obtain a final real-time intelligence keyword sequence, and converted into a final real-time intelligence keyword group.

[0040] Further, according to the preset real-time clue retrieval scheme, the real-time intelligence keyword group is searched for clues to obtain a number of corresponding real-time related clue files, including the following steps:

[0041] According to the preset real-time clue retrieval plan, set the domain parameters, range parameters, depth parameters and frequency parameters of clue retrieval;

[0042] According to the domain parameter, scope parameter, depth parameter and frequency parameter, a clue search is performed on several data sources on the Internet to obtain several heterogeneous real-time candidate files;

[0043] Obtaining semantic similarity between a real-time intelligence keyword group and a number of real-time candidate files, and using the real-time candidate files whose semantic similarity exceeds a threshold as real-time related clue files;

[0044] According to a preset data format conversion strategy, data format conversion is performed on heterogeneous real-time related clue files to obtain corresponding standardized real-time related clue files.

[0045] Furthermore, using the clue processing model, clue processing is performed on a number of real-time related clue files to obtain a corresponding real-time clue relationship network, including the following steps:

[0046] Preprocessing a number of real-time related clue files to obtain a number of corresponding real-time related clue text sequences, and inputting the number of real-time related clue text sequences into a clue processing model;

[0047] Extracting real-time clue keyword groups of a plurality of real-time related clue text sequences;

[0048] Constructing a graph structure for the real-time clue keyword group to obtain a real-time clue relationship network graph structure;

[0049] According to the real-time clue relationship network graph structure, a relationship network is constructed for several real-time clue keywords to obtain the corresponding real-time clue relationship network.

[0050] Further, using the clue analysis model, the real-time intelligence keyword group and the real-time clue relationship network are subjected to clue analysis to obtain the corresponding real-time clue analysis results, including the following steps:

[0051] Inputting the final real-time intelligence keyword sequence and the real-time clue relationship network corresponding to the final real-time intelligence keyword group into the clue analysis model;

[0052] Extract the real-time data features of the final real-time intelligence keyword sequence;

[0053] Extract real-time graph features of real-time clue relationship network;

[0054] According to the preset attention weight value, the real-time data features and the real-time graph features are spliced ​​to obtain the real-time spliced ​​features;

[0055] Conduct clue analysis based on the real-time splicing features to obtain corresponding real-time clue analysis results.

[0056] Further, according to the real-time clue analysis result, the preset real-time clue retrieval scheme is adjusted by using the retrieval adjustment model to obtain the corresponding adjusted real-time clue retrieval scheme, and the clue retrieval is performed again, including the following steps:

[0057] Extracting several historical retrieval adjustment strategy generation experiences, and updating the retrieval adjustment model according to the real-time clue analysis results and several historical retrieval adjustment strategy generation experiences, to obtain an updated retrieval adjustment model;

[0058] Using the updated retrieval adjustment model, generating a real-time retrieval adjustment strategy and corresponding real-time retrieval adjustment strategy generation experience, and storing the real-time retrieval adjustment strategy generation experience;

[0059] According to the real-time retrieval adjustment strategy, the preset real-time clue retrieval plan is adjusted to obtain the corresponding adjusted real-time clue retrieval plan, and the clue retrieval is performed again.

[0060] An intelligence clue analysis system based on artificial intelligence is used to implement an intelligence clue analysis method. The system includes a model building unit, an intelligence processing unit, a clue retrieval unit, a clue processing unit, a clue analysis unit, a report generation unit and a retrieval adjustment unit which are connected in sequence.

[0061] The beneficial effects of the present invention are:

[0062] The present invention discloses an intelligence clue analysis method and system based on artificial intelligence. The method and system combine artificial intelligence algorithms to realize automatic and intelligent intelligence clue analysis, improve the intelligence level and efficiency, and reduce cost investment. The method is suitable for application scenarios of massive data retrieval and analysis. The method extracts keywords from intelligence files through an intelligence processing model, simplifies the complexity of intelligence analysis, and improves the efficiency of intelligence analysis. The method performs automatic retrieval through intelligence keywords, improves data processing capability and clue retrieval efficiency, and performs batch processing on the retrieved clue files through the clue processing model, further improving data processing capability. The generated clue relationship network can accurately reflect the correlation between clues, and improves the accuracy of intelligence clue analysis. The method performs automatic analysis through the clue analysis model, avoids the deviation and error existing in manual methods, excavates the deep relationship between intelligence and clues, and improves the efficiency and accuracy of intelligence clue analysis. The method realizes a feedback mechanism through a retrieval adjustment model, and can change the retrieval strategy according to the clue analysis results, thereby improving practicality.

[0063] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a flowchart of the intelligence clue analysis method based on artificial intelligence in the present invention.

[0065] Figure 2 It is a structural block diagram of the intelligence clue analysis system based on artificial intelligence in the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0067] Embodiment 1:

[0068] like Figure 1 As shown, this embodiment provides an intelligence clue analysis method based on artificial intelligence, comprising the following steps:

[0069] S1: Use artificial intelligence algorithms to build intelligence processing models, clue processing models, clue analysis models, and retrieval adjustment models, including the following steps:

[0070] S1-1: Collect several historical target intelligence files, and perform preprocessing to obtain several corresponding preprocessed historical target intelligence files;

[0071] Preprocessing includes data cleaning, text extraction, sequence conversion, word segmentation, removal of stop words and noise, and word standardization, such as word form restoration and synonym merging, to improve data quality and provide data support for subsequent model training;

[0072] S1-2: Based on several pre-processed historical target intelligence files, use the natural language processing and deep learning fusion algorithm to build an intelligence processing model and generate several corresponding historical intelligence keyword groups;

[0073] The intelligence processing model is based on the BERT-LSTM-CRF algorithm, and includes a vector representation layer based on the Bidirectional Encoder Representations from Transformers (BERT) algorithm, a semantic feature extraction layer based on the Long Short-Term Memory (LSTM) algorithm, and an intelligence keyword annotation layer based on the Conditional Random Fields (CRF) module, which are sequentially connected;

[0074] The BERT layer is used to extract deep bidirectional contextual representations of text in related clue files and generate a vector representation for each word or tag. These vectors capture the contextual information of the input text. The LSTM layer further extracts the global semantic features of the sequence based on BERT, which can capture long-distance dependencies and keep the memory of previous information in the sequence. Adding the LSTM layer after the BERT layer can help the model better understand the temporal relationships and complex semantic structures in the sequence. The CRF layer is used to perform dependency processing on each keyword entity in the sequence of character-level vectors based on global semantic features, and keyword entity labels are added to predict the optimal label sequence.

[0075] S1-3: Collect historical clue files related to several data sources, and set corresponding data format conversion strategies for historical clue files related to each data source;

[0076] S1-4: According to the data format conversion strategy, data format conversion is performed on a number of heterogeneous history-related clue files to obtain a number of corresponding homogeneous standardized history-related clue files;

[0077] S1-5: preprocessing a number of standardized history-related clue files to obtain a number of corresponding preprocessed standardized history-related clue files;

[0078] S1-6: Based on several pre-processed and standardized historical clue files, a clue processing model is constructed using a natural language processing and deep learning fusion algorithm, and several corresponding historical clue relationship networks are generated;

[0079] The clue processing model is constructed based on the TF-IDF-CI-GCN algorithm, and the clue processing model includes a clue keyword extraction layer constructed based on the Term Frequency-Inverse Document Frequency with ClassInformation (TF-IDF-CI) algorithm, a graph structure construction layer, and a relationship network construction layer constructed based on the Graph Convolutional Network (GCN) algorithm, which are connected in sequence;

[0080] The clue keyword extraction layer extracts the TF-IDF-CI scores of the texts in the relevant clue files. According to the TF-IDF-CI scores, the most important feature words in each file are selected as clue keywords. Compared with the keyword extraction of a single file, the TF-IDF-CI algorithm is suitable for keyword extraction and query in a large number of files or text collections, which improves the efficiency of clue processing. The graph structure construction layer takes each keyword as a node, uses word embedding (such as Word2Vec or GloVe) to define features for each node, and uses node features to calculate the similarity between nodes as the edge construction condition to form a graph structure and provide a framework for the subsequent relationship network. The GCN layer obtains the embedded representation of the nodes, which can capture the semantic information of the nodes and their relationship in the graph. According to the embedded representation, the similarity between the nodes is calculated, and the edges are established according to the similarity threshold or by selecting the most similar nodes, and finally the relationship network is output;

[0081] S1-7: Based on several historical clue relationship networks and several historical intelligence keyword groups, a clue analysis model is constructed using a deep learning algorithm, and several corresponding historical clue analysis results are generated;

[0082] The clue analysis model is built based on the LSTM-GCN-Attention-Random Forest (RF) algorithm, and the clue analysis model includes a data feature extraction layer built based on the LSTM algorithm, a graph feature extraction layer built based on the GCN algorithm, an attention weight layer built based on the Attention mechanism, and a clue analysis layer built based on the RF algorithm. The data feature extraction layer and the graph feature extraction layer are both connected to the attention weight layer, and the attention weight layer is connected to the clue analysis layer.

[0083] The data feature extraction layer extracts the data features of intelligence keyword groups, and the graph feature extraction layer extracts the graph features of the clue relationship network, which can capture the deep relationship between intelligence and clues, as well as the correlation features between clues. The attention weight layer splices data features and graph features according to the preset attention weights to improve the ability to represent features. The clue analysis layer performs label prediction and classification through the internally trained Classification And Regression Tree (CART);

[0084] S1-9: Based on the analysis results of several historical clues, a retrieval adjustment model is constructed using a reinforcement learning algorithm, and several corresponding historical retrieval adjustment strategy generation experiences are generated;

[0085] The retrieval adjustment model is built based on the Deep Q Network (DQN) algorithm, and the retrieval adjustment model includes an intelligent agent, a deep Q network, and an experience replay pool, which stores several historical retrieval adjustment strategy generation experiences;

[0086] The steps include:

[0087] S1-9-1: Generate questions based on the retrieval adjustment strategy, define the simulation environment of the DQN algorithm, and build the agent and experience replay pool;

[0088] S1-9-2: Convert the historical clue analysis results into a number of historical state parameters, and define the state space of the DQN algorithm based on the several historical state parameters;

[0089] S1-9-3: Convert the preset historical retrieval adjustment strategy into a number of historical action parameters, and define the action space of the DQN algorithm based on the number of historical action parameters;

[0090] S1-9-4: Define the reward function of the DQN algorithm based on the influence of each action parameter in the action space;

[0091] S1-9-5: Construct the input layer, several hidden layers, and output layer of the deep Q network, connect the input layer to the state space, and connect the output layer to the action space;

[0092] S1-9-6: Based on the state space, action space and reward function, and according to the analysis results of several historical clues, the deep Q network and the agent are optimized and trained to build a retrieval adjustment model;

[0093] S1-9-7: Generate a number of historical retrieval adjustment strategy generation experiences, and store the number of historical retrieval adjustment strategy generation experiences in an experience playback pool;

[0094] S2: Collect real-time target intelligence files, and perform intelligence processing on the target intelligence files according to the intelligence processing model to obtain corresponding real-time intelligence keyword groups, including the following steps:

[0095] S2-1: Collect real-time target intelligence files, perform preprocessing, obtain corresponding real-time target intelligence text sequences, and input the real-time target intelligence text sequences into the intelligence processing model;

[0096] Preprocessing includes text extraction, sequence conversion, word segmentation, removal of stop words and noise, and word standardization, such as word form restoration and synonym merging.

[0097] S2-2: Use the vector representation layer of the intelligence processing model to perform vector representation on the real-time target intelligence text sequence to obtain the corresponding real-time target intelligence text word vector;

[0098] S2-3: Use the semantic feature extraction layer of the intelligence processing model to extract semantic features from the word vectors of the real-time target intelligence text to obtain the corresponding semantic features of the real-time target intelligence text;

[0099] S2-4: Based on the semantic features of the real-time target intelligence text, the intelligence keyword annotation layer of the intelligence processing model is used to annotate the real-time target intelligence text sequence with intelligence keywords to obtain an initial real-time intelligence keyword group;

[0100] S2-5: Using the preset corpus, the initial real-time intelligence keyword group is modified and expanded to obtain the final real-time intelligence keyword group, including the following steps:

[0101] S2-5-1: Perform sequence conversion on the initial real-time intelligence keyword group to obtain the initial real-time intelligence keyword sequence.

[0102] S2-5-2: obtaining semantic similarity between the real-time intelligence keywords in the initial real-time intelligence keyword sequence and the corpus keywords in the preset corpus, and taking the corpus keywords whose semantic similarity exceeds a threshold as correction keywords;

[0103] S2-5-3: obtaining a number of extended keywords according to the entity relationship between the modified keyword and other corpus keywords in the preset corpus;

[0104] S2-5-4: adding a number of modified keywords and a number of extended keywords to the initial real-time intelligence keyword sequence to obtain a final real-time intelligence keyword sequence, and converting it into a final real-time intelligence keyword group;

[0105] S3: According to the preset real-time clue retrieval scheme, a clue search is performed on the real-time intelligence keyword group to obtain a number of corresponding real-time related clue files, including the following steps:

[0106] S3-1: According to the preset real-time clue retrieval plan, set the domain parameters, range parameters, depth parameters and frequency parameters of clue retrieval;

[0107] The domain parameters of clue retrieval include the laws and regulations involved, the intelligence industry, the intelligence type, etc., the scope parameters include the websites and databases involved, etc., the depth parameters include the degree of retrieval required, the access process, etc., and the frequency parameters include the retrieval time, the number of retrievals, etc.;

[0108] S3-2: Based on the domain parameter, scope parameter, depth parameter and frequency parameter, a clue search is performed on several data sources on the Internet to obtain several heterogeneous real-time candidate files;

[0109] S3-3: Obtain the semantic similarity between the real-time intelligence keyword group and a number of real-time alternative files, and use the real-time alternative files with semantic similarity exceeding the threshold as real-time relevant clue files;

[0110] S3-4: According to the preset data format conversion strategy, convert the data formats of heterogeneous real-time relevant clue files to obtain a number of corresponding standardized real-time relevant clue files;

[0111] S4: Use the clue processing model to process a number of real-time relevant clue files to obtain the corresponding real-time clue relationship network, including the following steps:

[0112] S4-1: Preprocess a number of real-time relevant clue files to obtain a number of corresponding real-time relevant clue text sequences, and input the number of real-time relevant clue text sequences into the clue processing model;

[0113] The preprocessing includes the following contents:

[0114] Word segmentation: Split the text into words or terms (in Chinese, this usually involves splitting the text into individual characters or words);

[0115] Stop word removal: Delete common words that carry little information, such as "de", "he", "shi", etc.;

[0116] Stem extraction: Reduce words to their basic forms;

[0117] Part-of-speech tagging: Identify the part of speech of words and retain or delete certain parts of speech as needed;

[0118] Case normalization: Convert all words to lowercase to ensure consistency;

[0119] S4-2: Use the clue keyword extraction layer of the clue processing model to extract the real-time clue keyword groups of a number of real-time relevant clue text sequences, including the following steps:

[0120] S4-2-1: Obtain the word frequency and inverse document frequency of each feature word in each real-time relevant clue text sequence;

[0121] S4-2-2: The between-class dispersion factor and within-class dispersion factor of each feature word in different real-time relevant clue text sequences;

[0122] The formula is:

[0123]

[0124] In the formula, CI ac is the between-class dispersion factor; S(t j ) is the standard deviation of the feature word t j ; c kThe feature word t j categories; |C| is the total number of categories; TF(t j ,c k ) is the feature word t j and category c k The frequency of occurrence in The feature word t j The frequency of occurrence in each category; j is the characteristic word indicator; k is the category indicator;

[0125]

[0126] In the formula, CI ic is the discrete factor within the class; s(t j ,c k ) is category c k The characteristic word t appears in j The total number of j ,c k ) includes the feature word t j Category c k The number of real-time relevant clue text sequences; N(c k ) is category c k The number of real-time relevant clue text sequences;

[0127] S4-2-3: Generate the corresponding TF-IDF-CI score according to the word frequency, inverse document frequency, inter-class discrete factor and intra-class discrete factor of the feature word;

[0128]

[0129] In the formula, TF-IDF-CI is the feature word t j TF-IDF-CI score; TF is term frequency; IDF is inverse text frequency; CI is ic is the discrete factor within the class; CI ac is the inter-class discrete factor;

[0130] S4-2-4: Use the top several feature words in the TF-IDF-CI score ranking as clue keywords to obtain the corresponding real-time clue keyword sequence;

[0131] S4-2-5: integrating the real-time clue keyword sequences of all real-time related clue text sequences, including removing duplicate words, to obtain corresponding real-time clue keyword groups;

[0132] S4-3: Use the graph structure construction layer of the clue processing model to construct a graph structure for the real-time clue keyword group to obtain a real-time clue relationship network graph structure;

[0133] S4-4: According to the real-time clue relationship network graph structure, using the relationship network construction layer of the clue processing model, a relationship network is constructed for a number of real-time clue keywords to obtain a corresponding real-time clue relationship network;

[0134] S5: Using the clue analysis model, the real-time intelligence keyword group and the real-time clue relationship network are analyzed to obtain the corresponding real-time clue analysis results, including the following steps:

[0135] S5-1: inputting the final real-time intelligence keyword sequence and the real-time clue relationship network corresponding to the final real-time intelligence keyword group into the clue analysis model;

[0136] S5-2: Use the data feature extraction layer of the clue analysis model to extract the real-time data features of the final real-time intelligence keyword sequence;

[0137] S5-3: Use the graph feature extraction layer of the clue analysis model to extract real-time graph features of the real-time clue relationship network;

[0138] S5-4: according to the attention weight value preset in the attention weight layer of the clue analysis model, the real-time data feature and the real-time graph feature are spliced ​​to obtain the real-time spliced ​​feature;

[0139] S5-5: Use the clue analysis layer to perform clue analysis based on the real-time splicing features to obtain corresponding real-time clue analysis results;

[0140] S6: If the real-time clue analysis result does not meet the requirements, proceed to the next step, otherwise generate a corresponding real-time intelligence clue report according to the real-time clue analysis result and the real-time clue relationship network, and end the method;

[0141] In this embodiment, the existing GPT-3 model is used to process the real-time clue analysis results and the real-time clue relationship network to generate a corresponding real-time intelligence clue report;

[0142] S7: According to the real-time clue analysis result, the preset real-time clue retrieval scheme is adjusted using the retrieval adjustment model to obtain the corresponding adjusted real-time clue retrieval scheme, and the clue retrieval is performed again, including the following steps:

[0143] S7-1: extracting several historical retrieval adjustment strategy generation experiences from the experience playback pool, and updating the retrieval adjustment model according to the real-time clue analysis results and the several historical retrieval adjustment strategy generation experiences to obtain an updated retrieval adjustment model, including the following steps:

[0144] S7-1-1: Extract several historical retrieval adjustment strategies from the experience replay pool, and update the action space of the retrieval adjustment model based on the several historical retrieval adjustment strategies to obtain the updated action space A'=[a'1,...,a' j" ,...,a' I ], where a' j" is the updated j-th "action value", j" is the action indicator, and I is the total number of action space dimensions;

[0145] S7-1-2: Update the state space of the retrieval adjustment model according to the real-time clue analysis results to obtain the updated state space S'=[s'1,...,s' i" ,...,s' I ], where s' i" is the updated i'th state value, i' is the state indicator, and I' is the total number of state space dimensions;

[0146] S7-1-3: Update the input layer of the deep Q network according to the updated state space, and update the output layer of the deep Q network according to the updated action space to obtain an updated retrieval adjustment model;

[0147] S7-2: using the updated retrieval adjustment model, generating a real-time retrieval adjustment strategy and corresponding real-time retrieval adjustment strategy generation experience, and storing the real-time retrieval adjustment strategy generation experience in an experience playback pool, including the following steps:

[0148] S7-2-1: Use the agent to control the updated deep Q network in the updated retrieval adjustment model and output the updated action space A' = [a'1, ..., a' j' ,...,a' I ] is the Q-value sequence of the possible action parameter sequence;

[0149] S7-2-2: Iteratively update the Q value sequence according to the preset reward function to obtain an updated Q value sequence until the number of iterations reaches the threshold;

[0150] The formula is:

[0151] Q(s' p' ,a' p' )=(1-α")·Q(s p' ,a p' )+α"·(R(s p' ,a p' ,s' p' )+γ·Q max (s p' ,a p' ))

[0152] In the formula, Q(s'p' ,a' p' ) is the updated state value s' p' and the updated action value a' p' The corresponding updated Q value; Q(s p' ,a p' ) is the state value s p' and action value a p' The corresponding predicted Q value; α" is the learning rate; Q max (s p' ,a p' ) is the highest predicted Q value; p' is the comprehensive indicator; γ is the update parameter;

[0153] S7-2-3: According to the greedy strategy, the possible action parameter with the highest updated Q value in the updated Q value sequence of each iteration is used as the execution action of the corresponding state parameter;

[0154] S7-2-4: Integrate the execution actions corresponding to all state parameters in the updated state space to obtain a real-time retrieval adjustment strategy;

[0155] S7-3: According to the real-time retrieval adjustment strategy, the preset real-time clue retrieval plan is adjusted to obtain the corresponding adjusted real-time clue retrieval plan, and the clue retrieval is performed again.

[0156] Embodiment 2:

[0157] like Figure 2 As shown, this embodiment provides an intelligence clue analysis system based on artificial intelligence, which is used to implement an intelligence clue analysis method. The system includes a model building unit, an intelligence processing unit, a clue retrieval unit, a clue processing unit, a clue analysis unit, a report generation unit, and a retrieval adjustment unit connected in sequence;

[0158] A model building unit, for building an intelligence processing model, a clue processing model, a clue analysis model, and a retrieval adjustment model using an artificial intelligence algorithm;

[0159] The intelligence processing unit is used to collect real-time target intelligence files and perform intelligence processing on the target intelligence files according to the intelligence processing model to obtain corresponding real-time intelligence keyword groups;

[0160] A clue retrieval unit is used to perform clue retrieval on the real-time intelligence keyword group according to a preset real-time clue retrieval scheme to obtain a corresponding number of real-time related clue files;

[0161] A clue processing unit, used to use a clue processing model to process a number of real-time related clue files to obtain a corresponding real-time clue relationship network;

[0162] A clue analysis unit, used to use a clue analysis model to perform clue analysis on the real-time intelligence keyword group and the real-time clue relationship network to obtain corresponding real-time clue analysis results;

[0163] A report generation unit, used to generate a corresponding real-time intelligence clue report according to the real-time clue analysis results and the real-time clue relationship network;

[0164] The retrieval adjustment unit is used to adjust the preset real-time clue retrieval plan according to the real-time clue analysis result by using the retrieval adjustment model, obtain the corresponding adjusted real-time clue retrieval plan, and re-perform clue retrieval.

[0165] The present invention discloses an intelligence clue analysis method and system based on artificial intelligence. The method and system combine artificial intelligence algorithms to realize automatic and intelligent intelligence clue analysis, improve the intelligence level and efficiency, and reduce cost investment. The method is suitable for application scenarios of massive data retrieval and analysis. The method extracts keywords from intelligence files through an intelligence processing model, simplifies the complexity of intelligence analysis, and improves the efficiency of intelligence analysis. The method performs automatic retrieval through intelligence keywords, improves data processing capability and clue retrieval efficiency, and performs batch processing on the retrieved clue files through the clue processing model, further improving data processing capability. The generated clue relationship network can accurately reflect the correlation between clues, and improves the accuracy of intelligence clue analysis. The method performs automatic analysis through the clue analysis model, avoids the deviation and error existing in manual methods, excavates the deep relationship between intelligence and clues, and improves the efficiency and accuracy of intelligence clue analysis. The method realizes a feedback mechanism through a retrieval adjustment model, and can change the retrieval strategy according to the clue analysis results, thereby improving practicality.

[0166] The present invention is not limited to the above optional implementations, and anyone can derive other various forms of products under the enlightenment of the present invention. The above specific implementations should not be understood as limiting the scope of protection of the present invention. The scope of protection of the present invention should be based on the definition in the claims, and the description can be used to interpret the claims.

Claims

1. An intelligence clue analysis method based on artificial intelligence, characterized by: The steps include: Use artificial intelligence algorithms to build intelligence processing models, clue processing models, clue analysis models, and retrieval adjustment models; Collect real-time target intelligence files, and perform intelligence processing on the target intelligence files according to the intelligence processing model to obtain the corresponding real-time intelligence keyword groups; According to the preset real-time clue retrieval scheme, the real-time intelligence keyword group is searched for clues to obtain a number of corresponding real-time related clue files; Using the clue processing model, several real-time related clue files are processed to obtain the corresponding real-time clue relationship network, including the following steps: Preprocessing a number of real-time related clue files to obtain a number of corresponding real-time related clue text sequences, and inputting the number of real-time related clue text sequences into a clue processing model; Extracting a number of real-time clue keyword groups from real-time related clue text sequences includes the following steps: Obtain the word frequency and inverse document frequency of each feature word in each real-time relevant clue text sequence; Obtain the inter-class discrete factor and intra-class discrete factor of each feature word in different real-time related clue text sequences; The formula is: In the formula, is the inter-class discrete factor; Characteristic word The standard deviation of Characteristic word Category of; is the total number of categories; Characteristic word In category The frequency of occurrence in Characteristic word frequency of occurrence in each category; is the indicator of the characteristic word; is the category indicator; In the formula, is the discrete factor within the class; For Category Feature words appear in Total number of; To include feature words Category The number of real-time relevant clue text sequences; For Category The number of real-time relevant clue text sequences; Generate the corresponding TF-IDF-CI score based on the word frequency, inverse document frequency, inter-class discrete factor and intra-class discrete factor of the feature word; In the formula, Characteristic word TF-IDF-CI score; is the word frequency; is the inverse text frequency; is the discrete factor within the class; is the inter-class discrete factor; The top several feature words ranked by TF-IDF-CI scores are used as clue keywords to obtain the corresponding real-time clue keyword sequence; Integrate the real-time clue keyword sequences of all real-time related clue text sequences, including removing duplicate words, to obtain corresponding real-time clue keyword groups; Constructing a graph structure for the real-time clue keyword group to obtain a real-time clue relationship network graph structure; According to the real-time clue relationship network graph structure, a relationship network is constructed for several real-time clue keywords to obtain the corresponding real-time clue relationship network; Use the clue analysis model to analyze the real-time intelligence keyword groups and the real-time clue relationship network to obtain the corresponding real-time clue analysis results; If the real-time clue analysis result does not meet the requirements, then proceed to the next step, otherwise, generate a corresponding real-time intelligence clue report based on the real-time clue analysis result and the real-time clue relationship network, and end the method; According to the real-time clue analysis results, the preset real-time clue retrieval plan is adjusted using the retrieval adjustment model to obtain the corresponding adjusted real-time clue retrieval plan, and the clue retrieval is performed again.

2. The intelligence clue analysis method based on artificial intelligence according to claim 1 is characterized in that: Using artificial intelligence algorithms, constructing intelligence processing models, clue processing models, clue analysis models, and retrieval adjustment models includes the following steps: Collecting a number of historical target intelligence files, and preprocessing them to obtain a number of corresponding preprocessed historical target intelligence files; Based on several pre-processed historical target intelligence files, a natural language processing and deep learning fusion algorithm is used to build an intelligence processing model and generate several corresponding historical intelligence keyword groups; Collect historical related clue files of several data sources, and set corresponding data format conversion strategies for historical related clue files of each data source; According to the data format conversion strategy, data format conversion is performed on several heterogeneous historical clue files to obtain several corresponding homogeneous standardized historical clue files; Preprocessing a number of standardized history-related clue files to obtain a number of corresponding preprocessed standardized history-related clue files; Based on several pre-processed and standardized historical clue files, a clue processing model is constructed using a fusion algorithm of natural language processing and deep learning, and several corresponding historical clue relationship networks are generated; Based on several historical clue relationship networks and several historical intelligence keyword groups, a clue analysis model is constructed using a deep learning algorithm, and several corresponding historical clue analysis results are generated; According to the analysis results of several historical clues, a retrieval adjustment model is constructed using a reinforcement learning algorithm, and several corresponding historical retrieval adjustment strategy generation experiences are generated.

3. The intelligence clue analysis method based on artificial intelligence according to claim 2 is characterized in that: The intelligence processing model is based on the BERT-LSTM-CRF algorithm; The clue processing model is constructed based on the TF-IDF-CI-GCN algorithm; The clue analysis model is built based on the LSTM-GCN-Attention-RF algorithm; The retrieval adjustment model is constructed based on the DQN algorithm.

4. The intelligence clue analysis method based on artificial intelligence according to claim 3 is characterized in that: Collect real-time target intelligence files, and perform intelligence processing on the target intelligence files according to the intelligence processing model to obtain the corresponding real-time intelligence keyword group, including the following steps: Collect real-time target intelligence files, perform preprocessing, obtain corresponding real-time target intelligence text sequences, and input the real-time target intelligence text sequences into the intelligence processing model; Perform vector representation on the real-time target intelligence text sequence to obtain the corresponding real-time target intelligence text word vector; Extract semantic features from the word vectors of the real-time target intelligence text to obtain the corresponding semantic features of the real-time target intelligence text; Tagging intelligence keywords on the real-time target intelligence text sequence to obtain the initial real-time intelligence keyword group; Using the preset corpus, the initial real-time intelligence keyword group is modified and expanded to obtain the final real-time intelligence keyword group.

5. The intelligence clue analysis method based on artificial intelligence according to claim 4 is characterized in that: Using the preset corpus, the initial real-time intelligence keyword group is modified and expanded to obtain the final real-time intelligence keyword group, including the following steps: The initial real-time intelligence keyword group is sequenced to obtain the initial real-time intelligence keyword sequence. Obtaining semantic similarity between real-time intelligence keywords in the initial real-time intelligence keyword sequence and corpus keywords in a preset corpus, and using corpus keywords whose semantic similarity exceeds a threshold as correction keywords; According to the entity relationship between the revised keyword and other corpus keywords in the preset corpus, a number of extended keywords are obtained; A number of modified keywords and a number of extended keywords are added to the initial real-time intelligence keyword sequence to obtain a final real-time intelligence keyword sequence, and converted into a final real-time intelligence keyword group.

6. The intelligence clue analysis method based on artificial intelligence according to claim 3 is characterized in that: According to the preset real-time clue retrieval scheme, the real-time intelligence keyword group is searched for clues to obtain a number of corresponding real-time related clue files, including the following steps: According to the preset real-time clue retrieval plan, set the domain parameters, range parameters, depth parameters and frequency parameters of clue retrieval; According to the domain parameter, scope parameter, depth parameter and frequency parameter, a clue search is performed on several data sources on the Internet to obtain several heterogeneous real-time candidate files; Obtaining semantic similarity between a real-time intelligence keyword group and a number of real-time candidate files, and using the real-time candidate files whose semantic similarity exceeds a threshold as real-time related clue files; According to a preset data format conversion strategy, data format conversion is performed on heterogeneous real-time related clue files to obtain corresponding standardized real-time related clue files.

7. The intelligence clue analysis method based on artificial intelligence according to claim 5 is characterized in that: Using the clue analysis model, the real-time intelligence keyword group and the real-time clue relationship network are analyzed to obtain the corresponding real-time clue analysis results, including the following steps: Inputting the final real-time intelligence keyword sequence and the real-time clue relationship network corresponding to the final real-time intelligence keyword group into the clue analysis model; Extract the real-time data features of the final real-time intelligence keyword sequence; Extract real-time graph features of real-time clue relationship network; According to the preset attention weight value, the real-time data features and the real-time graph features are spliced ​​to obtain the real-time spliced ​​features; Conduct clue analysis based on the real-time splicing features to obtain corresponding real-time clue analysis results.

8. The intelligence clue analysis method based on artificial intelligence according to claim 3 is characterized in that: According to the real-time clue analysis result, the preset real-time clue retrieval plan is adjusted by using the retrieval adjustment model to obtain the corresponding adjusted real-time clue retrieval plan, and the clue retrieval is performed again, including the following steps: Extracting several historical retrieval adjustment strategy generation experiences, and updating the retrieval adjustment model according to the real-time clue analysis results and several historical retrieval adjustment strategy generation experiences, to obtain an updated retrieval adjustment model; Using the updated retrieval adjustment model, generating a real-time retrieval adjustment strategy and corresponding real-time retrieval adjustment strategy generation experience, and storing the real-time retrieval adjustment strategy generation experience; According to the real-time retrieval adjustment strategy, the preset real-time clue retrieval plan is adjusted to obtain the corresponding adjusted real-time clue retrieval plan, and the clue retrieval is performed again.

9. An intelligence clue analysis system based on artificial intelligence, used to implement the intelligence clue analysis method according to any one of claims 1 to 8, characterized in that: The system comprises a model building unit, an intelligence processing unit, a clue retrieval unit, a clue processing unit, a clue analysis unit, a report generating unit and a retrieval adjustment unit which are connected in sequence.

Images