A knowledge perception-based geographic information text intelligent extraction method

Abstract

The application belongs to the technical field of data processing, and discloses a geographic information text intelligent extraction method based on knowledge perception; comprising: constructing a geographic knowledge ontology library, and constructing a corresponding semantic relationship network based on the same; obtaining text geographic entities and text semantic relationships corresponding to input geographic information text; constructing a corresponding initial tree based on the obtained text geographic entities and text semantic relationships, and combining a semantic relationship network; performing knowledge perception and semantic understanding on the corresponding initial tree based on the geographic knowledge ontology library, and obtaining a corresponding semantic tree; performing information extraction based on the obtained semantic tree, and obtaining required geographic information; the application realizes in-depth understanding and intelligent extraction of geographic information text by constructing a geographic knowledge ontology library and a semantic relationship network, and improves the accuracy and efficiency of geographic information extraction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, more particularly, the present application relates to a kind of geographic information text intelligent extraction method based on knowledge perception. BACKGROUND

[0002] In the field of geographic information processing, accurately and efficiently extracting geographic information from text is of great significance;With the wide application of geographic information technology in many fields, such as geographic research, urban planning, environmental monitoring, etc., the demand for geographic information is growing, and higher requirements for its accuracy and integrity are put forward.

[0003] However, traditional geographic information extraction methods face many challenges;On the one hand, the complexity and diversity of geographic information text make information extraction difficult;Geographic text contains a large number of different types of geographic entities, and their semantic relationships are complex and diverse;These complex relationships make it difficult to accurately identify and extract geographic information, and traditional methods often have difficulty fully and accurately grasping these relationships, and are prone to information omission or incorrect extraction;

[0004] In view of this, the present application proposes a kind of geographic information text intelligent extraction method based on knowledge perception to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] A kind of geographic information text intelligent extraction method based on knowledge perception, comprising:

[0007] Step one: build a geographic knowledge ontology library, and build a corresponding semantic relationship network based on it;

[0008] Step two: obtain the text geographic entity and text semantic relationship corresponding to the input geographic information text;

[0009] Step three: based on the obtained text geographic entity and text semantic relationship, and combined with the semantic relationship network, build a corresponding initial tree-shaped tree;Based on the geographic knowledge ontology library, knowledge perception and semantic understanding are carried out on the corresponding initial tree-shaped tree, and a corresponding semantic tree is obtained;

[0010] Step four: based on the obtained semantic tree, information extraction is carried out, and the required geographic information is obtained.

[0011] Further, the process of building a geographic knowledge ontology library includes:

[0012] The existing geographical entity concept is obtained, and is summarized and abstracted as a basic geographical entity, including a natural geographical entity, an artificial geographical entity, and a combined geographical entity; the basic geographical entity is composed of several different geographical entity categories;

[0013] A data collection node is arranged, and the data collection node collects geographical data related to the corresponding basic geographical entity based on different data sources, to obtain corresponding basic geographical data;

[0014] The obtained basic geographical data is preprocessed to obtain corresponding basic text data;

[0015] After the preprocessing is completed, the geographical entities in the corresponding basic text data are identified based on a pre-constructed geographical entity identification model, and the semantic relationship between the corresponding geographical entities is identified;

[0016] Meanwhile, the attribute information of the corresponding geographical entities is extracted based on the basic text data to obtain corresponding entity attributes;

[0017] The corresponding knowledge triple is constructed based on the obtained geographical entities, semantic relationships and entity attributes, and is stored under the corresponding geographical entity category, and the corresponding geographical knowledge ontology library is constructed based thereon.

[0018] Further, the construction process of the geographical entity identification model includes:

[0019] The backbone network of the geographical entity identification model is a multi-level neural network, and the basic framework of the multi-level neural network is an input layer, a double subject and object extraction layer, a full connection layer and an output layer;

[0020] The input layer is used for encoding processing of the input text sentence to obtain the encoding vector corresponding to each word in the corresponding text sentence;

[0021] The double layer includes a first layer and a second layer; the first layer extracts semantic features of the input encoding vector by introducing a control gate mechanism to obtain a word semantic feature vector, and assigns weights to each word in the corresponding text sentence based on the obtained word semantic feature vector to obtain a weighted sentence; wherein the control gate mechanism includes a forget gate, an enhancement gate and an output gate;

[0022] The second layer is used for extracting the word semantic feature vector corresponding to each word in the received weighted sentence, and performing feature fusion to obtain a sentence semantic feature vector;

[0023] The host-guest extraction layer is configured to receive the word semantic feature vector output by the second layer, and obtain the host word position of the geographic entity within the corresponding text sentence based on a pre-set prediction classifier; and obtain the word representation vector corresponding to the host word based on the host word position of the geographic entity;

[0024] Based on the host word position of the geographic entity and the obtaining process of the word representation vector, the guest word position of the geographic entity within the corresponding text sentence and the corresponding word representation vector are obtained, and the host word and the guest word of the corresponding geographic entity within the corresponding text sentence are marked;

[0025] The full connection layer is configured to receive the obtained word representation vector and the sentence semantic feature vector, and combine them to obtain the corresponding entity vector;

[0026] The output layer is configured to map the obtained entity vector to the class probability of a certain semantic class; and perform threshold comparison on the obtained class probability and the corresponding class threshold value, if the class probability is greater than the class threshold value, the corresponding semantic class is taken as the semantic relationship between the corresponding physical entities, and if the class probability is not greater than the class threshold value, no other operation is performed;

[0027] A plurality of geographic text sentence sets with artificial annotation are obtained, and a corresponding training sample set is constructed based thereon; and a geographic entity recognition model is defined as an optimizer to continuously optimize the parameters of the geographic entity recognition model during the training process, the training sample set is input into the geographic entity recognition model in batches, the value of the corresponding loss function is recorded, when the values of the loss function of the continuous L2 batches no longer decrease or change, the parameters of the geographic entity recognition model at this time are saved, that is, the training of the geographic entity recognition model is completed, and L2 is a constant.

[0028] Further, the formula for obtaining the word semantic feature vector is: ; in the formula, and respectively represent the word semantic feature vector corresponding to the th and word within the corresponding text sentence; represents the weight matrix of the output gate in the control gate mechanism; represents the encoding vector corresponding to the th word within the text sentence; represents the bias term of the output gate; is a natural number, and I represents the total number of words within the text sentence; represents vector splicing; represents a pre-selected activation function;

[0029] Indicates the current time Next First Network state data corresponding to the layer, In the formula, and Each represents the weight matrix of the enhancement gate; and Both represent bias terms; the enhancement gate is used to adjust the corresponding first... Update the network status data of the layer; This represents the operation of the hyperbolic tangent function;

[0030] This represents the output of the forget gate corresponding to the i-th word in the text statement. In the formula, and These represent the weight matrix and bias term of the forget gate, respectively;

[0031] The formula for assigning weights is: ; Indicates the first text statement The word weights corresponding to each word; Q, K, and V respectively represent the semantic feature vectors of the corresponding words. The weight matrix obtained by performing a linear transformation; Represents matrix permutation; Represents the weight matrix The matrix dimension.

[0032] Furthermore, the formula for obtaining the keyword position is: In the formula, The main words representing geographical entities in the corresponding text statements The probability of being located at position S1, where S1 represents the position identifier of the subject word of the geographic entity, i.e. the range from the beginning to the end of the subject word; Representing the corresponding number The probability of a word being the starting position of the main word of a geographic entity; where, In the formula, Indicates the training weights; Indicates the bias term; Indicates the first weighted statement The semantic feature vector of each word; ;in, Based on the corresponding probability Decision, if appropriate If the probability exceeds a pre-set threshold, then Conversely, ; represent trainable parameters of the model; and respectively represent the start position marker and the end position marker of the subject word; n represents the position index of the subject word between the start position and the end position of the subject word; represents the total number of words in the text sentence; represents the product operation;

[0033] category probability ; in the formula, represents the probability that the entity vector ST belongs to the th semantic category in the semantic relation, and respectively represent the exponential operation and the activation function operation; represents all semantic categories of the predefined semantic relation;

[0034] Loss function for defining the geographic entity recognition model ; in the formula, and represent the weight ratio, represents the position probability of the object word as a geographic entity; Ld represents the th training sample in the training sample set, represents the semantic category label of the th training sample; represents the th training sample in the training sample set, .

[0035] Further, the construction process of the semantic relation network includes:

[0036] Obtain the geographic entity stored in the corresponding geographic knowledge ontology library, and perform entity disambiguation on it, and mark the entity disambiguated geographic entity as a concept entity; obtain different concept entities and their corresponding semantic relations; and convert the corresponding concept entities into corresponding graph nodes, and convert the semantic relations into corresponding edges, and mark the corresponding directions and relation types;

[0037] Based on the pre-selected directed graph construction algorithm, and based on the graph nodes constructed by the corresponding concept entities and the edges corresponding to the semantic relations, the semantic relation network between the corresponding concept entities is constructed.

[0038] Further, the process of obtaining the text geographic entity and the text semantic relation corresponding to the input geographic information text includes:

[0039] The input geographical information text is pre-processed and input into the geographical entity recognition model to obtain geographical entities and semantic relationships in the corresponding geographical information text, which are marked as text geographical entities and text semantic relationships respectively.

[0040] Further, based on the obtained text geographical entities and text semantic relationships, and in combination with the semantic relationship network, an initial tree-shaped tree is constructed.

[0041] The extracted text geographical entities and text semantic relationships are mapped into the constructed semantic relationship network, and the corresponding graph nodes and edges of the text geographical entities and text semantic relationships are obtained.

[0042] The graph nodes corresponding to the text geographical entities are used as clustering centers to perform clustering operations on other graph nodes in the semantic relationship network, and the graph nodes in the same clustering set as the corresponding text geographical entities are obtained. The concept entities and corresponding semantic relationships of the edges corresponding to the graph nodes are obtained and marked as expanded geographical entities and expanded semantic relationships.

[0043] Further, all the obtained text semantic relationships and expanded semantic relationships are structured and classified to obtain the upstream-downstream relationships or association relationships between different text geographical entities and expanded geographical entities, and a relationship determination matrix is constructed based thereon.

[0044] An enhanced tree structure framework is constructed, which includes a trunk hierarchical structure and a horizontal semantic link.

[0045] The text geographical entity with the highest upstream-downstream relationship in the semantic relationship network is selected as the root node in the trunk hierarchical structure, and other text geographical entities or expanded geographical entities are added to the corresponding trunk hierarchical structure layer by layer based on the upstream-downstream relationships in the relationship determination matrix.

[0046] After the addition is completed, the association relationships are stored in the form of tree node attributes in the enhanced tree structure framework based on the horizontal semantic link, and an initial tree-shaped tree is obtained.

[0047] The obtained initial tree-shaped tree is input into the constructed geographical knowledge base to perform semantic mapping and semantic reasoning, and a corresponding semantic tree is obtained based thereon.

[0048] After the semantic mapping and semantic reasoning are completed, the entity attributes stored in the corresponding address knowledge base and the corresponding semantic reasoning results are stored in the tree nodes in the corresponding semantic tree.

[0049] Further, the semantic mapping refers to mapping the node entity corresponding to the initial tree-shaped tree node and the edge semantic relationship corresponding to the edge between the tree nodes into the geographic knowledge triples in the geographic knowledge ontology library, the node entity refers to the text geographic entity or the expanded geographic entity, the edge semantic relationship refers to the text semantic relationship or the expanded semantic relationship, and the semantic reasoning refers to performing semantic reasoning on other tree nodes adjacent to the corresponding tree node according to the entity attributes in the mapped knowledge triples.

[0050] Further, the process of obtaining geographic information based on the obtained semantic tree includes:

[0051] The root node of the corresponding semantic tree is obtained as a starting node, the corresponding semantic tree is traversed based on a preset extraction rule, the tree nodes and the edges corresponding to the tree nodes in the corresponding semantic tree that meet the preset extraction rule are marked, the entity attributes and the semantic reasoning results stored in the corresponding tree nodes and the edge semantic relationship of the tree nodes to the edges are read, and the corresponding geographic information is obtained; the obtained geographic information is de-duplicated and summarized to obtain integrated information, and whether the semantic relationship between geographic entities involved in the integrated information meets the requirements is checked, if not, the corresponding integrated information is sorted and adjusted according to the geographic knowledge base; after the adjustment is completed, the corresponding integrated information is converted into a required format file and is visually output.

[0052] The technical effects and advantages of the geographic information text intelligent extraction method based on knowledge perception of the application are as follows:

[0053] 1. By constructing the geographic knowledge ontology library and the semantic relationship network, combining the input text geographic entity and the semantic feature to construct the semantic tree, and using the ontology library for knowledge perception and semantic understanding, the semantic relationship between the geographic entities can be accurately grasped, such as accurately determining the semantic of the text geographic entity through semantic mapping and mining deep semantic through semantic reasoning, so that the accuracy of geographic information extraction is improved.

[0054] 2. The artificial intelligence and geographic information system technology are fused, the intelligent extraction and processing of the geographic information text are realized, the accuracy and efficiency of geographic information extraction are improved, the semantic understanding ability of geographic information is enhanced, flexible and diverse geographic information extraction requirements are supported, the integration and application of geographic information are promoted, and the intelligent level of geographic information processing is improved. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 It is a schematic diagram of the geographic information text intelligent extraction method based on knowledge perception of the application;

[0056] Figure 2 It is a schematic diagram of the geographic information text intelligent extraction system based on knowledge perception of the application. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0058] Embodiment 1

[0059] Please refer to Figure 1 The method for intelligently extracting geographic information text based on knowledge perception comprises the following steps:

[0060] Step 1: Construct a geographic knowledge ontology library, and construct a corresponding semantic relationship network based on the library;

[0061] Step 2: Obtain text geographic entities and text semantic relationships corresponding to the input geographic information text;

[0062] Step 3: Based on the obtained text geographic entities and text semantic relationships, and in combination with the semantic relationship network, construct a corresponding initial tree-shaped tree; based on the geographic knowledge ontology library, perform knowledge perception and semantic understanding on the initial tree-shaped tree to obtain a corresponding semantic tree;

[0063] Step 4: Based on the obtained semantic tree, perform information extraction to obtain the required geographic information;

[0064] It needs to be further explained that in the specific implementation process, the process of constructing the geographic knowledge ontology library comprises:

[0065] Based on the pre-constructed corpus, obtain existing geographic entity concepts, and generalize and abstract the corresponding geographic entity concepts into corresponding basic geographic entities, wherein the basic geographic entities comprise natural geographic entities, artificial geographic entities and combined geographic entities; wherein the basic geographic entities are composed of several different geographic entity categories; the natural geographic entity refers to a geographic phenomenon or object existing in nature, such as mountains, rivers, etc.; the artificial geographic entity refers to a building or other structure produced by human activities, such as roads, bridges, etc.; the combined geographic entity usually refers to a region division with certain function or significance, for example, administrative division, nature reserve, etc.

[0066] Setting up data collection nodes, which collect geographic data related to corresponding basic geographic entities based on different data sources, to obtain corresponding basic geographic data; the data sources include social media, academic research papers and works, geographic data institutions, etc.; the basic geographic data includes vector layers provided by map services, remote sensing image data, and statistical data and related geographic academic papers publicly released by the government, etc.;

[0067] Performing data cleaning on the obtained basic geographic data, which includes removing special characters, numbers, HTML tags, and other non-text characters in the text, as well as processing abbreviations, acronyms, and non-standard writing methods in the text;

[0068] Further, performing data preprocessing on the basic geographic data after data cleaning to obtain corresponding basic text data, which includes steps such as word segmentation, stop word removal, and part-of-speech restoration;

[0069] After preprocessing is completed, the subject words and object words of the geographic entities in the corresponding basic text data are identified based on a pre-constructed geographic entity recognition model, and the semantic relationship between the subject words and object words of the corresponding geographic entities is identified to obtain the corresponding geographic entities and the semantic relationship between the geographic entities; the semantic relationship includes topological relationship, subordinate relationship, and time relationship; the topological relationship includes spatial relationships such as adjacency, association, inclusion, and connectivity between geographic entities; the subordinate relationship refers to the logical composition relationship between geographic entities; the time relationship includes the generation time, change time, and extinction time of geographic entities;

[0070] Further, based on the geographic encyclopedia, the attribute information of the corresponding geographic entities is obtained, and based on the attribute information, the attribute extraction of the corresponding basic text data is performed, and the attribute information of the corresponding geographic entities is supplemented and updated to obtain the corresponding entity attributes; the entity attributes include basic attributes and special attributes; the basic attributes refer to common data attributes, including classification code, entity code, name, other name, geometric form, etc.; the special attributes refer to specific attributes; taking a river as an example, the special attributes include attributes such as starting point, ending point, length, river level, and river function;

[0071] Based on the obtained geographic entities, semantic relationships, and entity attributes, corresponding knowledge triples are constructed and stored under the corresponding geographic entity category, and based on them, a corresponding geographic knowledge ontology library is constructed;

[0072] It should be further explained that in the specific implementation process, the construction process of the geographic entity recognition model includes:

[0073] The backbone network of the geographic entity recognition model is a multi-level neural network, and the basic framework of the multi-level neural network is an input layer, a double-layer Subject-object extraction layer, fully connected layer, and output layer;

[0074] The input layer is used to encode the input text statement to obtain the encoding vector corresponding to each word in the corresponding text statement;

[0075] The double layer Including the first Layer and second Layer; the first The layer extracts semantic features from the input encoding vector by introducing a control gate mechanism to obtain the corresponding word semantic feature vectors, and assigns weights to each word in the corresponding text sentence based on the obtained word semantic feature vectors to obtain the corresponding weighted sentence; wherein, the control gate mechanism includes a forget gate, an enhancement gate and an output gate;

[0076] The formula for obtaining the word semantic feature vector is as follows: In the formula, and These represent the first and second characters in the corresponding text statements. The semantic feature vector of each word; Indicates the first text statement The encoding vector corresponding to each word; and These represent the weight matrix and bias term of the output gate, respectively; the output gate is used to control the corresponding first... The output content of the layer; This indicates vector concatenation; This represents a pre-selected activation function;

[0077] Indicates the current time Next First The network state data corresponding to each layer is used to support the perception of long-distance context by word semantic features, serving the localization of geographic entities (subject / object words) and relationship identification; among them, In the formula, and Each represents the weight matrix of the enhancement gate; and Both represent bias terms; the enhancement gate is used to adjust the corresponding first... Update the network status data of the layer; This represents the operation of the hyperbolic tangent function;

[0078] Indicates the first text statement The forget gate output for each word In the formula, and These represent the weight matrix and bias term of the forget gate, respectively;

[0079] The formula for assigning weights is: ; Indicates the first text statement The word weights corresponding to each word; , , These represent the semantic feature vectors of the corresponding words. The weight matrix obtained by performing a linear transformation; Represents matrix permutation; Represents the weight matrix Matrix dimensions;

[0080] The second BiLSTM layer is used to extract the semantic feature vectors of each word in the received weighted sentence and fuse them to obtain the corresponding sentence semantic feature vectors.

[0081] The subject-object extraction layer receives the word semantic feature vector output by the second BiLSTM layer and obtains the subject word position of the corresponding geographic entity within the text sentence based on a pre-set prediction classifier. The formula for obtaining the subject word position is as follows: In the formula, This represents the probability of the subject word of a geographic entity being located at position S1 within the corresponding text sentence x, where S1 represents the position identifier of the subject word of the geographic entity, i.e., the range from the beginning to the end of the subject word. This represents the probability that the i-th word is the starting position of the main word of a geographic entity; where, In the formula, Indicates the training weights; Indicates deviation; This represents the semantic feature vector of the i-th word in the weighted sentence. Indicates the total number of words in a text sentence; Represents an exponential variable and ;in, Based on the corresponding probability Decision, if appropriate If the probability exceeds a pre-set threshold, then Conversely, ; Represents the trainable parameters of the model; and These represent the start and end positions of the main word, respectively; n represents the index of the main word between its start and end positions. Indicates the total number of words in a text sentence; Indicates product operation;

[0082] Based on the location of the subject word of the geographic entity, the word representation vector corresponding to the subject word is obtained, wherein the word representation vector is composed of at least one word semantic feature vector;

[0083] Furthermore, based on the process of obtaining the subject word position and corresponding word representation vector of the geographic entity, the object word position and corresponding word representation vector of the geographic entity in the corresponding text statement are obtained, and the subject word and object word of the corresponding geographic entity are marked in the corresponding text statement.

[0084] The fully connected layer is used to receive the obtained word representation vectors and sentence semantic vectors, and merge them to obtain the corresponding entity vector ST;

[0085] The output layer is used to map the obtained entity vectors to the category probabilities of a certain semantic category. ;

[0086] In the formula, This indicates that the entity vector ST belongs to the first semantic relation. The category probability of each semantic category. and These represent exponentiation and activation function operations, respectively. This represents all semantic categories of a predefined semantic relationship;

[0087] The obtained category probabilities are compared with the preset category threshold. If the category probability is greater than the category threshold, the corresponding semantic category is used as the semantic relationship between the corresponding geographic entities. If the category probability is not greater than the category threshold, no other operation is performed.

[0088] Define the loss function of the geographic entity recognition model. In the formula, and Indicates the weight ratio. The location probability of object words, which are represented as geographic entities, is obtained through a process similar to that of subject word location acquisition, and will not be elaborated upon in this application. Represents the first in the training sample set training samples, Indicates the first The semantic category labels of each training sample; Represents the first in the training sample set The potential triples corresponding to each training sample ;

[0089] A plurality of sets of geographical text sentences with artificial annotations are acquired, and a corresponding training sample set is constructed based thereon; AdaGrad is defined as an optimizer to continuously optimize the parameters of the geographical entity recognition model during the training process, the training sample set is input into the geographical entity recognition model in batches, the value of the corresponding loss function is recorded, and when the value of the loss function of L2 consecutive batches no longer decreases or changes, the parameters of the geographical entity recognition model at this time are saved, that is, the training of the geographical entity recognition model is completed, L2 being a constant;

[0090] It needs to be further explained that, in the specific implementation process, the construction process of the semantic relation network includes:

[0091] The geographical entities stored in the corresponding geographical knowledge ontology library are acquired, and entity disambiguation is performed thereon, and the entity disambiguated geographical entities are marked as concept entities, the entity disambiguation refers to distinguishing and marking the geographical entities with the same name but different addresses, and merging the repeated knowledge triples corresponding to the geographical entities with different names but the same address;

[0092] Further, different concept entities and their corresponding semantic relations are acquired; and the corresponding concept entities are converted into corresponding graph nodes, the semantic relations are converted into corresponding directed edges, and the corresponding directions and relation types are marked, the relation types include membership relation, proximity relation, connectivity relation, etc.; for example: if city A belongs to province B, there is a directed edge from node "city A" to node "province B" in the graph, and the relation type of the edge is marked as "belongs to";

[0093] Further, based on the pre-selected directed graph construction algorithm and based on the graph nodes constructed by the corresponding concept entities and the edges corresponding to the semantic relations, a semantic relation network between the corresponding concept entities is constructed; the directed graph construction algorithm is used to organize and connect the constructed graph nodes and edges; to ensure the accuracy and integrity of the relations, such as correctly reflecting the inclusion hierarchy relationship between different geographical regions and the interaction relationship between geographical elements in the graph.

[0094] It needs to be further explained that, in the specific implementation process, the process of acquiring the text geographical entities and text semantic relations corresponding to the input geographical information text includes:

[0095] The input geographical information text is preprocessed, and is input into the geographical entity recognition model, the geographical entities and semantic relations in the corresponding geographical information text are obtained, and are marked as text geographical entities and text semantic relations, respectively.

[0096] It needs to be further explained that in the specific implementation process, based on the obtained text geographic entity and text semantic relationship, and combined with the semantic relationship network, the corresponding initial tree-shaped tree is constructed; based on the geographic knowledge ontology library, the corresponding initial tree-shaped tree is subjected to knowledge perception and semantic understanding, and the process of obtaining the corresponding semantic tree includes:

[0097] The extracted text geographic entity and text semantic relationship are obtained and mapped into the constructed semantic relationship network; and the graph nodes and edges corresponding to the corresponding text geographic entity and text semantic relationship are obtained;

[0098] The graph nodes corresponding to the respective text geographic entity are taken as the clustering center, and the clustering operation is performed on the other graph nodes in the corresponding semantic relationship network, to obtain the graph nodes in the same clustering set with the corresponding text geographic entity, obtain the concept entity corresponding to the corresponding graph node and the semantic relationship of the corresponding edge, and mark them as expanded geographic entity and expanded semantic relationship;

[0099] The process of clustering operation includes:

[0100] The graph node corresponding to the text geographic entity in the corresponding semantic relationship network is obtained and set as the clustering center;

[0101] Further, based on the pre-set clustering radius, the other graph nodes connected with the clustering center in the semantic relationship network are traversed, and the clustering radius refers to the maximum hop limit between the clustering center node in the semantic relationship network, which is usually set to 2-3 hops, that is, considering other graph nodes within 2-3 edges from the clustering center;

[0102] Further, based on the traversal result, the knowledge triple corresponding to each graph node within the clustering radius is obtained, and the semantic similarity between the corresponding graph node and the clustering center is obtained based on it, and the graph nodes with a semantic similarity greater than a pre-set semantic similarity threshold are divided into the clustering set in which the clustering center is located; wherein the clustering set retains the node attributes and edge types in the original semantic relationship network, and a new clustering identification attribute is added, which is used to mark the clustering categories to which different nodes belong; for example: the clustering set formed by taking the Yangtze River as the clustering center contains the graph nodes corresponding to the geographic entities related to the Yangtze River (such as “Three Gorges of the Yangtze River”, “Yangtze River Basin”, “Yangtze River Estuary” and other associated geographic entities), and the semantic relationships between them such as “flowing through”, “converging”, “located in”; the semantic similarity threshold is usually set to a value between 0.6-0.8, which can be adjusted according to the application scene requirement;

[0103] The mathematical calculation formula of the semantic similarity is: ; in the formula, the semantic similarity between the clustering center zx and the qth graph node within the clustering radius. It represents the normalized reciprocal of the shortest path distance between the cluster center zx and the q-th graph node within the cluster radius; This represents the similarity of node attributes, obtained by comparing the overlap between the attributes of the corresponding entities of the nodes; ;and and Indicates the weighting coefficient;

[0104] Furthermore, based on the textual semantic relationships and extended semantic relationships, upstream and downstream relationships between different textual geographic entities, between different extended geographic entities, and between textual geographic entities and extended geographic entities are obtained; for example, the "inclusion" relationship can be reflected as an upstream node containing a downstream node (such as a province containing a city), and the "flow through" relationship can be reflected as a connection at the same level between a node representing a river and a node flowing through a city or region, to indicate their relationship in terms of geographical location and mutual connection, thereby constructing a complete tree structure;

[0105] Furthermore, all obtained text semantic relationships and extended semantic relationships are acquired and structurally classified to obtain the upstream and downstream relationships or association relationships between different text geographic entities and extended geographic entities. For example, hierarchical relationships (such as "contains", "belongs to", "composed of" etc.) are mapped to specific upstream and downstream relationships, while interactive relationships (such as "flows through", "adjacent", "connected" etc.) are mapped to association relationships with specific tags.

[0106] A relationship determination matrix is ​​constructed based on the obtained upstream and downstream relationships and association relationships. The relationship determination matrix is ​​used to store and clarify the upstream and downstream relationships or association relationships between different text geographic entities and extended geographic entities. The matrix element values ​​of the relationship determination matrix are used to represent the relationship type and direction between entities. Upstream and downstream relationships are recorded with directional values ​​(e.g., +1 indicates from top to bottom, -1 indicates from bottom to top), while relationships at the same level are recorded using specific identifiers (e.g., 2) and additional semantic type tags.

[0107] Furthermore, an enhanced tree structure framework is constructed, which includes two parts: a trunk hierarchical structure and horizontal semantic links. The trunk hierarchical structure strictly follows the hierarchical characteristics of a tree and is constructed based on upstream and downstream relationships. Horizontal semantic links are stored in the relevant nodes in the form of metadata, which does not affect the basic structure of the tree, but is considered simultaneously during semantic processing. For example, the "flow through" relationship is represented in the data structure as follows: the river node and the city node each maintain their original positions in the tree, and at the same time, horizontal semantic links are added to the node attributes to record their association and the semantic type "flow through".

[0108] selecting a text geographic entity with the highest hierarchical relationship in the semantic relationship network as a root node of a main hierarchical structure, and adding other text geographic entities or expanded geographic entities to the main hierarchical structure layer by layer based on upstream and downstream relationships in the relationship determination matrix, wherein it is ensured that each geographic entity appears only once in the tree during the adding process;

[0109] After the adding is completed, interactive relationships (such as “flows through”, “adjacent” and the like) are stored in the form of attributes of tree nodes in the enhanced tree structure framework based on the transverse semantic link, and an initial tree-shaped tree is obtained, wherein each attribute contains relationship type and direction information; for example, the attribute of the node “Yangtze River” contains [type: “flows through”, target: “Wuhan City”, direction: “one-way”], and the node “Wuhan City” also contains a corresponding attribute pointing to the “Yangtze River”;

[0110] It should be noted that one embodiment of the present application further comprises: when there are multiple semantic relationships between two geographic entities, the most representative relationship is selected as the basis for constructing the tree structure according to a preset semantic relationship priority order; for example, if a city is both “located” in a mountain range (geographical inclusion relationship) and “belongs to” a province (administrative division relationship), the “belongs to” relationship is preferred to determine its position in the tree; the semantic relationship priority order is in descending order of administrative division relationship, geographical inclusion relationship, spatial positioning relationship and functional interaction relationship;

[0111] Further, the obtained initial tree-shaped tree is input into the constructed geographic knowledge ontology library for semantic mapping and semantic reasoning to obtain a corresponding semantic tree;

[0112] The semantic mapping refers to mapping the node entities corresponding to the tree nodes in the initial tree-shaped tree and the edge semantic relationships corresponding to the edges between the tree nodes to the geographic entities and semantic relationships in the geographic knowledge ontology library based on a preselected tree algorithm; for example, a text geographic entity extracted is mapped to the geographic entity category of “river”, and the relationship (such as flowing through a city) with the surrounding geographic entities can also be mapped to the semantic relationship in the geographic knowledge ontology library, so as to determine the accurate semantics;

[0113] The semantic reasoning refers to reasoning adjacent tree nodes on the semantic level according to the entity attributes in the knowledge triple corresponding to the mapped geographic entities; obtaining a corresponding semantic reasoning result; for example, it is known that a lake is upstream of a river, and in the knowledge system constructed by the geographic knowledge ontology library, it can be inferred that the water resources in the area where the lake is located may have an impact on the water quantity, ecology and other aspects of the downstream of the river, and other deeper semantic understandings, thereby expanding the grasp of the meanings implied by the geographic entities and relationships;

[0114] After the semantic mapping and semantic reasoning are completed, the entity attributes stored in the local library of the corresponding address knowledge and the corresponding semantic reasoning results are stored in the tree nodes in the corresponding semantic tree.

[0115] It needs to be further explained that, in the specific implementation process, the process of obtaining the geographic information required based on the obtained semantic tree includes:

[0116] The root node of the corresponding semantic tree is taken as the starting node, and the corresponding semantic tree is traversed based on the preset extraction rule, and the tree nodes and the edges corresponding to the tree nodes in the corresponding semantic tree that meet the preset extraction rule are marked, and then the entity attributes and the semantic reasoning results stored in the corresponding tree nodes and the edge semantic relationship of the tree nodes to the edges are read to obtain the corresponding geographic information; wherein the preset extraction rule refers to a condition or standard preset for selectively extracting specific geographic information from the semantic tree; for example: the preset extraction rule can be to extract all lakes in a river basin, in the traversal process, when a river node and a lake node having a specific semantic relationship (such as a "contains lake" relationship, which can be defined when the semantic tree is constructed) are encountered, if the limitation on this relationship in the rule is met, the geographic information (such as lake name, area, and region) of the lake node is extracted; wherein the river node and the lake node are tree nodes corresponding to the corresponding geographic entities; the river and the lake are both one of the corresponding geographic entity categories; wherein the predicted extraction rule is determined according to actual needs.

[0117] Further, the obtained geographic information is de-duplicated and summarized to obtain corresponding integrated information, and whether the semantic relationship between the geographic entities involved in the extracted and integrated information is clear and reasonable is examined, if not, the corresponding integrated information is sorted and adjusted according to the geographic knowledge local library; after the adjustment is completed, the corresponding integrated information is converted into a required format file and is visually outputted;

[0118] The present application builds a geographic knowledge ontology library and a semantic relationship network, and uses a multi-level neural network to build a geographic entity recognition model, aiming to realize efficient and accurate extraction and integration of geographic information, and to provide more powerful technical support for geographic information processing.

[0119] Embodiment 2

[0120] Please refer to Figure 2 The present embodiment is not described in detail, and the description of part of the present embodiment is seen in the description of embodiment 1, and a kind of geographic information text intelligent extraction system based on knowledge perception is provided;Including:

[0121] Data construction module, for constructing geographic knowledge ontology library, and based on its corresponding semantic relationship network is constructed;

[0122] The feature recognition module is configured to acquire a text geographical entity and a text semantic relationship corresponding to the input geographical information text;

[0123] The information expansion module is configured to construct an initial tree based on the acquired text geographical entity and text semantic relationship and in combination with a semantic relationship network; and perform knowledge perception and semantic understanding on the initial tree based on the geographical knowledge ontology library to acquire a corresponding semantic tree.

[0124] The information extraction module is configured to perform information extraction based on the acquired semantic tree to acquire required geographical information.

[0125] The various modules are connected through wired and / or wireless means to realize data transmission between the modules.

[0126] Embodiment 3

[0127] The embodiment provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the operation mode of the above-provided geographical information text intelligent extraction method based on knowledge perception when executing the computer program.

[0128] Since the electronic device introduced in the embodiment is the electronic device used to implement the geographical information text intelligent extraction method based on knowledge perception in the embodiment, the specific implementation mode of the electronic device in the embodiment and various changes thereof can be understood by those skilled in the art based on the geographical information text intelligent extraction method based on knowledge perception introduced in the embodiment, and therefore, how the electronic device implements the method in the embodiment will not be introduced in detail. As long as the electronic device used to implement the geographical information text intelligent extraction method based on knowledge perception in the embodiment is implemented by those skilled in the art, it belongs to the scope to be protected by the present application.

[0129] The above formulas are dimensionless values calculated, the formula is obtained by collecting a large amount of data to simulate a formula of the nearest real situation, and the preset parameters and threshold values in the formula are set by those skilled in the art according to the actual situation.

[0130] The above description is only the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above embodiments. Any technical solution falling within the idea of the present application belongs to the protection scope of the present application. It should be noted that, for ordinary technical users in the technical field, some improvements and decorations without departing from the principle of the present application are also considered to be within the protection scope of the present application.

Claims

1. A knowledge-aware based geographic information text intelligent extraction method, characterized in that, The application comprises the following steps: Step 1: constructing a geographic knowledge ontology library and constructing a corresponding semantic relationship network based on the same; Step 2: obtaining text geographic entities and text semantic relationships corresponding to input geographic information text; Step 3: constructing a corresponding initial tree based on the obtained text geographic entities and text semantic relationships and combining the same with the semantic relationship network; carrying out knowledge perception and semantic understanding on the corresponding initial tree based on the geographic knowledge ontology library to obtain a corresponding semantic tree; Step 4: carrying out information extraction based on the obtained semantic tree to obtain required geographic information; The process of constructing the geographic knowledge ontology library comprises the following steps: obtaining existing geographic entity concepts and generalizing and abstracting the same into basic geographic entities, wherein the basic geographic entities comprise natural geographic entities, artificial geographic entities and combined geographic entities, and the basic geographic entities are composed of several different geographic entity categories; setting data collection nodes, wherein the data collection nodes collect geographic data related to corresponding basic geographic entities based on different data sources to obtain corresponding basic geographic data; carrying out data preprocessing on the obtained basic geographic data to obtain corresponding basic text data, wherein the data preprocessing comprises word segmentation, stop word removal and word type restoration; after the preprocessing is completed, identifying geographic entities in the corresponding basic text data based on a pre-constructed geographic entity identification model and identifying semantic relationships between the corresponding geographic entities; simultaneously, extracting attribute information of the corresponding geographic entities based on the basic text data to obtain corresponding entity attributes; constructing corresponding knowledge triples based on the obtained geographic entities, semantic relationships and entity attributes and storing the same under corresponding geographic entity categories and constructing a corresponding geographic knowledge ontology library based on the same. 2.The knowledge perception based geo-information text intelligent extraction method according to claim 1, characterized in that, The process of constructing the geographic entity identification model comprises the following steps: The backbone network of the geographic entity recognition model is a multi-level neural network, and a basic framework of the multi-level neural network is an input layer, a double-layer , a host-guest extraction layer, a fully connected layer, and an output layer. the input layer is used for encoding processing of input text sentences to obtain encoding vectors corresponding to each word in the corresponding text sentences; The double layer comprises a first layer and a second layer; the first layer performs semantic feature extraction on an input encoding vector by introducing a control gate mechanism, obtains a corresponding word semantic feature vector, and assigns weights to each word in a corresponding text sentence based on the obtained word semantic feature vector to obtain a corresponding weighted sentence; wherein the control gate mechanism comprises a forget gate, an enhance gate and an output gate; The second The layer is used for extracting word semantic feature vectors corresponding to each word in the received weighted sentence, and performing feature fusion to obtain a corresponding sentence semantic feature vector; The host-guest extraction layer is configured to receive a second layer output word semantic feature vector, and obtain a host word position of a geographic entity within a corresponding text sentence based on a pre-set prediction classifier; and obtain a word representation vector corresponding to the corresponding host word based on the host word position of the geographic entity. based on the process of obtaining the subject word position of the geographic entity and the corresponding word representation vector, the object word position of the geographic entity in the corresponding text sentence and the corresponding word representation vector are obtained, and the subject word and the object word of the corresponding geographic entity in the corresponding text sentence are marked; the full connection layer is used for receiving the obtained word representation vector and sentence semantic feature vector and combining the same to obtain a corresponding entity vector; the output layer is used for mapping the obtained entity vector to a class probability of a certain semantic category; the obtained class probability is compared with a corresponding class threshold value, if the class probability is greater than the class threshold value, the corresponding semantic category is taken as the semantic relationship between the corresponding physical entities, if the class probability is not greater than the class threshold value, no other operation is performed; A plurality of geographical text sentence sets with artificial annotations are acquired, and a corresponding training sample set is constructed based on the geographical text sentence sets; a geographical entity recognition model is defined As an optimizer, the parameters of the geographical entity recognition model are continuously optimized during the training process. The training sample set is input into the geographical entity recognition model in batches, and the value of the corresponding loss function is recorded. When the values of the loss function of the continuous L2 batches no longer decrease or change, the parameters of the geographical entity recognition model at this time are saved, that is, the training of the geographical entity recognition model is completed, and L2 is a constant. 3.The knowledge perception based geo-information text intelligent extraction method according to claim 2, characterized in that, the formula for obtaining the word semantic feature vector is: wherein, and denote the word semantic feature vector corresponding to the th and th word in the corresponding text sentence, respectively; denotes the weight matrix of the output gate in the control gate mechanism; denotes the encoding vector corresponding to the th word in the text sentence; denotes the bias term of the output gate; denotes vector concatenation; denotes a pre-selected activation function; represents the current time the first layer corresponds to network state data, ; in which, and both represent the weight matrix of the enhanced gate; and both represent the bias term; the enhanced gate is used to update the network state data of the corresponding first layer; represents the hyperbolic tangent function operation; represents the forget gate output corresponding to the i-th word within the text sentence, ; where, and represent the weight matrix and bias term of the forget gate, respectively;​ The formula for assigning weights is: ; Indicates the first text statement The word weights corresponding to each word; , , These represent the semantic feature vectors of the corresponding words. The weight matrix obtained by performing a linear transformation; Represents matrix permutation; Represents the weight matrix The matrix dimension. 4.The knowledge perception based geo-information text intelligent extraction method according to claim 2, characterized in that, The formula for obtaining the position of the main word is: In the formula, The main words representing geographical entities in the corresponding text statements The probability at location S1 within the geographic entity, where S1 represents the location identifier of the main word of the geographic entity, i.e. the range from the beginning to the end of the main word; Representing the corresponding number The probability of a word being the starting position of the main word of a geographic entity; where, In the formula, Indicates the training weights; Indicates the bias term; Indicates the first weighted statement The semantic feature vector of each word Indicates the total number of words within a text sentence; ;in, Based on the corresponding probability Decision, if appropriate If the probability exceeds a pre-set threshold, then Conversely, ; Represents the trainable parameters of the model; and These represent the start and end positions of the main word, respectively; n represents the index of the main word between its start and end positions. Indicates the total number of words in a text sentence; Indicates product operation; category probability ; where, denotes an entity vector belongs to the i-th semantic category within a semantic relation, category probability, and softmax() respectively denote an exponential operation and an activation function operation; denotes all semantic categories of a predefined semantic relation; Define the loss function of the geographic entity recognition model. In the formula, and Indicates the weight ratio. The location probability of object words representing geographical entities; Represents the first in the training sample set training samples, Indicates the first The semantic category labels of each training sample; Represents the first in the training sample set The potential triples corresponding to each training sample . 5.The knowledge perception based geo-information text intelligent extraction method according to claim 1, characterized in that, the process of constructing the semantic relationship network comprises the following steps: The geographic entity stored in the corresponding geographic knowledge base is obtained, entity disambiguation is performed, and the entity disambiguated geographic entity is marked as a concept entity; different concept entities and their corresponding semantic relationships are obtained; and the corresponding concept entities are converted into corresponding graph nodes, the semantic relationships are converted into corresponding directed edges, and the corresponding directions and relationship types are marked; Based on the preselected directed graph construction algorithm and the graph nodes constructed by the corresponding concept entities and the edges corresponding to the semantic relationships, a semantic relationship network between the corresponding concept entities is constructed. 6.The knowledge perception based geo-information text intelligent extraction method according to claim 5, characterized in that, The process of obtaining the text geographic entity and the text semantic relationship corresponding to the input geographic information text includes: The input geographic information text is preprocessed, and the preprocessed geographic information text is input into the geographic entity recognition model to obtain the geographic entity and the semantic relationship in the geographic information text, which are marked as a text geographic entity and a text semantic relationship, respectively. 7.The knowledge perception based geo-information text intelligent extraction method according to claim 6, characterized in that, Based on the obtained text geographic entity and text semantic relationship, and in combination with the semantic relationship network, a corresponding initial tree is constructed, which includes: The extracted text geographic entity and text semantic relationship are mapped into the constructed semantic relationship network, and the graph nodes and edges corresponding to the text geographic entity and the text semantic relationship are obtained; The graph nodes corresponding to the text geographic entity are taken as clustering centers, and clustering is performed on the other graph nodes in the semantic relationship network to obtain the graph nodes in the same clustering set as the text geographic entity, the concept entities corresponding to the graph nodes and the semantic relationships of the corresponding edges are obtained, and the concept entities and the semantic relationships are marked as expanded geographic entities and expanded semantic relationships; All the text semantic relationships and the expanded semantic relationships are obtained, and are classified and structured to obtain the upstream and downstream relationships or the associated relationships between the different text geographic entities and the expanded geographic entities, and a relationship determination matrix is constructed based on the relationships; An enhanced tree structure framework is constructed, which includes a trunk hierarchical structure and a horizontal semantic link; The text geographic entity with the highest upstream and downstream relationship in the semantic relationship network is selected as the root node in the trunk hierarchical structure, and other text geographic entities or expanded geographic entities are added to the corresponding trunk hierarchical structure layer by layer based on the upstream and downstream relationships in the relationship determination matrix; After the addition is completed, the associated relationships are stored in the form of tree node attributes in the enhanced tree structure framework based on the horizontal semantic link, and an initial tree is obtained; The obtained initial tree is input into the constructed geographic knowledge base for semantic mapping and semantic reasoning, and a semantic tree is obtained based on the initial tree. After the semantic mapping and the semantic reasoning are completed, the entity attributes stored in the address knowledge base and the semantic reasoning results are stored in the tree nodes in the semantic tree. 8.The knowledge-aware based geographic information text intelligent extraction method according to claim 7, characterized in that, The semantic mapping refers to mapping the node entity corresponding to the tree node in the initial tree and the edge semantic relationship corresponding to the edge between the tree nodes into the geographic knowledge triplets in the geographic knowledge base. The node entity refers to a text geographic entity or an expanded geographic entity. The edge semantic relationship refers to a text semantic relationship or an extended semantic relationship; and the semantic reasoning refers to reasoning of other tree nodes adjacent to the corresponding tree node in a semantic level according to the entity attributes in the knowledge triple mapped by the corresponding tree node. 9.The knowledge perception based geo-information text intelligent extraction method according to claim 7, characterized in that, The process of information extraction based on the obtained semantic tree to obtain the required geographic information includes: Taking the root node of the corresponding semantic tree as a starting node, traversing the corresponding semantic tree based on a preset extraction rule, marking the tree nodes and edges corresponding to the tree nodes in the corresponding semantic tree that meet the preset extraction rule, reading the entity attributes and semantic reasoning results stored in the corresponding tree nodes and the edge semantic relationship of the tree nodes to the edges, obtaining the corresponding geographic information, removing and summarizing the obtained geographic information to obtain corresponding integrated information, and checking whether the semantic relationship between the geographic entities involved in the integrated information meets the requirements, if not, combing and adjusting the corresponding integrated information according to the geographic knowledge local library, and converting the corresponding integrated information into a required format file and outputting the same in a visualized manner.

Images