Knowledge graph construction method and system based on archaeological literature and medium

Through the data cleaning of archaeological document materials, the combination of word segmentation and deep learning models, the entity relationships in archaeological texts are extracted, and the complex problem of inter-entity relationship extraction in the Chinese context is solved, and the accuracy of archaeological data processing and knowledge graph are improved. The construction effect.

CN118917312APending Publication Date: 2024-11-08ZHEJIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410940712.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the Chinese context, the process of extracting relationships between entities in archaeological document data is complex and difficult, which affects the in-depth understanding of archaeological data structures and the application of knowledge graph technology in archaeological literature analysis.

Method used

Using a knowledge graph construction method based on archaeological literature materials, feature extraction and entity recognition are carried out through data cleaning, word segmentation, two-way long and short-term memory recurrent neural network model and conditional random field model, and entity relationships are extracted in combination with attention mechanism model to build an archaeological field. knowledge graph.

Benefits of technology

It improves the accuracy of key entity recognition and the accuracy of relationship extraction in archaeological texts. It is suitable for small sample domains, has strong expansion, and is suitable for situations where large sample training sets are lacking.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917312A_ABST
    Figure CN118917312A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of natural language processing, and discloses a knowledge graph construction method and system based on archaeological literature literature and a medium, and the method comprises the steps: collecting archaeological mining literature literature, and carrying out data cleaning and preprocessing to obtain an archaeological text; setting a word segmentation tool according to the archaeological information features, and performing sentence segmentation and word segmentation on the archaeological text based on the word segmentation tool to obtain a word vector text; performing feature extraction on the word vector text based on BILSTM to obtain a word feature sequence; labeling the word feature sequence based on CRF, and identifying a target entity in the word feature sequence; based on an entity relationship extraction model, extracting a relationship between target entities to obtain an entity relationship triple; and disambiguating the entity relationship triad, and constructing a knowledge graph oriented to the archaeological field. According to the method, the unique text features and relationships in the archaeological data are effectively processed, and the accuracy and discriminability of key entity recognition in the archaeological data are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of natural language processing technology, and in particular to a method, system and medium for constructing a knowledge graph based on archaeological literature. Background Art

[0002] As direct witnesses of ancient human activities, archaeological sites reflect the lifestyle and civilization process of ancient human society. The data contained in these sites are diverse in form and contain profound archaeological value and knowledge. The archaeological documents that record this information have detailed important data such as the name, geographical location and historical period of the site, becoming a valuable resource for archaeological research. With the development of network technology, the number of online archaeological documents is increasing, and the overload and dispersion of information are becoming increasingly significant. It is urgent to effectively utilize these data through integration and mining. Traditional information recognition methods that rely on manual work are not only time-consuming but also inefficient, especially when processing large-scale network text data.

[0003] In this context, with the development of computer and network technology, the rise of artificial intelligence, especially knowledge graph technology, has brought revolutionary changes to the processing of archaeological data. Information extraction technology combining deep learning and natural language processing can efficiently extract key data from a large amount of unstructured text, significantly improving the efficiency of information processing. These technologies not only optimize the extraction process of spatiotemporal information in archaeological texts, but also provide new tools and methods for constructing archaeological knowledge graphs and the integration and fusion of knowledge. Textual materials in the field of archaeology contain rich historical and cultural information, and the relationships between entities provide indispensable historical clues. However, the extraction process of these relationships in the Chinese context is complex and difficult, which constitutes a major difficulty in current technical research. It not only affects the in-depth understanding of the structure of archaeological data, but also restricts the wider application of knowledge graph technology in archaeological literature analysis. Therefore, there is an urgent need for a knowledge graph construction method, system and medium based on archaeological literature data to realize the construction of knowledge graphs. Summary of the invention

[0004] The purpose of the present invention is to provide a method, system and medium for constructing a knowledge graph based on archaeological literature, so as to solve the problem that the extraction process of the relationship between entities in the Chinese context is complex and difficult, which not only affects the in-depth understanding of the archaeological data structure, but also restricts the wider application of knowledge graph technology in archaeological literature analysis.

[0005] The present invention provides a method for constructing a knowledge graph based on archaeological literature, the method comprising:

[0006] Collecting archaeological excavation literature and data, performing data cleaning and preprocessing on the archaeological excavation literature and data, and obtaining archaeological texts;

[0007] According to the information features of the archaeological text, a word segmentation tool of archaeological knowledge is set, and the archaeological text is segmented into sentences and words based on the word segmentation tool to obtain a word vector text;

[0008] Performing feature extraction on the word vector text based on a bidirectional long short-term memory recurrent neural network model to obtain a word feature sequence;

[0009] Annotating the word feature sequence based on a conditional random field model, and identifying a target entity in the word feature sequence;

[0010] Extract the relationship between the target entities based on the entity relationship extraction model to obtain entity relationship triples;

[0011] The entity relationship triples are disambiguated to construct a knowledge graph for the archaeological field.

[0012] Preferably, the archaeological excavation literature is subjected to data cleaning and preprocessing, including: removing special symbols and repeated fields in the archaeological excavation literature, and retaining periods and commas;

[0013] Perform ambiguity judgment on the archaeological excavation literature and remove ambiguous data.

[0014] Preferably, the sentence and word segmentation of the archaeological text based on the word segmentation tool includes:

[0015] Defining extraction rules of temporal and spatial features of the archaeological text;

[0016] Based on the extraction rules, the archaeological text is segmented into sentences and words according to the word segmentation tool;

[0017] wherein said temporal characteristics are determined according to the relative age of the archaeological sites;

[0018] The spatial features are determined according to the latitude and longitude coordinates of the geographical location of the archaeological site.

[0019] Preferably, before extracting features from the word vector text based on the bidirectional long short-term memory recurrent neural network model, the method further includes:

[0020] The archaeological texts are annotated using the BIO data annotation strategy and the Yedda data annotation tool to obtain a training set for a bidirectional long short-term memory recurrent neural network model.

[0021] Preferably, feature extraction is performed on the word vector text based on a bidirectional long short-term memory recurrent neural network model, including:

[0022] Define the word vector input at time t as xt , the hidden layer state vector is h t , input x in different states t The weight matrix is ​​U, and the hidden layer state vector is h t The weight matrix is ​​W, the bias vector is b, the activation function is σ and tanh, the forget gate is used to decide the discarded information, the input gate is used to decide the updated information, and the forget gate f at time t t It is expressed as:

[0023] f t =σ(W f h t-1 +U f x t +b f );

[0024] Update h at time t t-1 , input gate i t It is expressed as:

[0025] i t =σ(W i h t-1 +U i x t +b i );

[0026] The candidate value vector added to the cell state at time t It is expressed as:

[0027]

[0028] At time t, the information stored at time t-1 is updated to the information stored at time t, and the updated state vector C is obtained. t It is expressed as:

[0029]

[0030] Output gate o at time t t It is expressed as:

[0031] o t =σ(W o h t-1 +U o X t +b o );

[0032] Update the hidden layer state vector h t :

[0033] h t =o t ×tanh(C t );

[0034] By combining the forward LSTM and the reverse LSTM for each word vector, the outputs at the same time t are merged to obtain the forward and backward state information at each time, so that the feature sequence extracted by the bidirectional long short-term memory recurrent neural network model can fully reflect the bidirectional semantic dependency information:

[0035]

[0036] Preferably, the word feature sequence is labeled based on a conditional random field model, and a target entity in the word feature sequence is identified, including:

[0037] Define the input word sequence z = {z1, z2, ..., z n}, n is the length of the input word sequence, z i is the input vector of the i-th word, and the actual output label sequence y = {y1, y2, ..., y n}, the possible output label sequence of word sequence z is Y(z) = {y'1, y'2, ..., y' n};

[0038] For a word sequence z, define the conditional probability over all possible output label sequences y':

[0039]

[0040] Potential gain function ψ i It is expressed as:

[0041]

[0042] in, and b y',y Represents the weight vector and bias vector corresponding to the label pair (y', y);

[0043] Using the maximum conditional likelihood estimation method, for the training set {(z i ,y i )}, the logarithm of the likelihood is expressed as:

[0044]

[0045] Select the parameters that maximize the likelihood L(W, b), find the highest conditional probability y in the possible output label sequence, and obtain the required entity recognition relationship:

[0046]

[0047] Preferably, extracting the relationship between the target entities based on the entity relationship extraction model includes: the entity relationship extraction model classifies and extracts the entity recognition relationship by constructing a bidirectional long short-term memory recurrent neural network model;

[0048] And use the attention mechanism model to train the weight matrix of the output sequence, specifically:

[0049] Define H = {h1, h2, ..., h n} is the output vector matrix of the BILSTM layer, M is the weight vector of the sequence, and w T is the transpose of the parameter vector. By multiplying the input vector with the sequence weight vector, the word-level features are transformed into sentence-level features, and the relationship between entities is classified and extracted.

[0050] Using the activation function tanh, we get M:

[0051] M = tanh(H);

[0052] Through the softmax fully connected layer, the input vector is multiplied by the sequence weight vector:

[0053] δ=softmax(w T M);

[0054] Calculate the weighted sum of the output vectors to get the feature representation r of the sentence sequence, and use the activation function tanh to get the sentence representation h for classification * :

[0055] r=Hδ T ;

[0056] h * =tanh(r);

[0057] Using the softmax function as a classifier, we get the predicted label y of sentence S as the result of relation extraction:

[0058] p(y|S;W,b)=softmax(W s h * +b s );

[0059] y=argmaxp(y|S;W,b).

[0060] Preferably, disambiguating the entity relationship triples includes:

[0061] By customizing and building a synonym dictionary of the archaeological text, and setting weights for target entities and candidate entities;

[0062] Filter words with weights higher than the weight threshold for entity disambiguation;

[0063] Construct a knowledge representation method based on entity / feature-relationship and generate an expression model of the knowledge graph.

[0064] The present invention also discloses a knowledge graph construction system based on archaeological literature data, including several processors for executing the above-mentioned knowledge graph construction method based on archaeological literature data.

[0065] The present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned knowledge graph construction method based on archaeological literature.

[0066] Compared with the prior art, the beneficial effect of the present invention lies in that the present invention provides a method, system and medium for constructing a knowledge graph based on archaeological literature data, by designing feature extraction rules for archaeological texts, using a bidirectional long short-term memory recurrent neural network model (BILSTM) and a conditional random field model (CRF) to identify archaeological entities in archaeological texts, using an entity relationship extraction model based on BILSTM and an attention mechanism model (Attention) to identify the relationship between entities, constructing a knowledge representation method based on "entity / feature-relationship", and generating an expression model of the knowledge graph.

[0067] The present invention improves the accuracy and discrimination of key entity recognition in archaeology by designing feature extraction rules for archaeological texts and effectively processing unique text features and complex relationships in archaeological literature.

[0068] The present invention fine-tunes and trains small sample corpora for specific domains based on the general large model, which can effectively improve the accuracy of generating knowledge graphs for the specified small sample domains, and is suitable for various situations where large sample training sets are lacking, and has strong scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0070] Figure 1 It is a flow chart of a method for constructing a knowledge graph based on archaeological literature data of the present invention;

[0071] Figure 2 Schematic diagram of a bidirectional long short-term memory recurrent neural network model and a conditional random field model in an embodiment of the present invention;

[0072] Figure 3 is a schematic diagram of an entity relationship extraction model in an embodiment of the present invention;

[0073] Figure 4 It is a schematic diagram of the knowledge graph expression model in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0075] like Figure 1 As shown, the present invention provides a method for constructing a knowledge graph based on archaeological literature, the method comprising:

[0076] S1, collecting archaeological excavation literature data, performing data cleaning and preprocessing on the archaeological excavation literature data, and obtaining archaeological texts.

[0077] In some embodiments of the present application, the archaeological excavation document data is cleaned and preprocessed, including: removing special symbols and repeated fields in the archaeological excavation document data, and retaining periods and commas; performing ambiguity judgment on the archaeological excavation document data, and removing ambiguous data.

[0078] In this embodiment, the archaeological excavation literature used in this application is 500 text data of site encyclopedia entries crawled from Baidu Encyclopedia (https: / / baike.baidu.com / ) and 300 archaeological excavation reports downloaded from HowNet (https: / / www.cnki.net / ). By removing special symbols and redundant information and retaining important punctuation marks such as commas and periods, the collected archaeological excavation literature is subjected to data cleaning tasks to obtain an archaeological corpus dataset.

[0079] S2, according to the information features of the archaeological text, set a word segmentation tool for archaeological knowledge, and perform sentence and word segmentation on the archaeological text based on the word segmentation tool to obtain a word vector text.

[0080] In some embodiments of the present application, the archaeological text is segmented into sentences and words based on the word segmentation tool, including: defining extraction rules for the time characteristics and spatial characteristics of the archaeological text; based on the extraction rules, the archaeological text is segmented into sentences and words according to the word segmentation tool; wherein the time characteristics are determined according to the relative age of the archaeological site; and the spatial characteristics are determined according to the latitude and longitude coordinates of the geographical location of the archaeological site.

[0081] In this embodiment, the site encyclopedia entries used in this application mainly include graphic information such as basic information of the site, historical evolution, cultural relics, and protection measures. The archaeological excavation report mainly includes information such as the geographical location of the site, the investigation process, preservation status, stratigraphic accumulation, and excavated relics. According to the defined archaeological spatiotemporal feature extraction rules, natural language processing tools are used to segment the text into sentences and words, so that each sentence in the document is on a line, and words are separated by spaces.

[0082] S3, extracting features of the word vector text based on a bidirectional long short-term memory recurrent neural network model to obtain a word feature sequence.

[0083] In some embodiments of the present application, before performing feature extraction on the word vector text based on the bidirectional long short-term memory recurrent neural network model, it also includes: using the BIO data annotation strategy as the annotation strategy and the Yedda data annotation tool as the annotation tool to annotate the archaeological text and obtain a training set of the bidirectional long short-term memory recurrent neural network model.

[0084] In this embodiment, the annotation strategy selects the BIO data annotation strategy, the annotation tool selects the Yedda data annotation tool, and a part of the archaeological text corpus is selected for annotation as the training input of the bidirectional long short-term memory recurrent neural network model. Subsequently, the bidirectional long short-term memory recurrent neural network model is used to extract features from the word vector text.

[0085] In some embodiments of the present application, Figure 2 As shown, the bidirectional long short-term memory recurrent neural network model includes a forward LSTM layer, a reverse LSTM layer and a hidden layer. The feature extraction of the word vector text based on the bidirectional long short-term memory recurrent neural network model includes: defining the word vector input at time t as x t , the hidden layer state vector is h t , input x in different states t The weight matrix is ​​U, and the hidden layer state vector is h t The weight matrix is ​​W, the bias vector is b, the activation function is σ and tanh, the forget gate is used to decide the discarded information, the input gate is used to decide the updated information, and the forget gate f at time t t It is expressed as:

[0086] f t =σ(W f h t-1 +U f x t +b f ).

[0087] Update h at time t t-1 , input gate i t It is expressed as:

[0088] i t =σ(W i h t-1 +U i x t +b i ).

[0089] The candidate value vector added to the cell state at time t It is expressed as:

[0090]

[0091] At time t, the information stored at time t-1 is updated to the information stored at time t, and the updated state vector C is obtained. t It is expressed as:

[0092]

[0093] Output gate o at time t t It is expressed as:

[0094] o t =σ(W o h t-1 +U o X t +b o ).

[0095] Update the hidden layer state vector h t :

[0096] h t =o t ×tanh(C t ).

[0097] By combining the forward LSTM and the reverse LSTM for each word vector, the outputs at the same time t are merged to obtain the forward and backward state information at each time, so that the feature sequence extracted by the bidirectional long short-term memory recurrent neural network model can fully reflect the bidirectional semantic dependency information:

[0098]

[0099] S4, labeling the word feature sequence based on a conditional random field model, and identifying a target entity in the word feature sequence.

[0100] In some embodiments of the present application, Figure 2 As shown, the conditional random field model is a CRF layer. The word feature sequence is labeled based on the conditional random field model, and the target entity in the word feature sequence is identified, including: defining an input word sequence z = {z1, z2, ..., z n}, n is the length of the input word sequence, z i is the input vector of the i-th word, and the actual output label sequence y = {y1, y2, ..., y n}, the possible output label sequence of word sequence z is Y(z) = {y'1, y'2, ..., y' n};

[0101] For a word sequence z, define the conditional probability over all possible output label sequences y':

[0102]

[0103] Potential gain function ψ i It is expressed as:

[0104]

[0105] in, and b y',y Represents the weight vector and bias vector corresponding to the label pair (y', y);

[0106] Using the maximum conditional likelihood estimation method, for the training set {(z i ,y i )}, the logarithm of the likelihood is expressed as:

[0107]

[0108] Select the parameters that maximize the likelihood L(W, b), find the highest conditional probability y in the possible output label sequence, and obtain the required entity recognition relationship:

[0109]

[0110] S5, extracting the relationship between the target entities based on the entity relationship extraction model to obtain entity relationship triples.

[0111] In some embodiments of the present application, by constructing a bidirectional long short-term memory recurrent neural network model and an attention mechanism model, the entity recognition relationship obtained in step S4 is classified and extracted. The bidirectional long short-term memory recurrent neural network model has the same processing process as that described in step S4, and the attention mechanism model is used to train the weight matrix of the output sequence. Figure 3 As shown, the entity relationship extraction model includes a BILSTM layer and an Attention layer. The relationship between the target entities is extracted based on the entity relationship extraction model, including: the entity relationship extraction model classifies and extracts the entity recognition relationship by constructing a bidirectional long short-term memory recurrent neural network model; and uses the attention mechanism model to train the weight matrix of the output sequence, specifically: define H = {h1, h2, ..., h n} is the output vector matrix of the BILSTM layer, M is the weight vector of the sequence, and w T is the transpose of the parameter vector. By multiplying the input vector with the sequence weight vector, the word-level features are transformed into sentence-level features, and the relationship between entities is classified and extracted.

[0112] Using the activation function tanh, we get M:

[0113] M = tanh(H).

[0114] Through the softmax fully connected layer, the input vector is multiplied by the sequence weight vector:

[0115] δ=softmax(w T M).

[0116] Calculate the weighted sum of the output vectors to get the feature representation r of the sentence sequence, and use the activation function tanh to get the sentence representation h for classification * :

[0117] r=Hδ T .

[0118] h * =tanh(r).

[0119] Using the softmax function as a classifier, we get the predicted label y of sentence S as the result of relation extraction:

[0120] p(y|S;W,b)=softmax(W s h * +b s ).

[0121] y=argmaxp(y|S;W,b).

[0122] It is understandable that in order to objectively evaluate the performance of the method of this embodiment, this implementation method designs the accuracy P and recall R as evaluation indicators of the relationship extraction model. The larger the values ​​of P and R, the better the performance of the model accuracy and recall. The calculation formulas of P and R are defined as follows:

[0123]

[0124] After experimental training, the overall performance of the model is good. The experimental results show that the model accuracy is 92.3% and the recall rate is 79.7%. The entity recognition and relationship extraction results obtained on the selected data set are shown in Table 1.

[0125] Table 1 Example of experimental results of the method of the present invention on the data set

[0126]

[0127] S6, disambiguating the entity relationship triples and constructing a knowledge graph for the archaeological field.

[0128] In some embodiments of the present application, the entity relationship triples are disambiguated, including: customizing the construction of a synonym dictionary for the archaeological text and setting weights for the target entity and candidate entities; screening words with weights higher than a weight threshold for entity disambiguation; and constructing an entity / feature-relationship-based knowledge representation method to generate an expression model of the knowledge graph.

[0129] It can be understood that this embodiment, based on the entity recognition and relationship extraction results obtained, constructs a synonym dictionary of the archaeological text by customizing it, sets weights for the target entity and candidate entities, and performs entity disambiguation by screening phrases with higher weights; constructs a knowledge representation method based on "entity / feature-relationship" to generate an expression model of the archaeological knowledge graph, such as Figure 4 shown.

[0130] The present application also discloses a knowledge graph construction system based on archaeological literature data, including several processors for executing the above-mentioned knowledge graph construction method based on archaeological literature data.

[0131] It is understandable that in this embodiment, each functional unit may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0132] The present application also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned knowledge graph construction method based on archaeological literature.

[0133] It is understandable that the storage medium in this embodiment includes: a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, and other media that can store program codes.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

[0135] The system provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiment can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.

[0136] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

Claims

1. A method for constructing a knowledge graph based on archaeological literature, characterized in that: The method comprises: Collecting archaeological excavation literature and data, performing data cleaning and preprocessing on the archaeological excavation literature and data, and obtaining archaeological texts; According to the information features of the archaeological text, a word segmentation tool of archaeological knowledge is set, and the archaeological text is segmented into sentences and words based on the word segmentation tool to obtain a word vector text; Performing feature extraction on the word vector text based on a bidirectional long short-term memory recurrent neural network model to obtain a word feature sequence; Annotating the word feature sequence based on a conditional random field model, and identifying a target entity in the word feature sequence; Extract the relationship between the target entities based on the entity relationship extraction model to obtain entity relationship triples; The entity relationship triples are disambiguated to construct a knowledge graph for the archaeological field.

2. The method for constructing a knowledge graph based on archaeological literature according to claim 1, characterized in that: Performing data cleaning and preprocessing on the archaeological excavation literature data, including: removing special symbols and repeated fields in the archaeological excavation literature data, and retaining periods and commas; Perform ambiguity judgment on the archaeological excavation literature and remove ambiguous data.

3. The method for constructing a knowledge graph based on archaeological literature according to claim 1, characterized in that: Sentence and word segmentation are performed on the archaeological text based on the word segmentation tool, including: Defining extraction rules of temporal and spatial features of the archaeological text; Based on the extraction rules, the archaeological text is segmented into sentences and words according to the word segmentation tool; wherein said temporal characteristics are determined according to the relative age of the archaeological sites; The spatial features are determined according to the latitude and longitude coordinates of the geographical location of the archaeological site.

4. The method for constructing a knowledge graph based on archaeological literature according to claim 1, characterized in that: Before extracting features from the word vector text based on the bidirectional long short-term memory recurrent neural network model, the method further includes: The archaeological texts are annotated using the BIO data annotation strategy and the Yedda data annotation tool to obtain a training set for a bidirectional long short-term memory recurrent neural network model.

5. The method for constructing a knowledge graph based on archaeological literature according to claim 4 is characterized in that: Feature extraction of the word vector text is performed based on a bidirectional long short-term memory recurrent neural network model, including: Define the word vector input at time t as x t , the hidden layer state vector is h t , input x in different states t The weight matrix is ​​U, and the hidden layer state vector is h t The weight matrix is ​​W, the bias vector is b, the activation function is σ and tanh, the forget gate is used to decide the discarded information, the input gate is used to decide the updated information, and the forget gate f at time t t It is expressed as: f t =σ(W f h t-1 +U f x t +b f ); Update h at time t t-1 , input gate i t It is expressed as: i t =σ(W i h t-1 +U i x t +b i ); The candidate value vector added to the cell state at time t It is expressed as: At time t, the information stored at time t-1 is updated to the information stored at time t, and the updated state vector C is obtained. t It is expressed as: Output gate o at time t t It is expressed as: the t =σ(W o h t-1 +U o X t +b o ); Update the hidden layer state vector h t : h t =o t ×tanh(C t ); By combining the forward LSTM and the reverse LSTM for each word vector, the outputs at the same time t are merged to obtain the forward and backward state information at each time, so that the feature sequence extracted by the bidirectional long short-term memory recurrent neural network model can fully reflect the bidirectional semantic dependency information:

6. The method for constructing a knowledge graph based on archaeological literature according to claim 5, characterized in that: The word feature sequence is labeled based on a conditional random field model, and a target entity in the word feature sequence is identified, including: Define the input word sequence z = {z1, z2, ..., z n }, n is the length of the input word sequence, z i is the input vector of the i-th word, and the actual output label sequence y = {y1, y2, ..., y n }, the possible output label sequence of word sequence z is Y(z) = {y'1, y'2, ..., y' n }; For a word sequence z, define the conditional probability over all possible output label sequences y': Potential gain function ψ i It is expressed as: in, and b y',y Represents the weight vector and bias vector corresponding to the label pair (y', y); Using the maximum conditional likelihood estimation method, for the training set {(z i ,y i )}, the logarithm of the likelihood is expressed as: Select the parameters that maximize the likelihood L(W, b), find the highest conditional probability y in the possible output label sequence, and obtain the required entity recognition relationship:

7. The method for constructing a knowledge graph based on archaeological literature according to claim 6, characterized in that: Extracting the relationship between the target entities based on the entity relationship extraction model, including: the entity relationship extraction model classifies and extracts the entity recognition relationship by constructing a bidirectional long short-term memory recurrent neural network model; And use the attention mechanism model to train the weight matrix of the output sequence, specifically: Define H = {h1, h2, ..., h n } is the output vector matrix of the BILSTM layer, M is the weight vector of the sequence, and w T is the transpose of the parameter vector. By multiplying the input vector with the sequence weight vector, the word-level features are transformed into sentence-level features, and the relationship between entities is classified and extracted. Using the activation function tanh, we get M: M = tanh(H); Through the softmax fully connected layer, the input vector is multiplied by the sequence weight vector: δ=softmax(w T M); Calculate the weighted sum of the output vectors to get the feature representation r of the sentence sequence, and use the activation function tanh to get the sentence representation h for classification * : r=Hδ T ; h * =tanh(r); Using the softmax function as a classifier, we get the predicted label y of sentence S as the result of relation extraction: p(y∣S;W,b)=softmax(W s h * +b s ); y=argmaxp(y|S;W,b).

8. The method for constructing a knowledge graph based on archaeological literature according to claim 1, characterized in that: Disambiguating the entity relationship triples includes: By customizing and building a synonym dictionary of the archaeological text, and setting weights for target entities and candidate entities; Filter words with weights higher than the weight threshold for entity disambiguation; Construct a knowledge representation method based on entity / feature-relationship and generate an expression model of the knowledge graph.

9. A knowledge graph construction system based on archaeological literature, characterized in that: It includes several processors for executing the knowledge graph construction method based on archaeological literature data as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the method for constructing a knowledge graph based on archaeological literature data as described in any one of claims 1 to 8 is implemented.