A method, device and server for automatic question and answer of land survey data

Abstract

The present invention relates to the field of big data technology, and specifically to a method, device and server for automatic question and answer of land survey data. A method for automatic question and answer of land survey data comprises obtaining a question sentence; obtaining an entity set, wherein the entity set is obtained by semantically parsing the question sentence according to a preset semantic parsing rule, and the entity set includes location, time and land type; obtaining an answer, wherein the answer is obtained according to the entity set; and outputting the answer. The present application automatically parses according to the user's demand for obtaining land survey information, without the need for manual query summary, and the information obtained is more rigorous and accurate.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and specifically to a method, device and server for automatic question and answer of land survey data. Background Art

[0002] At present, the ownership, land type data, planting attributes, arable land quality and other information of land surveys require users to conduct statistical analysis according to their own needs. The statistical granularity of many systems is limited and cannot be adjusted in real time according to user needs. It is time-consuming and laborious for users to obtain information, and the timeliness of the information is poor. Summary of the invention

[0003] The purpose of the present invention is to propose a method, device and server for automatic question and answer of land survey data, which has the advantages of automatic analysis according to the user's demand for obtaining land survey information, without the need for manual query summary, and has the advantages of strong timeliness, high rigor and high accuracy in obtaining information.

[0004] A method for automatic question-answering of land survey data comprises: obtaining a question sentence; obtaining an entity set, wherein the entity set is obtained by semantically parsing the question sentence according to a preset semantic parsing rule, and the entity set includes a location, a time, and a land type; obtaining a first land survey keyword, wherein the first land survey keyword is obtained according to a land survey keyword process, and the first land survey keyword is used to indicate a land survey order; obtaining an answer, wherein the answer is obtained according to the entity set and the first land survey keyword; and outputting the answer.

[0005] In one embodiment of the present invention, after obtaining the question sentence, the method further includes obtaining a preprocessed question sentence, wherein the preprocessed question sentence includes: the preprocessed question sentence is obtained by removing emoticons from the question sentence, the traditional Chinese characters in the question sentence are converted into simplified Chinese characters, the full-width input format in the question sentence is converted into a diagonal input format, and homophonic and similar input errors in the question sentence are corrected.

[0006] In one embodiment of the present invention, an entity set is obtained, and the entity set is obtained by semantically parsing the question according to preset semantic parsing rules, and the entity set includes location, time and place type, including: obtaining location and time, and the location and the time are both obtained based on a BERT-BILSTM-CRF model; obtaining place type, and the place type is obtained according to a string matching method.

[0007] In one embodiment of the present invention, a location and a time are obtained, and the location and the time are obtained based on a BERT-BILSTM-CRF model and preset rules, including: the location contains several levels of administrative divisions, the several levels of administrative divisions are divided according to a word segmentation method of a dictionary, and the dictionary consists of administrative divisions and aliases of the administrative divisions.

[0008] In one embodiment of the present invention, after obtaining the position and time, wherein the position and the time are obtained based on the BERT-BILSTM-CRF model and preset rules, it also includes: outputting the first position and the second position when the obtained position result is not unique.

[0009] In one embodiment of the present invention, a first land survey keyword is obtained, the first land survey keyword is obtained according to a land survey keyword process, and the first land survey keyword is used to indicate a land survey order, including: when the question sentence successfully matches the preset second land survey keyword string, the first land survey keyword is obtained, and the preset second land survey keyword contains words or phrases used to indicate the survey order; when the question sentence fails to match the preset second land survey keyword string, and the land class time attribute is unique, the first land survey keyword is obtained according to the unique time attribute of the land class; when the question sentence fails to match the preset second land survey keyword string, the land class time attribute is not unique, and the time is different from the preset land survey time attribute. When there is a corresponding relationship, the first land survey keyword is obtained according to the time; when the question sentence fails to match the preset second land survey keyword character string, the land class time attribute is not unique, and there is no corresponding relationship between the time and the preset land survey time, the first land survey keyword is obtained according to the latest time attribute in the land class to correct the time; when the time cannot be obtained based on the BERT-BILSTM-CRF model and the question sentence contains the first land survey keyword, the time is the time corresponding to the first land survey keyword; when the time cannot be obtained based on the BERT-BILSTM-CRF model and the question sentence does not contain the first land survey keyword, the time is the time input for the question sentence.

[0010] In one embodiment of the present invention, after obtaining the first land survey keyword, the question classification is also included, and the question classification includes: obtaining a first type of question, the entity set of the first type of question includes location and land type; obtaining a second type of question, the entity set of the second type of question includes one of location or land type; obtaining a third type of question, the entity set of the third type of question does not include location and land type.

[0011] In one embodiment of the present invention, after obtaining the first category of questions, it also includes classification of the first category of questions, the first category of questions includes the first question, the second question and the third question; the first question is a single-category area problem, the second question is an area sum problem, and the third question is an area proportion problem: a first similarity set is obtained, and the elements in the first similarity set are obtained by calculating the cosine similarity of the first question vector set; the first question vector set is obtained by inputting the question and multiple preset first typical questions into the Sentence-Bert model; the preset first typical question is a typical question of the first question; a second similarity set is obtained, and the elements in the second similarity set are obtained by calculating the cosine similarity of the second question vector set; the second question vector set is obtained by inputting the question and multiple preset second typical questions into the Sentence-Bert model The method is as follows: inputting the Sentence-Bert model; the preset second typical question is the typical question of the second question; obtaining a third similarity set, wherein the elements in the third similarity set are obtained by calculating the cosine similarity of the third question vector set; the third question vector set is obtained by inputting the question and multiple preset third typical questions into the Sentence-Bert model; the preset third typical question is the typical question of the third question; obtaining the maximum value of the elements in the first similarity set, the second similarity set and the third similarity set; obtaining the first question, when the element in the first similarity set is the maximum value; obtaining the second question, when the element in the second similarity set is the maximum value; obtaining the third question, when the element in the third similarity set is the maximum value.

[0012] In one embodiment of the present invention, after obtaining the second category of questions, it also includes classification of the second category of questions, the second category of questions includes fourth questions and fifth questions, the fourth question is an inclusion relationship question, and the fifth question is a distribution question: obtaining a first probability, the first probability is obtained by the question through a naive Bayes classification algorithm; obtaining a second probability, the second probability is obtained by the question through a naive Bayes classification algorithm; obtaining the fourth question when the first probability is not less than the second probability; obtaining the fifth question when the first probability is less than the second probability.

[0013] In one embodiment of the present invention, an answer is obtained, and the answer is obtained according to the entity set and the first land survey keyword, including: obtaining a unique subgraph, and the unique subgraph is retrieved according to the entity set and the first land survey keyword; the subgraph is obtained by querying the graph database to obtain the nodes and relationships corresponding to the graph database; the unique subgraph is determined by the query language; obtaining the land area information of the location when the question is the first question; obtaining the sum of the land area of ​​the location when the question is the second question; obtaining the subordinate land area of ​​the location Obtain the ratio of the sum of the areas of the land classes at the location to the sum of the areas of the superior land classes at the location, when the question is the third question and the land classes are two-level land classes with an inclusion relationship; obtain the ratio of the sum of the areas of the land classes at the location to the sum of the areas of the location, when the question is the third question and the land classes do not satisfy the two-level land classes with an inclusion relationship; obtain the superior-subordinate inclusion relationship of the location, when the question is the fourth question; obtain the land class to which the map of the location belongs or the map distribution area of ​​the land class, when the question is the fifth question; obtain the answer is empty, when the question is the sixth question.

[0014] A land survey data automatic question-answering device comprises: an input module for obtaining a question sentence; a first obtaining module for obtaining an entity set, wherein the entity set is obtained by semantically parsing the question sentence according to a preset semantic parsing rule, and the entity set includes a location, a time and a land type; a second obtaining module for obtaining a first land survey keyword, wherein the first land survey keyword is obtained according to a land survey keyword process, and the first land survey keyword is used to indicate a survey order; a third obtaining module for obtaining an answer, wherein the answer is obtained according to the entity set; and an output module for outputting the answer.

[0015] A land survey data automatic question and answer server, comprising a memory and a processor, wherein the memory stores a computer program, and wherein when the computer program is executed by the processor, the processor executes the land survey data automatic question and answer method as described in any one of claims 1 to 10.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. Through semantic analysis of the question, we can obtain entities and land survey keywords such as location, time and land type, determine the knowledge graph search directory based on the entities and land survey keywords, and get the answer by integrating the search results. It can realize automatic adjustment according to user needs without the need for manual query summary, and the information obtained is highly rigorous, accurate and timely.

[0018] 2. By performing semantic analysis on the question and retrieving the knowledge graph, multiple data can be obtained and displayed at the same time in one query. Compared with the existing technology that can only obtain and display one data in one query, it has richer functions and is easier to use. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is an application scenario diagram of the present invention;

[0020] Figure 2 It is the main flow chart of the present invention;

[0021] Figure 3 Flow chart of question preprocessing of the present invention;

[0022] Figure 4 A flow chart for obtaining an entity set for the present invention;

[0023] Figure 5 A flow chart for question classification of the present invention;

[0024] Figure 6 A flow chart for classifying the first category of problems of the present invention;

[0025] Figure 7 Flowchart for classifying the second type of problems in the present invention

[0026] Figure 8 A flowchart for answer retrieval of the present invention;

[0027] Fig. 9 It is a structural block diagram of the device of the present invention;

[0028] Fig.10 This is a structural block diagram of the server of the present invention;

[0029] Fig.11 This is the structure diagram of the existing BERT-BILSTM-CRF model. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] Land and resources data are stored in the form of a knowledge graph library, including five levels of administrative division entities: province, city, county, township, and village, land parcel entities uniquely identified by land parcel numbers, and land class entities uniquely identified by land class codes. Land class entities include primary, secondary, tertiary, eight categories, three categories, cultivated land refinement, planting attributes, and cultivated land quality, among which all land class entities have area attributes. According to the characteristics of provinces, cities, counties, townships, and villages, their respective alias entity sets are constructed. The aliases of administrative division data collected on the Internet, the historical names obtained from the websites of the Ministry of Civil Affairs and the Bureau of Statistics, and time change information are added to the alias relationship of administrative division data.

[0032] The knowledge graph relationship is set as follows. The ownership relationship between administrative divisions and between plots and village entities has time attributes and name attributes. The time attribute is now or before. If the time is now, the name attribute value is the start time of the relationship. If the time is before, the name attribute value is the time period of the relationship. The plot entity has an ownership relationship with the lowest level of land, and there is an inclusion relationship between land types. Each land type has a time attribute, which corresponds to the first, second, and third survey attribute values. When the network collects ownership change information, the change time is added to the name attribute of the ownership relationship, and the time attribute is modified at the same time.

[0033] At present, the words and phrases used to indicate the survey order include the first survey, the second survey, the third survey, the first national land survey, the second national land survey, and the third national land survey; in the future, they will also include the fourth survey, the fifth survey, the fourth national land survey, and the fifth national land survey.

[0034] The correspondence between time and the words and phrases used to indicate the order of the first three surveys is: 1984-2007 corresponds to the first national land survey; 2007-2017 corresponds to the second national land survey; 2017-X corresponds to the third national land survey; X is the year when the fourth national land survey began; and so on to correspond to land surveys of different orders.

[0035] There is a direct correlation between the plots and the refinement of cultivated land (including cultivated land in river channels, cultivated land in lake areas, cultivated land in forest areas, cultivated land in pastoral areas, cultivated land in sandy areas and cultivated land in rocky desertification), and the refinement of cultivated land belongs to cultivated land.

[0036] There is a direct correlation between the plots and planting attributes (including cultivation, fallow, temporary garden planting, temporary tree planting, temporary pasture planting, temporary ponds, intercropping of forest and grain, ornamental gardening, fast-growing trees, green grassland, and uncultivated).

[0037] like Fig.11As shown in the figure, the BERT-BILSTM-CRF model combines the BILSTM network and the CRF model, that is, a CRF linear layer is added after the hidden layer of the BILSTM network.

[0038] The basic idea of ​​bidirectional LSTM is to apply a forward and backward LSTM network to each training sequence, and these two are connected to the same output layer. Such a network structure can provide the output layer with complete context information for each sequence point, and effectively consider the label information before and after the sequence through CRF.

[0039] The BERT-BILSTM-CRF named entity recognition model is built by using the BERT model as a feature representation layer added to the bidirectional LSTM model.

[0040] like Figure 1 As shown, the self-service terminal 101 is provided with an input module, a display module, a storage module and a processor; the input module is used to input query sentences; the display module is used to display the query answers; the storage module is used to store land and resources data and training models for different periods; the processor is used to analyze and judge the results based on the query sentences.

[0041] like Figure 2 A method for automatic question and answer of land survey data is shown in the main flow chart, comprising steps S101, S102, S103, S104 and S105, wherein:

[0042] Step S101 , obtaining a question sentence, where the question sentence is input by a user of the self-service terminal 101 .

[0043] Step S102, obtaining an entity set, the entity set includes three attributes: location, time and land type, and the entity set is obtained by inputting the question according to preset semantic parsing rules.

[0044] Step S103, obtaining a first land survey keyword, wherein the first land survey keyword is obtained according to a land survey keyword process, and the first land survey keyword is used to indicate a survey order.

[0045] Step S104, obtaining an answer, wherein the answer is obtained based on the entity set and the first land survey keyword.

[0046] Step S105, output the answer.

[0047] like Figure 3 The flowchart of question preprocessing includes step S201, step S202, step S203 and step S204, wherein:

[0048] Step S201, deleting emoticons from the question.

[0049] Step S202: converting the traditional Chinese characters in the question into simplified Chinese characters.

[0050] Step S203: the full-width input format in the question sentence is converted into a diagonal input format.

[0051] Step S204, correcting homophonic and similar input errors in the question.

[0052] For example: if the question is "Henan Province Zhengzhou City Flooded District o(╯□╰)o", then according to step S201, the question is adjusted to "Henan Province Zhengzhou City Flooded District", according to step S202, the question is adjusted to "Henan Province Zhengzhou City Flooded District", and according to step S204, the question is adjusted to "Henan Province Zhengzhou City Jinshui District". By pre-processing the questions, redundant information is deleted and the question format is unified, which is conducive to improving the accuracy of semantic information understanding.

[0053] like Figure 4 The flowchart of obtaining the entity set includes step S301, step S302, step S303, step S304, step S305, step S306, step S307, step S308, step S309 and step S310, wherein:

[0054] Step S301, obtaining a position and a time, wherein the position and the time are obtained according to a BERT-BILSTM-CRF model. Since the BERT-BILSTM-CRF model is a prior art, the implementation method is not specifically described.

[0055] Step S302, determining whether the position is unique.

[0056] Step S303, when the position is not unique, output the first position and the second position, and when the position result obtained is not unique; for the user to select.

[0057] Step S304: when the location is unique, obtain the land type, wherein the land type is obtained according to a string matching method.

[0058] Step S305, determining whether the question sentence successfully matches the preset second land survey keyword character string, wherein the preset second land survey keyword includes words or phrases used to indicate the survey order, and if successful, proceeding to S309.

[0059] Step S306, when the question fails to match the preset second land survey keyword character string, determine whether the land type time attribute is unique. If it is unique, enter S309 to obtain the first land survey keyword based on the unique time attribute of the land type.

[0060] Step S307, when the land type time attribute is not unique, determine whether there is a corresponding relationship between the time and the preset land survey time. If yes, enter S309 to obtain the first land survey keyword based on the time.

[0061] Step S308, when there is no corresponding relationship between the time and the preset land survey time, the latest time attribute in the land category is used.

[0062] Step S309, obtaining the first land survey keyword.

[0063] Step S310, correct the time. When the time cannot be obtained based on the BERT-BILSTM-CRF model and the question contains the first land survey keyword, the time is the time corresponding to the first land survey keyword; when the time cannot be obtained based on the BERT-BILSTM-CRF model and the question does not contain the first land survey keyword, the time is the time when the question is input.

[0064] For example: the question is "Hui District in Henan Province", according to step S301, both "Guancheng Hui District in Zhengzhou City, Henan Province" and "Shunhe Hui District in Kaifeng City, Henan Province" meet the conditions. According to step S302, it is determined that the location information is not unique. According to step S303, these two location information are output and the user selects one of them. When the user selects the location to be queried and enters step S304, further search is performed.

[0065] For example, if the question is "the first survey of cultivated land area in Jinshui District, Zhengzhou City, Henan Province", the location information "Jinshui District, Zhengzhou City, Henan Province" is obtained according to step S301, and no time information is obtained; according to step S302, the location information is determined to be unique, and step S304 is entered; according to step S304, the land type information "cultivated land" is obtained; according to steps S301 and S304, the first entity set is obtained, and the first entity set contains the location information "Jinshui District, Zhengzhou City, Henan Province", the land type information "cultivated land", and the time information is empty; according to step S305, it is determined that the first land survey keyword in the entity set has "first survey"; because the end time of "first survey" is 1997, according to step S310, the time information in the entity set is corrected to 1997. By obtaining the location, time and land type information in the entity set, and correcting the first land survey keyword, the accuracy of semantic understanding is improved, and the entities in the knowledge graph library can be accurately located, which is conducive to the accuracy of information retrieval.

[0066] for example:

[0067] like Figure 5 The flowchart of question classification includes step S401, step S402, step S403, step S404 and step S405, wherein:

[0068] Step S401, determining whether the entity set contains a location or a land type.

[0069] Step S402: If the entity set does not contain location or land type, a third type of question is obtained.

[0070] Step S403, if the entity set includes a location or a land type, determining whether the entity set includes both a location and a land type;

[0071] Step S404: If the entity set includes both location and land type, the first type of question is obtained.

[0072] Step S405: If the entity set does not include both location and land type, a second type of question is obtained.

[0073] like Figure 6 The flowchart of the first type of problem classification is shown, which includes step S501, step S502, step S503, step S504, step S505, step S506, step S507, step S508 and step S509, wherein:

[0074] Step S501, obtaining a first similarity set, wherein the elements in the first similarity set are obtained by calculating the cosine similarity of a first question vector set; the first question vector set is obtained by inputting the question and a plurality of preset first typical questions into a Sentence-Bert model; the preset first typical question is a typical question of the first question.

[0075] Step S502, obtaining a second similarity set, wherein the elements in the second similarity set are obtained by calculating the cosine similarity of the second question vector set; the second question vector set is obtained by inputting the question and multiple preset second typical questions into the Sentence-Bert model; the preset second typical question is a typical question of the second problem.

[0076] Step S503, obtaining a third similarity set, wherein the elements in the third similarity set are obtained by calculating the cosine similarity of the third question vector set; the third question vector set is obtained by inputting the question and multiple preset third typical questions into the Sentence-Bert model; the preset third typical question is a typical question of the third problem.

[0077] Step S504: obtaining a maximum value of elements in the first similarity set, the second similarity set and the third similarity set.

[0078] Step S505 : determining the magnitude of the maximum value of the elements in the first similarity set, the second similarity set and the third similarity set relative to the threshold.

[0079] Step S506, obtaining the sixth question, when the maximum value of the elements in the first similarity set, the second similarity set and the third similarity set is less than a preset threshold; the third type of question belongs to the sixth question, and the sixth question is a non-related question.

[0080] Step S507, obtaining the first question when the element in the first similarity set is a maximum value.

[0081] Step S508, obtaining the second question when the element in the second similarity set is a maximum value.

[0082] Step S509, obtaining the third question when the element in the third similarity set is the maximum value.

[0083] In one embodiment of the present invention, the first question vector set includes ten vectors; the first question vector set is obtained by inputting the question and nine preset first typical questions into the Sentence-Bert model. The second question vector set includes ten vectors; the second question vector set is obtained by inputting the question and nine preset second typical questions into the Sentence-Bert model. The third question vector set includes ten vectors; the third question vector set is obtained by inputting the question and nine preset third typical questions into the Sentence-Bert model.

[0084] like Figure 7 The flowchart of the second type of problem classification is shown, which includes step S601, step S602, step S603, step S604 and step S605, wherein:

[0085] Step S601, obtaining a first probability, wherein the first probability is obtained by applying a naive Bayes classification algorithm to the question;

[0086] Step S602, obtaining a second probability, where the second probability is obtained by applying a naive Bayes classification algorithm to the question;

[0087] Step S603: determine whether the first probability is not less than the second probability.

[0088] Step S604, obtaining the fourth question when the first probability is not less than the second probability.

[0089] Step S605, obtaining the fifth question when the first probability is less than the second probability.

[0090] For example, if the question is "What is the cultivated land area in Jinshui District, Zhengzhou City, Henan Province?", and the entity set contains both the location information "Jinshui District, Zhengzhou City, Henan Province" and the land type information "cultivated land", then the question is determined to be classified as the first type of question, that is, the area type question.

[0091] Then, the question "A survey of cultivated land area in Jinshui District, Zhengzhou City, Henan Province" is input into the Sentence-Bert model with the preset first typical question, the preset second typical question, and the preset third typical question, and the cosine similarity is calculated to obtain the first similarity set, the second similarity set, and the third similarity set;

[0092] It is determined that the maximum value of the elements of the three similarity sets is greater than a preset threshold; it is determined that the maximum value of the three similarity sets is an element of the first similarity; thus, it is concluded that the question "the first survey of cultivated land area in Jinshui District, Zhengzhou City, Henan Province" is classified as the first problem, that is, a single-category area problem.

[0093] like Figure 8 The flowchart of answer retrieval includes steps S701, S702, S703, S704, S705, S706, S707 and S708, wherein:

[0094] Step S701, obtain a unique subgraph, the unique subgraph is retrieved according to the entity set and the first land survey keyword; the subgraph is obtained by querying the graph database to obtain the nodes and relationships corresponding to the graph database; the unique subgraph is determined by the query language.

[0095] Step S702, obtaining the land area information of the location when the question is the first question.

[0096] Step S703, obtaining the sum of the land area of ​​the location when the question is the second question.

[0097] Step S704, obtaining the ratio of the sum of the lower-level land category areas of the location to the upper-level land category area of ​​the location, when the question is the third question and the land category is a two-level land category with an inclusion relationship.

[0098] Step S705, obtaining the ratio of the sum of the land category areas of the location to the sum of the location areas, when the question is the third question and the land category does not satisfy the two-level land category inclusion relationship.

[0099] Step S706, obtaining the superior-subordinate inclusion relationship of the position when the question is the fourth question.

[0100] Step S707, obtaining the land category to which the location map belongs or the map distribution area of ​​the land category, when the question is the fifth question.

[0101] Step S708, the answer obtained is empty when the question is the sixth question.

[0102] For example: the question is "The first survey of cultivated land area in Jinshui District, Zhengzhou City, Henan Province". Since this question is the first question, and the entity set contains the location "Jinshui District, Zhengzhou City, Henan Province", the land type "cultivated land", the time and the land survey keyword "first survey", the unique subgraph can be directly obtained in the knowledge graph, and the corresponding area data in the database can be quickly obtained and output.

[0103] like Fig. 9 The structure block diagram of the device is shown as follows:

[0104] The input module 201 is used to obtain a question sentence.

[0105] The first obtaining module 202 is used to obtain an entity set, wherein the entity set is obtained by semantically parsing the question sentence according to a preset semantic parsing rule, and the entity set includes location, time and place type.

[0106] The second obtaining module 203 is used to obtain a first land survey keyword, where the first land survey keyword is obtained according to a land survey keyword process, and the first land survey keyword is used to indicate a survey order.

[0107] The third obtaining module 204 is used to obtain an answer, and the answer is obtained according to the entity set and the first land survey keyword.

[0108] The output module 205 is used to output the answer.

[0109] like Fig.10 As shown in the server structure block diagram, the server includes a processor and a memory connected via a system bus. The processor is used to provide computing and control capabilities to support the operation of the entire server. The memory stores an operating system and a computer operating program. The computer program can be executed by the processor to implement a land survey data automatic question-and-answer method provided in each of the above embodiments.

[0110] Those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention and are not intended to limit the present invention. As long as they are within the spirit of the present invention, appropriate changes and modifications to the above embodiments are within the scope of protection of the present invention.

Claims

1. A method for automatic question and answer of land survey data, characterized in that: include: Land and resources data are stored in the form of knowledge graphs, including five levels of administrative division entities: province, city, county, township, and village, land parcel entities uniquely identified by land parcel numbers, and land type entities uniquely identified by land type codes. Land type entities include primary, secondary, tertiary, eight categories, three categories, cultivated land refinement, planting attributes, and cultivated land quality. All land type entities have area attributes. In the knowledge graph relationship, the ownership relationship between administrative divisions and between plots and village entities has time attributes and name attributes; The land parcel entity has an ownership relationship with the lowest level land class, and the land classes have a containment relationship. Each land class has a time attribute, which corresponds to the attribute values ​​of the first adjustment, the second adjustment, and the third adjustment respectively; When the network collects ownership change information, the change time is added to the name attribute of the ownership relationship, and the time attribute is modified at the same time; The time attribute of the land class is a time attribute; Get the question sentence; Obtaining an entity set, wherein the entity set is obtained by semantically parsing the question sentence according to a preset semantic parsing rule, the entity set includes location, time, and land type, and the location includes several levels of administrative divisions; Obtaining a first land survey keyword, where the first land survey keyword is obtained according to a land survey keyword process, and the first land survey keyword is used to indicate a land survey order; Obtain an answer, wherein the answer is retrieved from the knowledge graph according to the entity set and the first land survey keyword; outputting the answer; The first land survey keyword is obtained according to a land survey keyword process, including: When the question sentence successfully matches the preset second land survey keyword string, the first land survey keyword is obtained, and the preset second land survey keyword includes a word or phrase used to indicate the survey order; when the question sentence fails to match the preset second land survey keyword string, and the land class time attribute is unique, the first land survey keyword is obtained according to the unique time attribute of the land class; when the question sentence fails to match the preset second land survey keyword string, the land class time attribute is not unique, and there is a corresponding relationship between the time and the preset land survey time, the first land survey keyword is obtained according to the time; when the question sentence fails to match the preset second land survey keyword string, the land class time attribute is not unique, and there is no corresponding relationship between the time and the preset land survey time, the first land survey keyword is obtained according to the latest time attribute in the land class; Correct the time of the entity set. When the time cannot be obtained based on the BERT-BILSTM-CRF model and the question contains the first land survey keyword, the time is the time corresponding to the first land survey keyword; when the time cannot be obtained based on the BERT-BILSTM-CRF model and the question does not contain the first land survey keyword, the time is the time when the question is input.

2. The automatic question-answering method for land survey data according to claim 1 is characterized in that: After obtaining the question sentence, the method further includes obtaining the preprocessed question sentence, wherein the preprocessed question sentence includes: The preprocessed question sentence is obtained by removing emoticons from the question sentence, converting traditional Chinese characters in the question sentence into simplified Chinese characters, converting the full-width input format in the question sentence into the diagonal input format, and correcting homophonic and similar input errors in the question sentence.

3. The automatic question-answering method for land survey data according to claim 1 is characterized in that: An entity set is obtained by semantically parsing the question according to a preset semantic parsing rule, and the entity set includes location, time, and land type, including: Obtaining a location and a time, where both the location and the time are obtained based on a BERT-BILSTM-CRF model; Obtain a land type, wherein the land type is obtained according to a string matching method.

4. The automatic question-answering method for land survey data according to claim 3 is characterized in that: After obtaining the location and time, the location and time are obtained based on the BERT-BILSTM-CRF model and the preset rules, the method further includes: Output the first position and the second position when the position results obtained are not unique.

5. The automatic question-answering method for land survey data according to claim 3 is characterized in that: Obtaining a location and a time, where the location and the time are obtained based on a BERT-BILSTM-CRF model and a preset rule, including: The several levels of administrative divisions are divided according to a word segmentation method of a dictionary, and the dictionary consists of administrative divisions and aliases of the administrative divisions.

6. The automatic question-answering method for land survey data according to claim 5 is characterized in that: After obtaining the first land survey keyword, the question classification is also included, and the question classification includes: A first type of question is obtained, wherein the entity set of the first type of question includes location and land type; A second type of question is obtained, wherein the entity set of the second type of question includes one of a location or a land type; A third type of question is obtained, and the entity set of the third type of question does not include location and land type.

7. The automatic question-answering method for land survey data according to claim 6 is characterized in that: After obtaining the first type of problems, the method further includes classifying the first type of problems, wherein the first type of problems includes the first problem, the second problem and the third problem; the first problem is a single type of area problem, the second problem is an area sum problem, and the third problem is an area proportion problem: A first similarity set is obtained, wherein the elements in the first similarity set are obtained by calculating the cosine similarity of the first question vector set; the first question vector set is obtained by inputting the question and a plurality of preset first typical questions into a Sentence-Bert model; the preset first typical question is a typical question of the first question; A second similarity set is obtained, wherein the elements in the second similarity set are obtained by calculating the cosine similarity of the second question vector set; the second question vector set is obtained by inputting the question and a plurality of preset second typical questions into a Sentence-Bert model; the preset second typical question is a typical question of the second question; A third similarity set is obtained, wherein the elements in the third similarity set are obtained by calculating the cosine similarity of the third question vector set; the third question vector set is obtained by inputting the question and a plurality of preset third typical questions into a Sentence-Bert model; the preset third typical question is a typical question of the third problem; Obtaining a maximum value of elements in the first similarity set, the second similarity set, and the third similarity set; The first problem is obtained when an element in the first similarity set is a maximum value; The second problem is obtained when the element in the second similarity set is a maximum value; The third problem is obtained when the element in the third similarity set is a maximum value.

8. The automatic question-answering method for land survey data according to claim 7 is characterized in that: After obtaining the maximum value of the elements in the first similarity set, the second similarity set and the third similarity set, the following further includes threshold determination: Determine the magnitude of a maximum value of elements in the first similarity set, the second similarity set, and the third similarity set relative to the threshold; A sixth question is obtained when the maximum value of the elements in the first similarity set, the second similarity set and the third similarity set is less than a preset threshold; the third type of question belongs to the sixth question, and the sixth question is an irrelevant question.

9. The automatic question-answering method for land survey data according to claim 8 is characterized in that: After obtaining the second type of problems, the second type of problems is further classified. The second type of problems includes the fourth and fifth problems. The fourth problem is a containment relationship problem, and the fifth problem is a distribution problem: Obtaining a first probability, wherein the first probability is obtained by applying a naive Bayes classification algorithm to the question; Obtaining a second probability, where the second probability is obtained by applying a naive Bayes classification algorithm to the question; The fourth problem is obtained when the first probability is not less than the second probability; The fifth problem is obtained when the first probability is less than the second probability.

10. The automatic question-answering method for land survey data according to claim 9, characterized in that: Obtaining an answer, wherein the answer is obtained according to the entity set and the first land survey keyword, including: Obtain a unique subgraph, the unique subgraph is retrieved according to the entity set and the first land survey keyword; the subgraph is obtained by querying the graph database to obtain nodes and relationships corresponding to the graph database; the unique subgraph is determined by a query language; obtaining the land area information of the location when the question is the first question; obtaining the sum of the land areas of the location when the question is the second question; Obtaining the ratio of the sum of the lower-level land category areas of the location to the upper-level land category area of ​​the location, when the question is the third question and the land category is a two-level land category with an inclusion relationship; Obtaining a ratio of the sum of the land category areas of the location to the sum of the location areas, when the question is the third question and the land category is not a two-level land category containing relationship; Obtaining the superior-subordinate inclusion relationship of the position, when the question is the fourth question; Obtaining the land category to which the location map belongs or the map distribution area of ​​the land category, when the question is the fifth question; The answer obtained is empty when the question is the sixth question.

11. An automatic question-answering device for land survey data, characterized in that: The device comprises: the device executes the method as claimed in claim 1, an input module for obtaining a question sentence; A first obtaining module is used to obtain an entity set, wherein the entity set is obtained by semantically parsing the question sentence according to a preset semantic parsing rule, and the entity set includes location, time and land type; A second obtaining module obtains a first land survey keyword, wherein the first land survey keyword is obtained according to a land survey keyword process, and the first land survey keyword is used to indicate a survey order; A third obtaining module is used to obtain an answer, wherein the answer obtains a result according to the entity set; An output module is used to output the answer.

12. A land survey data automatic question-answering server, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by the processor, the processor executes the automatic question-and-answer method for land survey data according to any one of claims 1 to 10.

Images