CN113032418B - Method for converting complex natural language query into SQL (structured query language) based on tree model - Google Patents
Method for converting complex natural language query into SQL (structured query language) based on tree model Download PDFInfo
- Publication number
- CN113032418B CN113032418B CN202110183393.4A CN202110183393A CN113032418B CN 113032418 B CN113032418 B CN 113032418B CN 202110183393 A CN202110183393 A CN 202110183393A CN 113032418 B CN113032418 B CN 113032418B
- Authority
- CN
- China
- Prior art keywords
- sql
- natural language
- query
- model
- complex
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/242—Query formulation
- G06F16/243—Natural language query formulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/242—Query formulation
- G06F16/2433—Query languages
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a method for converting complex natural language query into SQL (structured query language) based on a tree model, which comprises the steps of firstly, pre-screening an input database to obtain TOP K tables related to natural language query, converting multi-table query into single-table query and adapting to the condition that the database has larger scale so that information cannot be input into the SQL conversion model; then, aiming at the complex SQL statement, the complex SQL statement is disassembled into a multi-branch tree form from top to bottom, each node represents one module, and therefore a tree model is constructed to predict the output of each module of the complex SQL statement; and finally, separating the value extraction from the tree model aiming at the task of SQL statement value extraction, and simultaneously extracting the values of different clauses in SQL by sharing the weight. The method frame for converting the complex natural language query into the SQL based on the tree model can solve the problem of multi-table complex query in the task of converting the natural language query into the SQL.
Description
Technical Field
The invention belongs to the technical field of natural language processing, and particularly relates to a technical method for converting complex natural language query into SQL (structured query language) based on a tree model.
Background
A great deal of information in the world is stored in a database, however, the acquisition and analysis of the data often need to be mastered in SQL and other formal query languages, so that the use of non-technical users is limited. The technology can understand spoken language query of a user and structural information of the database, fuse and analyze the spoken language query and the structural information to generate SQL sentences understandable by the database, and then execute query on the database to return results, thereby realizing automatic question answering capability based on the database.
At present, research on conversion of natural language query to SQL in academic circles is rapidly progressed, however, most of existing work is based on an English scene, when the method is applied to the Chinese field, the accuracy of conversion of the complex SQL is greatly reduced due to special spoken language expression of Chinese, and new requirements for business intelligence-oriented calculation query cannot be met. Meanwhile, the current research work ignores the possibility that the information cannot be completely input into the model due to the large scale of the database in reality and the problem that the performance of the model is reduced due to the fact that a new database with a changed structure faces in the process of cross-domain query. Aiming at the problems, a set of complex natural language query-to-SQL system framework based on a tree model is provided, different test sets on a Chinese DuSQL data set respectively obtain the scores of a first name and a second name, and the effectiveness of a solution is proved, so that the method has certain academic value and practical significance.
Disclosure of Invention
In order to overcome the defects of the existing solution aiming at the problem of converting complex query into SQL, the invention provides a natural language query into SQL technical framework based on a tree model, which takes the problem of converting complex and changeable Chinese spoken language query into SQL as a core research point, and decomposes complex SQL sentences into a multi-branch tree form from top to bottom, thereby providing a growing tree model; meanwhile, the new demand of computing query facing business intelligence is solved by upgrading the system; aiming at the problem of performance reduction caused by structural change of a cross-domain query database, a table information enhancement algorithm is provided to improve the stability of a model; aiming at the problem that the database with any scale cannot be processed by the model under the condition of low resources, the screening model for pre-screening the database table is provided. Finally, a set of complex natural language query SQL system framework based on a tree model is set up, excellent results are obtained on a Chinese DuSQL data set, and the effectiveness of the solution is proved.
The purpose of the invention is realized by the following technical scheme: a method for converting complex natural language query into SQL based on tree model includes the following steps:
(1) Preprocessing input natural language spoken language queries, wherein the preprocessing specifically comprises the steps of describing time, date and numerical values of the input natural language spoken language queries according to a record format stored in a database;
(2) Pre-screening all tables in an input database through a screening model to obtain K tables with the highest relevance degree to spoken language query;
(3) For the input natural language query and the screened table, predicting the result of each component module of the complex SQL statement through a tree model;
(4) Complementing the SQL result predicted by the tree model, firstly identifying the condition columns needing complementing the condition values, and then respectively obtaining the condition values corresponding to each condition column through a value extraction model;
(5) And constructing a json-form prediction result, and obtaining the expression form of the SQL statement through an SQL analysis post-processing algorithm.
Further: the step (2) includes the sub-steps of:
(2.1) the input section is a natural language query Q and a database schema, wherein only the table name and field information of each data table in the database schema are concerned, and the representation form of each table information T is defined as "table name __ column name 1_ column name 2 …";
and (2.2) inputting Q and all T in the step (2.1) into a screening model, obtaining the relevance score of each table T and the natural language Q, and storing K tables with the highest relevance.
Further: the step (3) includes the substeps of:
(3.1) dividing the nested SQL problem of the complex SQL statement core, dividing the nested SQL into a main SQL part and an auxiliary SQL part, decomposing clause hierarchies such as Select, where and the like in each SQL hierarchy, and performing inductive division on each clause hierarchy to obtain target hierarchies such as column names, operators and the like, so as to form a multi-branch tree and build a tree model;
(3.2) the input part is a data table of natural language query Q and TOP K, all the tables are merged, column name information c in each table is represented in a form of 'table name _ column name', and the column name information c is spliced together and input into the tree model;
(3.3) obtaining the coding characteristics of the natural language q and each column name C in (3.2) through a pre-training language model coder, performing dimensionality reduction on each C, and splicing to obtain a code C of the whole column name sequence, wherein the dimensionality reduction form is as follows:
RfAttentionPooling(q,c)＝score T ·c
wherein D represents the dimension of the coding feature,
representing trainable weight parameters.
(3.4) recoding the problem query code q entering the auxiliary SQL module through a Transformer coding layer to obtain q';
and (3.5) sending the codes obtained in the steps (3.3) and (3.4) into a tree model, and acquiring results of all components of the SQL statement in a depth-first search (DFS) mode.
Further: the step (4) comprises the following substeps:
(4.1) processing the prediction results of all modules of the SQL obtained in the step (3) to obtain the information of all condition columns needing value extraction in the form of 'table name __ column name' and natural language query concatenation as the input of an extraction model;
and (4.2) the extraction model can simultaneously predict the results of input column names possibly under different SQL clauses, share the weight and finally supplement condition value information to each condition column. So far, the prediction results of all components of the target SQL statement are obtained;
further: the step (5) comprises the following substeps:
(5.1) converting the prediction result of each clause level under each SQL level into a character string form according to a labeling format, and then adding keywords such as Select and the like according to the clause type;
(5.2) aiming at the analytic character strings (5.1) of each clause under each SQL layer, splicing according to the sequence of each keyword of the SQL sentence to obtain a simple SQL sentence;
and (5.3) aiming at the prediction result in the nested form, combining the simple SQL statements (5.2) analyzed by different SQL layers to finally obtain a result SQL statement. Further, in the training stage of the tree model in the step (3), a table information enhancement algorithm is adopted, and the algorithm specifically carries out operations of random discarding and random arrangement on the table level and the column name level in each table, so as to realize data enhancement.
Compared with the prior art, the invention has the following beneficial effects: a set of complex natural language query SQL system framework based on a tree model is built, complex SQL sentences are disassembled into a multi-way tree form from top to bottom, and a growing tree model is provided; meanwhile, the new demand of computing query facing business intelligence is solved by upgrading the system; aiming at the problem of performance reduction caused by structural change of a cross-domain query database, a table information enhancement algorithm is provided to improve the stability of a model; aiming at the problem that the database with any scale cannot be processed by the model under the condition of low resources, the screening model for pre-screening the database table is provided. And excellent results are obtained on the Chinese DuSQL data set, the effectiveness of the solution is proved, and certain academic value and practical significance are realized.
Drawings
FIG. 1 is an overall architecture diagram of a tree model-based complex natural language query-to-SQL system framework employed in the present invention;
FIG. 2 is an exemplary diagram of the parsing of a complex SQL statement into a multi-way tree form according to the invention;
FIG. 3 is an exemplary diagram of a tree model corresponding to the multi-way tree of FIG. 2 in accordance with the present invention;
FIG. 4 is an exemplary diagram of the tree model-based conversion of complex natural language query into SQL model according to the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, the present invention shall be described in further detail with reference to the following detailed description and accompanying drawings.
The invention provides a natural language query SQL conversion method framework based on a tree model, which is shown in figure 1. The invention takes the problem of converting complex and changeable Chinese spoken language query into SQL as a core research point, and decomposes complex SQL sentences into a multi-branch tree form from top to bottom, thereby providing a growing tree model; meanwhile, the new demand of computing query facing business intelligence is solved by upgrading the system; aiming at the problem of performance reduction caused by structural change of a cross-domain query database, a table information enhancement algorithm is provided to improve the stability of a model; aiming at the problem that the database with any scale cannot be processed by the model under the condition of low resources, the screening model for pre-screening the database table is provided. Finally, a set of complex natural language query SQL system framework based on a tree model is built, excellent results are obtained on a Chinese DuSQL data set, effectiveness of a solution is proved, and certain academic value and practical significance are achieved.
The natural language query SQL conversion technical framework based on the tree model comprises the following steps:
(1) Preprocessing input natural language spoken language queries, wherein the preprocessing comprises standardized operations such as time values and the like;
(2) Pre-screening all tables in an input database through a screening model to obtain TOP K tables relevant to spoken language query;
(3) For the input natural language query and the screened table, predicting the result of each component module of the complex SQL statement through a tree model;
(4) Completing the SQL result predicted by the tree model, and taking out the conditional column names needing value extraction to obtain answers through the value extraction model;
(5) And at this moment, all components of the whole target SQL are predicted, a json-form prediction result is constructed, and the expression form of the SQL statement is obtained through an SQL analysis post-processing algorithm.
Further, the step (2) includes the following sub-steps:
(2.1) the input part is a natural language query Q and a database mode, wherein only the table name and field information of each data table in the database mode are concerned, and the representation form of each table information T is defined as 'table name __ column name 1_ column name 2 …';
(2.2) inputting Q and all T in (2.1) into the screening model, obtaining the relevance score of each table T and the natural language Q, and storing TOP K tables.
Further, the calculation method of the loss value in the step (2) is a binary cross entropy loss function, and the optimization method is a gradient descent optimizer of Adam adaptive learning rate including a Warmup mechanism.
Further, step (3) comprises the following substeps:
(3.1) dividing the nested SQL problem of the complex SQL statement core, such as set operation nesting, conditional operation nesting and line calculation nesting, dividing the nested SQL into a main SQL part and an auxiliary SQL part, decomposing the 'select', 'where', 'having', 'groupby' and 'orderby' Clause (class) hierarchies again in each SQL hierarchy, such as the 'where' Clause 'year = 2017', and similar structures of different keyword clauses (class), such as all clauses containing column names, performing induction and division on each Clause hierarchy to obtain Target (Target) hierarchies such as column names and operators, and forming a multi-branch tree and building a tree model, such as the multi-branch tree shown in FIG. 2, and FIG. 3 shows a corresponding modular tree model;
(3.2) the input part is a natural language query Q and TOP K data tables, the tables are merged, column name information c in each table is represented in a form of table name _ column name, and the column name information c is spliced and input into a network structure containing a tree model, as shown in FIG. 4;
(3.3) firstly, acquiring the coding characteristics of the natural language q and each column name C in (3.2) through a pre-training language model coder, then performing dimension reduction processing on each C, and splicing to obtain a code C of the whole column name sequence, wherein the dimension reduction form is as follows:
RfAttentionPooling(q,c)＝score T ·c
wherein D represents the dimension of the coding feature,
representing trainable weight parameters.
(3.4) considering that the attention points of different SQL clauses of the complex SQL sentences to the natural language query are different, the problem query codes q input by the two SQL modules are the same and have the defect of information interference and confusion, and therefore the problem query codes q entering the secondary SQL module are recoded through a Transformer coding layer to obtain q';
and (3.5) sending the codes obtained in the step (3.3) and the step (3.4) into a tree model, and obtaining results of all components of the SQL statement in a depth-first search (DFS) mode.
Further, the calculation method of the loss value in the step (3) is a binary cross entropy loss function and a cross entropy loss function, different loss functions are adopted for different Target prediction outputs, the final loss is the weighted addition of the outputs of the modules, and the adopted optimization method is a gradient descent optimizer of Adam adaptive learning rate including a Warmup mechanism.
Further, the step (4) includes the following sub-steps:
(4.1) processing the prediction results of all the modules of the SQL acquired in the step (3) to acquire information of all condition columns needing to be extracted, wherein the information is in the form of 'table name __ column name' and is spliced with natural language query to be used as input of an extraction model;
(4.2) the extraction model can simultaneously predict the results of input column names possibly under different SQL clauses, shares weights, finally supplements complete value information, obtains the prediction results of all components of the target SQL sentence, and obtains a result SQL sentence through an analysis algorithm;
further, in the training stage of the tree model in the step (3), a table information enhancement algorithm is adopted, and data enhancement is performed by performing operations of random discarding and random arrangement on the table hierarchy and the column name hierarchy in each table.
Examples
In the following, natural language is used to query "which celebrities in colleges and universities are not more than five in the colleges and universities, and the category" of these colleges and universities is converted into SQL statement "select T2. Category, T2. The name is from celebrity as T1 join colleges as T2 on celebrities, college id = = colleges and universities, term id group by T1. College id having count (= 5" is taken as an example, the database corresponding to the problem is the chinese college database (colleges and universities, celebrities, awards, college attributes), and referring to fig. 1, the method for converting complex natural language query into SQL of the present invention is described.
(1) Preprocessing natural language queries, converting the natural language queries into 'in colleges and universities of famous graduation, which celebrities do not exceed 5, and giving categories of the colleges and universities';
(2) Setting K =2, and screening a database to obtain a TOP K related table, namely a college and celebrity data table;
(3) The method comprises the steps of inputting a natural language query 'in colleges and universities of famous people graduation, which celebrities do not exceed 5, giving categories of the colleges and universities' and a data table of the colleges and the celebrities into a tree model, obtaining SQL module results without values, wherein in order to realize visualization display, the results are visualized into a statement form, and finally obtaining 'select college category, colleges and universities' name group by celebrities 'id having count (= VAL'), so that the SQL statement value can be found to be replaced by VAL and a nested SQL situation is not included;
(4) Supplementing the value of the result of the tree model prediction, firstly obtaining the value involved in the prediction SQL:
(Q, college _ - >5;
(5) Then, the value extraction model extracts the input of the category and the column name of Chinese colleges of natural language query that the number of the celebrities in the famous colleges and universities is not more than 5 to obtain an answer of 5;
(6) And finally, analyzing to obtain a final SQL statement result, namely a "select T2. Type, a" name from a celebrity as T1 join colleges and universities as T2 on celebrity, colleges and universities id = = colleges and universities, entry id group by T1. Colleges and universities id having count (= 5 "). The above embodiments are only intended to illustrate the technical solution of the present invention and not to limit the same, and a person skilled in the art can modify the technical solution of the present invention or substitute the same without departing from the principle and scope of the present invention, and the scope of the present invention should be determined by the claims.
Claims (4)
1. A method for converting complex natural language query into SQL based on tree model is characterized by comprising the following steps:
(1) Preprocessing input natural language spoken language queries, wherein the preprocessing specifically comprises the steps of describing time, date and numerical values of the input natural language spoken language queries according to a record format stored in a database;
(2) Pre-screening all tables in an input database through a screening model to obtain K tables with the highest relevance degree to spoken language query;
(3) For the input natural language query and the screened table, predicting the result of each component module of the complex SQL statement through a tree model; the method comprises the following substeps:
(3.1) dividing the nested SQL problem of the complex SQL statement core, dividing the nested SQL into a main SQL part and a sub SQL part, decomposing a Select clause hierarchy and a Where clause hierarchy in each SQL hierarchy, and carrying out inductive division on each clause hierarchy to obtain column names and an operator target hierarchy, so as to form a multi-branch tree and build a tree model;
(3.2) the input part is a data table of natural language query Q and TOP K, all the tables are merged, column name information c in each table is represented in a form of 'table name _ column name', and the column name information c is spliced together and input into the tree model;
(3.3) obtaining the coding characteristics of the natural language q and each column name C in (3.2) through a pre-training language model coder, performing dimensionality reduction on each C, and splicing to obtain a code C of the whole column name sequence, wherein the dimensionality reduction form is as follows:
RfAttentionPooling(q,c)＝score T ·c
wherein D represents the dimension of the coding feature, W 1 ,
A weight parameter representing the training;
(3.4) recoding the question query code q entering the auxiliary SQL module through a Transformer coding layer to obtain q';
(3.5) sending the codes obtained in the steps (3.3) and (3.4) into a tree model, and obtaining results of all components of the SQL statement in a depth-first search (DFS) mode;
(4) Completing the SQL result predicted by the tree-shaped model, firstly identifying the condition columns needing condition value completion, and then respectively obtaining the condition value corresponding to each condition column through a value extraction model;
(5) And constructing a json-form prediction result, and obtaining an expression form of the SQL statement after SQL analysis.
2. The method of converting complex natural language queries to SQL as claimed in claim 1, wherein: the step (2) includes the sub-steps of:
(2.1) the input section is a natural language query Q and a database schema, wherein only the table name and field information of each data table in the database schema are concerned, and the representation form of each table information T is defined as "table name __ column name 1_ column name 2 …";
and (2.2) inputting the Q and all the T in the step (2.1) into a screening model to obtain a relevance score of each table T and the natural language Q, and storing K tables with highest relevance.
3. The method of converting complex natural language queries to SQL as claimed in claim 1, wherein: the step (4) comprises the following substeps:
(4.1) processing the prediction results of all modules of the SQL obtained in the step (3) to obtain the information of all condition columns needing value extraction in the form of 'table name __ column name' and natural language query concatenation as the input of an extraction model;
(4.2) the extraction model can simultaneously predict the results of the input column names under different SQL clauses, share the weight and finally supplement condition value information to each condition column; so far, the prediction results of the components of the target SQL statement are obtained.
4. The complex natural language query to SQL method according to claim 1, wherein the step (5) comprises the sub-steps of:
(5.1) converting the prediction result of each clause level under each SQL level into a character string form according to a labeling format, and adding Select keywords according to the clause type;
(5.2) aiming at the analysis character string (5.1) of each clause under each SQL level, splicing according to the sequence of each keyword of the SQL statement to obtain a simple SQL statement;
and (5.3) aiming at the prediction result in the nested form, combining the simple SQL sentences (5.2) analyzed by different SQL layers to finally obtain the result SQL sentences.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110183393.4A CN113032418B (en) | 2021-02-08 | 2021-02-08 | Method for converting complex natural language query into SQL (structured query language) based on tree model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110183393.4A CN113032418B (en) | 2021-02-08 | 2021-02-08 | Method for converting complex natural language query into SQL (structured query language) based on tree model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113032418A CN113032418A (en) | 2021-06-25 |
| CN113032418B true CN113032418B (en) | 2022-11-11 |
Family
ID=76460746
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110183393.4A Active CN113032418B (en) | 2021-02-08 | 2021-02-08 | Method for converting complex natural language query into SQL (structured query language) based on tree model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113032418B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113553414B (en) * | 2021-06-30 | 2023-08-25 | 北京百度网讯科技有限公司 | Intelligent dialogue method, intelligent dialogue device, electronic equipment and storage medium |
| CN113408298B (en) * | 2021-06-30 | 2024-05-31 | 北京百度网讯科技有限公司 | Semantic analysis method, semantic analysis device, electronic equipment and storage medium |
| CN114218287B (en) * | 2021-12-30 | 2022-11-04 | 北京诺司时空科技有限公司 | Query time prediction method for time sequence database |
| CN115221198A (en) * | 2022-01-19 | 2022-10-21 | 支付宝(杭州)信息技术有限公司 | Data query method and device |
Family Cites Families (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20180093157A (en) * | 2017-02-09 | 2018-08-21 | 서울대학교산학협력단 | A question translation system based on dependency tree and semantic representation and the method thereof |
| WO2018236886A1 (en) * | 2017-06-21 | 2018-12-27 | Opera Solutions Usa, Llc | System and method for code and data versioning in computerized data modeling and analysis |
| CN107516135A (en) * | 2017-07-14 | 2017-12-26 | 浙江大学 | A kind of automation monitoring learning method for supporting multi-source data |
| CN108519890B (en) * | 2018-04-08 | 2021-07-20 | 武汉大学 | Robust code abstract generation method based on self-attention mechanism |
| US11087088B2 (en) * | 2018-09-25 | 2021-08-10 | Accenture Global Solutions Limited | Automated and optimal encoding of text data features for machine learning models |
| CN110688394B (en) * | 2019-09-29 | 2021-11-23 | 浙江大学 | NL generation SQL method for novel power supply urban rail train big data operation and maintenance |
| CN112035506A (en) * | 2019-10-28 | 2020-12-04 | 竹间智能科技（上海）有限公司 | Semantic recognition method and equipment |
| CN110888897B (en) * | 2019-11-12 | 2020-07-14 | 杭州世平信息科技有限公司 | Method and device for generating SQ L statement according to natural language |
| CN111159220B (en) * | 2019-12-31 | 2023-06-23 | 北京百度网讯科技有限公司 | Method and apparatus for outputting structured query statement |
| CN112100426B (en) * | 2020-09-22 | 2024-05-24 | 哈尔滨工业大学（深圳） | Method and system for universal form information retrieval based on visual and text features |
| CN112185558A (en) * | 2020-09-22 | 2021-01-05 | 珠海中科先进技术研究院有限公司 | Mental health and rehabilitation evaluation method, device and medium based on deep learning |
-
2021
- 2021-02-08 CN CN202110183393.4A patent/CN113032418B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113032418A (en) | 2021-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113032418B (en) | Method for converting complex natural language query into SQL (structured query language) based on tree model | |
| CN105701253B (en) | The knowledge base automatic question-answering method of Chinese natural language question semanteme | |
| US8005815B2 (en) | Search engine | |
| CN111340661B (en) | Automatic application problem solving method based on graph neural network | |
| CN111858932A (en) | Multiple-feature Chinese and English emotion classification method and system based on Transformer | |
| CN115858758A (en) | Intelligent customer service knowledge graph system with multiple unstructured data identification | |
| CN110175585A (en) | It is a kind of letter answer correct system and method automatically | |
| CN114238653B (en) | Method for constructing programming education knowledge graph, completing and intelligently asking and answering | |
| CN114281968B (en) | Model training and corpus generation method, device, equipment and storage medium | |
| Jayaram et al. | A review: Information extraction techniques from research papers | |
| CN114625748A (en) | SQL query statement generation method and device, electronic equipment and readable storage medium | |
| CN113297251A (en) | Multi-source data retrieval method, device, equipment and storage medium | |
| CN111831624A (en) | Data table creating method and device, computer equipment and storage medium | |
| Xiong et al. | Transferable natural language interface to structured queries aided by adversarial generation | |
| CN117251455A (en) | Intelligent report generation method and system based on large model | |
| Chuprina et al. | Ontology based data access methods to teach students to transform traditional information systems and simplify decision making process | |
| CN117390198A (en) | Method, device, equipment and medium for constructing scientific and technological knowledge graph in electric power field | |
| CN117349311A (en) | Database natural language query method based on improved RetNet | |
| CN117473054A (en) | Knowledge graph-based general intelligent question-answering method and device | |
| CN117271558A (en) | Language query model construction method, query language acquisition method and related devices | |
| CN112183110A (en) | Artificial intelligence data application system and application method based on data center | |
| CN106776590A (en) | A kind of method and system for obtaining entry translation | |
| CN114722159B (en) | Multi-source heterogeneous data processing method and system for numerical control machine tool manufacturing resources | |
| CN114969087A (en) | NL2SQL method and device based on multi-view feature decoupling | |
| CN112347121B (en) | Configurable natural language sql conversion method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |