Structured data generation method and related device

Abstract

The invention discloses a structured data generation method and a related device, and relates to the technical field of data processing, and the structured data generation method comprises the following steps: obtaining target text data; using the large model, according to the target text data, under the constraint of a text unit generation rule contained in a security grammar specification constructed according to a predefined information extraction structured framework, generating structured data conforming to the information extraction structured framework, and obtaining structured data corresponding to the target text data; wherein the information extraction structured framework is used for indicating data needing to be extracted and an organization form of a data extraction result, and the text unit generation rule contained in the security grammar specification is a text unit generation rule related to a structure indicated by the information extraction structured framework. According to the structured data generation method provided by the invention, accurate structured data can be stably generated.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a structured data generation method and related devices. Background Art

[0002] In natural language processing (NLP) tasks, structured data facilitates accurate information transmission and analysis, and can be directly used in downstream tasks. With the widespread application of large models in natural language processing, knowledge extraction, task automation, and other fields, generating structured data (such as structured data in JSON format) has become a critical requirement, especially in data interaction, information transmission, and business system integration. However, how to use large models to stably and accurately generate structured data is a pressing issue. Summary of the Invention

[0003] In view of this, the present application provides a structured data generation method and related devices for stably and accurately generating structured data using a large model. The technical solution is as follows:

[0004] The first aspect of the present application provides a method for generating structured data, comprising:

[0005] Get target text data;

[0006] Utilizing a large model, based on the target text data, and under the constraints of the text unit generation rules contained in the security grammar specification constructed based on a pre-defined information extraction structured framework, structured data that conforms to the information extraction structured framework is generated to obtain structured data corresponding to the target text data; wherein, the information extraction structured framework is used to indicate the data to be extracted and the organizational form of the data extraction results, and the text unit generation rules contained in the security grammar specification are text unit generation rules related to the structure indicated by the information extraction structured framework.

[0007] In a possible implementation, the information extraction structured framework includes: a structured framework for entity extraction, a structured framework for entity relationship extraction, and a structured framework for event extraction.

[0008] In a possible implementation, the structured framework of the entity extraction is a first array object, the first array object includes a plurality of first dictionary objects, each first dictionary object represents structured information of an entity, and each first dictionary object includes keywords for entity extraction;

[0009] The structured framework of the entity relationship extraction is a second array object, the second array object includes a plurality of second dictionary objects, each second dictionary object represents the structured information of an entity relationship, and each second dictionary object includes keywords for entity relationship extraction;

[0010] The structured framework of the event extraction is a third array object, which includes several third dictionary objects. Each third dictionary object represents the structured information of an event. Each third dictionary object includes keywords for event extraction, and each third dictionary object includes several sub-dictionary objects.

[0011] In one possible implementation, utilizing the large model, under the constraints of text unit generation rules contained in a security grammar specification constructed according to a predefined information extraction structured framework, to generate structured data conforming to the predefined information extraction structured framework from the target text data includes:

[0012] Using the large model, based on the target text data and the text unit table, and under the constraints of the text unit generation rules contained in the security grammar specification, text units are generated one by one, and the sequence composed of the generated text units is structured data that conforms to the information extraction structured framework;

[0013] Among them, the text unit summary table includes a basic text unit table and a supplementary text unit table. The supplementary text unit table is a structure-related text unit table constructed according to the information extraction structured framework. The supplementary text unit table includes keywords and symbols involved in the information extraction structured framework.

[0014] In one possible implementation, the large model is used to generate text units one by one according to the target text data and the text unit table, under the constraints of the text unit generation rules included in the security grammar specification, including:

[0015] Extracting features from the target text data using the large model to obtain features of the target text data;

[0016] Use the large model to generate steps in each text unit:

[0017] Based on the characteristics of the target text data and the generated text units, predicting the probability that the next text unit will be each text unit in the text unit table, and obtaining the probability distribution of the next text unit on the text unit table;

[0018] Under the constraints of the text unit generation rules included in the security grammar specification, the next text unit is determined according to the probability distribution of the next text unit on the text unit master table to obtain the text unit generated in the current step.

[0019] In one possible implementation, determining the next text unit according to the probability distribution of the next text unit in the text unit master table under the constraints of the text unit generation rules included in the security grammar specification to obtain the text unit generated in the current step includes:

[0020] Screening text units that meet the text unit generation rules included in the security grammar specification from the text unit master table to obtain a plurality of first candidate text units;

[0021] Obtaining K maximum probabilities from the probability distribution of the next text unit on the text unit total table, determining the text units corresponding to the K maximum probabilities as second candidate text units, and obtaining K second candidate text units, where K is an integer greater than 0;

[0022] The next text unit is determined from the plurality of first candidate text units and the K second candidate text units to obtain the text unit generated in the current step.

[0023] In a possible implementation, the text units that meet the text unit generation rules included in the security grammar specification are screened from the text unit master table to obtain a plurality of first candidate text units, including:

[0024] For each text unit in the text unit summary table:

[0025] Determining whether the text unit can be generated after the last generated text unit according to the text unit generation rules included in the security grammar specification;

[0026] If it is determined that the text unit can be generated after the last text unit that has been generated, then the text unit is determined to comply with the security grammar specification;

[0027] If the text unit complies with the security grammar specification, the text unit is determined as the first candidate text unit.

[0028] In a possible implementation, determining the next text unit from the plurality of first candidate text units and the K second candidate text units to obtain the text unit generated in the current step includes:

[0029] Determine the intersection of the plurality of first candidate text units and the K second candidate text units to obtain a plurality of third candidate text units;

[0030] Obtaining the probabilities corresponding to the plurality of third candidate text units respectively from the probability distribution of the next text unit on the text unit total table;

[0031] According to the respective probabilities corresponding to the plurality of third candidate text units, the next text unit is determined from the plurality of third candidate text units to obtain the text unit generated in the current step.

[0032] A second aspect of the present application provides a structured data generating device, comprising: a text data acquisition module and a text data structuring module;

[0033] The text data acquisition module is used to acquire target text data;

[0034] The text data structuring module is used to use a large model to generate structured data that conforms to the information extraction structuring framework based on the target text data, under the constraints of the text unit generation rules contained in the security grammar specification constructed according to a pre-defined information extraction structuring framework, to obtain structured data corresponding to the target text data; wherein, the information extraction structuring framework is used to indicate the data to be extracted and the organizational form of the data extraction results, and the text unit generation rules contained in the security grammar specification are text unit generation rules related to the structure indicated by the information extraction structuring framework.

[0035] A third aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0036] The memory is used to store computer programs;

[0037] The processor is configured to execute the computer program so that the electronic device can implement the steps of any one of the above-mentioned structured data generation methods.

[0038] In a fourth aspect, the present application provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the steps of any one of the above-mentioned structured data generation methods.

[0039] In a fifth aspect, the present application provides a computer program product comprising computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the steps of any one of the above-described structured data generation methods.

[0040] By means of the above technical solution, the structured data generation method provided by the present application can, after obtaining the target text data, utilize the large model to generate structured data that conforms to the information extraction structured framework according to the target text data under the constraints of the text unit generation rules contained in the security grammar specification constructed according to the predefined information extraction structured framework. The structured data generation method provided by the embodiment of the present application introduces the security grammar specification in the process of generating structured data using the large model. Since the security grammar specification is constructed according to the information extraction structured framework (the information extraction structured framework indicates the structure of the data to be generated), it contains text unit generation rules related to the structure indicated by the information extraction structured framework. Therefore, under the constraints of the text unit generation rules contained in the security grammar specification, it can ensure that the large model generates structured data that conforms to the information extraction structured framework, that is, the structured data generation method provided by the present application can stably generate accurate structured data. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0042] Figure 1 A schematic diagram of a system architecture involved in this application;

[0043] Figure 2 A schematic diagram of the hardware structure of a terminal provided in an embodiment of the present application;

[0044] Figure 3 A schematic diagram of the hardware structure of a server provided in an embodiment of the present application;

[0045] Figure 4 A flow chart of a method for generating structured data according to an embodiment of the present application;

[0046] Figure 5 A schematic diagram of an embodiment of the present application using a large model to process target text data into structured data that conforms to a structured information extraction framework under the constraints of text unit generation rules contained in a security grammar specification;

[0047] Figure 6 A schematic diagram of the output format of structured data that complies with the structured framework for entity extraction provided in an embodiment of the present application;

[0048] Figure 7A schematic diagram of the output format of structured data that complies with the structured framework for entity relationship extraction provided in an embodiment of the present application;

[0049] Figure 8 Schematic diagram of the output format of structured data that conforms to the structured framework for event extraction provided in an embodiment of the present application;

[0050] Figure 9 A schematic diagram of a process for generating text units one by one using a large model according to target text data and a text unit table, under the constraints of text unit generation rules included in a security grammar specification, provided in an embodiment of the present application;

[0051] Figure 10 An example of predicting the next text unit based on the generated text unit provided in the embodiment of the present application;

[0052] Figure 11 A schematic diagram of the structure of a structured data generation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The following describes the embodiments of the present application in conjunction with the accompanying drawings. The terms used in the implementation methods of the present application are only used to explain the specific embodiments of the present application and are not intended to limit the present application.

[0054] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0055] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, and this is merely a way of distinguishing the objects of the same attributes when describing them in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0056] In one possible implementation, Figure 1 As shown, the system architecture involved in this application may include a terminal 101 and a server 102, and the terminal 101 may interact with the server 102 via a network (wired network or wireless network). The server 102 may include one or more servers ( Figure 1The terminal may obtain target text data, transmit the target text data to the server via a network, and the server may structure the target text data using the data structuring method provided in the present application to obtain structured data corresponding to the target text data.

[0057] Next, the product form of the above-mentioned terminal will be described.

[0058] The above-mentioned terminal can be a mobile phone, a tablet computer, a wearable device, an in-vehicle device, a robot, an augmented reality (AR) / virtual reality (VR) device, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), etc., and the embodiments of the present application do not impose any restrictions on this.

[0059] Figure 2 A schematic diagram of an optional hardware structure of a terminal is shown.

[0060] refer to Figure 2 As shown, the terminal may include a radio frequency unit 210, a memory 220, an input unit 230, a display unit 240, a camera 250 (optional), an audio circuit 260 (optional), a speaker 261 (optional), a microphone 262 (optional), an earphone jack 263 (optional), a processor 270, an external interface 280, a power supply 290 and other components. Those skilled in the art will understand that Figure 2 This is merely an example of a terminal and does not constitute a limitation on the terminal. The terminal may include more or fewer components than shown in the figure, or a combination of certain components, or different components.

[0061] The input unit 230 can be used to receive input digital or character information and generate key signal input related to the user settings and function control of the terminal. Specifically, the input unit 230 may include a touch screen 231 (optional) and / or other input devices 232. The touch screen 231 can collect user touch operations on or near it (for example, operations performed on or near the touch screen using a finger, joint, stylus, or any other suitable object) and drive corresponding connected devices according to a pre-set program. The touch screen can detect user touch actions on the touch screen, convert the touch actions into touch signals and transmit them to the processor 270, and can receive and execute commands sent by the processor 270; the touch signals include at least touch point coordinate information. The touch screen 231 can provide an input interface and an output interface between the terminal and the user. In addition, touch screens can be implemented using various types, such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch screen 231, the input unit 230 may also include other input devices. Specifically, the other input devices 232 may include, but are not limited to, one or more of a physical keyboard, function keys (such as a volume control button, a switch button, etc.), a trackball, a mouse, a joystick, and the like.

[0062] The display unit 240 may be used to display information input by a user or information provided to a user, various menus of the terminal, an interactive interface, file display, and / or playback of any multimedia file.

[0063] Memory 220 can be used to store instructions and data. Memory 220 primarily includes an instruction storage area and a data storage area. The data storage area can store various data, such as multimedia files and text. The instruction storage area can store software units such as the operating system, applications, and instructions required for at least one function, or subsets or extensions thereof. It may also include non-volatile random access memory (RAM). It provides processor 270 with management functions for the hardware, software, and data resources within the computing and processing device, supporting control software and applications. It is also used to store multimedia files and running programs and applications.

[0064] The processor 270 is the terminal's control center, connecting all components of the terminal using various interfaces and circuits. By executing instructions stored in memory 220 and accessing data stored in memory 220, it performs various terminal functions and processes data, thereby providing overall control of the terminal. Optionally, the processor 270 may include one or more processing units. Preferably, the processor 270 integrates an application processor and a modem processor, with the application processor primarily processing the operating system, user interface, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into the processor 270. In some embodiments, the processor and memory may be implemented on a single chip; in other embodiments, they may be implemented on separate chips. The processor 270 may also generate corresponding operational control signals and send them to corresponding components of the computing and processing device. It may also read and process data in the software, particularly the data and programs in memory 220, to enable the various functional modules therein to perform their corresponding functions, thereby controlling the corresponding components to operate as instructed.

[0065] Among them, the memory 220 can be used to store software codes related to the data structuring method, the processor 270 can execute the software codes in the memory 220, and can also schedule other units (such as the above-mentioned input unit 230 and display unit 240) to implement corresponding functions.

[0066] The RF unit 210 (optional) can be used to send and receive information or receive and send signals during calls. For example, it receives downlink information from the base station and passes it to the processor 270 for processing; in addition, it sends uplink data to the base station. Typically, the RF unit 210 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier (LNA), a duplexer, and the like. Furthermore, the RF unit 210 can communicate with network devices and other devices via wireless communication. This wireless communication can utilize any communication standard or protocol, including but not limited to Global System of Mobile Communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), and the like.

[0067] In the embodiment of the present application, the radio frequency unit 210 can send data to other devices and can also receive data sent by other devices. It should be understood that the radio frequency unit 210 is optional and can be replaced by other communication interfaces, such as a network port.

[0068] The terminal also includes a power supply 290 (such as a battery) for supplying power to various components. Preferably, the power supply can be logically connected to the processor 270 through a power management system, thereby implementing functions such as charging, discharging, and power consumption management through the power management system.

[0069] The terminal further includes an external interface 280 , which may be a standard Micro USB interface or a multi-pin connector, and may be used to connect the terminal to other devices for communication, or to connect a charger for charging the terminal.

[0070] Although not shown, the terminal may further include a flashlight, a wireless fidelity (WiFi) module, a Bluetooth module, sensors with different functions, etc., which will not be described in detail here.

[0071] Next, the product form of the above server will be described.

[0072] Figure 3 A structural diagram of the above server is provided, such as Figure 3 As shown, the server may include a bus 301 , a processor 302 , a communication interface 303 , and a memory 304 . The processor 302 , the memory 304 , and the communication interface 303 communicate with each other via the bus 301 .

[0073] The bus 301 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0074] The processor 302 may be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0075] The memory 304 may include volatile memory, such as random access memory (RAM). The memory 304 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0076] The memory 304 may be used to store software codes related to the data structuring method. The processor 302 may call the software codes stored in the memory 304 or schedule other units to implement corresponding functions.

[0077] The processors in the above-mentioned terminals and servers (such as processor 270 and processor 302) can be hardware circuits (such as application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), general-purpose processors, digital signal processors (DSP), microprocessors or microcontrollers, etc.), or a combination of these hardware circuits. For example, the processor can be a hardware system with an instruction execution function, such as a CPU, DSP, etc., or a hardware system without an instruction execution function, such as an ASIC, FPGA, etc., or a combination of the above-mentioned hardware systems without an instruction execution function and hardware systems with an instruction execution function.

[0078] In order to use the big model to generate structured data stably and accurately, the inventors of this case conducted research. The initial idea was: obtain a predefined prompt template, where the prompt template includes a structured template (such as a structured template in JSON format), and the prompt template is used to indicate to the big model the structure of the data to be generated; organize the predefined prompt template and the target text data into a prompt instruction prompt; input the prompt instruction prompt into the big model to obtain the structured data output by the big model (such as structured data in Json format).

[0079] The inventors of this case have studied the above-mentioned structured data generation method and found that it is not stable enough. That is, it can sometimes generate accurate structured data, but sometimes it cannot. Taking the generation of structured data in JSON format as an example, the above-mentioned structured data generation method sometimes results in missing brackets (such as missing curly brackets {} or square brackets []), mixing symbols (such as mixing single quotes '' with double quotes ""), missing key fields, and structural errors under specific nesting conditions. The occurrence of these situations can cause the final structured data to not meet user expectations.

[0080] To address the issues with the aforementioned structured data generation methods, the inventors of this application continued their research and ultimately proposed a more effective structured data generation method that can stably generate accurate structured data. The structured data generation method provided by this application is now described through the following examples.

[0081] See also Figure 4 , which shows a flow chart of a structured data generation method provided in an embodiment of the present application. The structured data generation method may include:

[0082] Step S401: Acquire target text data.

[0083] The target text data may be data that needs to be structured in any scenario.

[0084] Step S402: Using the large model, based on the target text data, under the constraints of the text unit generation rules contained in the security grammar specification constructed according to the pre-defined information extraction structured framework, generate structured data that conforms to the information extraction structured framework, and obtain structured data corresponding to the target text data.

[0085] In this embodiment, the information extraction structured framework is used to indicate the data to be extracted and the organizational form of the data extraction results, that is, the information extraction structured framework indicates the structure of the data to be generated.

[0086] The security grammar specification in this embodiment is constructed based on the information extraction structured framework, which includes text unit generation rules related to the structure indicated by the information extraction structured framework. The security grammar specification is used to constrain the large model to generate structured data that conforms to the information extraction structured framework, that is, the security grammar specification constrains the large model to generate data that conforms to the expected structure.

[0087] like Figure 5 As shown, this embodiment utilizes a large model and, under the constraints of security grammar specifications, can process target text data into structured data that conforms to the information extraction structured framework.

[0088] The structured data generation method provided by the embodiment of the present application, after obtaining the target text data, can use the big model to generate structured data that conforms to the information extraction structured framework based on the target text data under the constraints of the text unit generation rules contained in the security grammar specification constructed according to the pre-defined information extraction structured framework. The structured data generation method provided by the embodiment of the present application introduces the security grammar specification in the process of generating structured data using the big model. Since the security grammar specification is constructed according to the information extraction structured framework (the information extraction structured framework indicates the structure of the data to be generated), it contains text unit generation rules related to the structure indicated by the information extraction structured framework. Therefore, under the constraints of the text unit generation rules contained in the security grammar specification, it can ensure that the big model generates structured data that conforms to the information extraction structured framework. That is, the structured data generation method provided by the embodiment of the present application can stably generate accurate structured data.

[0089] In another embodiment of the present application, the information extraction structured framework in the above embodiment is introduced.

[0090] The information extraction structured framework may include: a structured framework for entity extraction, a structured framework for entity relationship extraction, and a structured framework for event extraction.

[0091] Taking the generation of structured data in JSON format as an example, the structured framework for entity extraction can be a first array object, which includes several first dictionary objects. Each first dictionary object represents the structured information of an entity and includes keywords for entity extraction. Table 1 below shows an example of a structured framework for entity extraction.

[0092] Table 1 Example of a structured framework for entity extraction

[0093]

[0094] Each curly bracket {} in Table 1 is a first dictionary object. Each first dictionary object represents the structured information of an entity. Each first dictionary object includes two keywords "label" and "value".

[0095] The structured framework for entity relationship extraction is a second array object, which includes a plurality of second dictionary objects. Each second dictionary object represents the structured information of an entity relationship, and each second dictionary object includes keywords for entity relationship extraction. Table 2 below shows an example of the structured framework for entity extraction.

[0096] Table 2 Example of a structured framework for entity extraction

[0097]

[0098]

[0099] Each curly brace {} in Table 2 represents a second dictionary object. Each second dictionary object represents the structured information of an entity relationship and includes the keywords "subject" (main entity), "relation" (relationship between entities), "object" (object entity), and "condition". Note that "condition" is an optional keyword and not every second dictionary object includes it. "Subject" and "object" are the entities at the two ends of the relationship, "relation" represents the relationship between two entities, and "condition" is supplementary constraint information.

[0100] The structured framework for event extraction is a third array object, which includes several third dictionary objects. Each third dictionary object represents the structured information of an event, includes keywords for event extraction, and includes several sub-dictionary objects. Table 3 below shows an example of the structured framework for event extraction.

[0101] Table 3 Example of a structured framework for event extraction

[0102]

[0103]

[0104] Each outermost curly brace {} in Table 3 represents a third dictionary object. Each third dictionary object represents the structured information of an event and includes the keywords "trigger," "type," and "arguments." "Arguments" includes several sub-dictionary objects. It should be noted that the trigger is the keyword or phrase that triggers the event, the event type is the category of the event, and describes the core nature of the event. Event arguments are the participants or related elements involved in the event.

[0105] In another embodiment of the present application, the security grammar specification in the above embodiment is introduced.

[0106] The security grammar specification in this application is constructed based on the above-mentioned information extraction structured framework. The security grammar specification includes several text unit generation rules related to the structure defined by the information extraction structured framework. Structured data is generated under the constraints of the security grammar specification.

[0107] Assuming that the information extraction structured framework is a structured framework in Json format, the security grammar specification constructed based on the above information extraction structured framework may include the following text unit generation rules:

[0108] (1)[→{|\n|\t

[0109] (2)]→}|\n|\t

[0110] (3){→"label"|"trigger"|"subject"|\n|\t

[0111] (4)}→,|]|\n|\t

[0112] (5):→"|[|\n|\t

[0113] (6)"→"|,|}|LLM|\n|\t

[0114] (7),→{|keywords|\n|\t

[0115] (8) keywords →: |\n|\t

[0116] (9)LLM→"|LLM|\n|\t

[0117] Among them: <→> means that the text unit (token) on the left can be forcibly normalized to generate the text unit (token) on the right. For example, rule (1) is that the text unit <[> is forced to be normalized to generate the text unit <{>, <\n> or <\t>; <|> means "or", for example, rule (2) is that the text unit <]> is forced to be normalized to generate the text unit <}>, <\n> or <\t>. <keywords>Indicates keywords involved in the information extraction structured framework, such as the aforementioned "label", "value", "subject", "relation", "object", "condition", "trigger", etc.; <"> is a double quote in English, and single quotes and Chinese quotes are not considered here; <llm>Represents a text unit generated by the large model (derived from the basic text unit table). It should be noted that each text unit (token) can be followed by <\n> or <\t>. This does not affect the semantics, but only the aesthetics of the structure.

[0118] In order to enable the large model to process the target data into data of the expected structure, in addition to introducing the security grammar specification, this application also introduces a supplementary text unit table (supplementary token table) based on the basic text unit table (basic token table). The basic text unit table (basic token table) is the token table based on which conventional data is generated. The supplementary text unit table is a structure-related token table constructed by this application based on the information extraction structured framework. The supplementary text unit table includes the keywords and symbols involved in the information extraction structured framework. Table 4 below shows an example of a supplementary text unit table constructed based on the information extraction structured framework in Json format.

[0119] Table 4 Supplementary Text Unit Table

[0120]

[0121] The supplementary text unit table shown in Table 4 includes two parts. One part is the symbols involved in the information extraction structured framework in Json format, where "\n" represents a line break and "\t" represents a space, both of which are invisible symbols. The other part is the keywords involved in the information extraction structured framework in Json format.

[0122] In another embodiment of the present application, the specific implementation process of "Step S402: using a large model, based on the target text data, under the constraints of the text unit generation rules contained in the security grammar specification constructed according to the pre-defined information extraction structured framework, generate structured data that conforms to the information extraction structured framework, and obtain structured data corresponding to the target text data" in the above embodiment is introduced.

[0123] It should be noted that, as for the data output format, it is expected that the large model will output flattened structured data (ie, the structured data that conforms to the information extraction structured framework will be flattened and output).

[0124] Take the task of generating structured data in Json format as an example: for structured data that conforms to the structured framework of entity extraction, it is expected to be Figure 6 Output in the form shown, such as Figure 6 As shown, the output data (entity Json) consists of multiple entity objects (EntityObject) and symbols. The entity object (EntityObject) consists of a series of characters and values. Among them, the nodes with gray background represent the content freely generated by the expected large model according to the context semantics (derived from the basic text unit table), and the nodes with white background represent the content generated by decoding according to the security grammar specification (derived from the supplementary text unit table); for structured data that conforms to the structured framework of entity relationship extraction, it is expected to be Figure 7 Output in the form shown, such as Figure 7 As shown in the figure, the output data (entity relationship Json) consists of multiple relationship objects (RelationObject) and symbols. The relationship object (RelationObject) consists of a series of characters and values. Among them, the nodes with gray background represent the content freely generated by the expected large model according to the context semantics (derived from the basic text unit table), and the nodes with white background represent the content generated by decoding according to the security grammar specification (derived from the supplementary text unit table); for structured data that conforms to the structured framework of event extraction, it is expected to be Figure 8 Output in the form shown, such as Figure 8 As shown, the output data (event Json) consists of multiple event objects (EventObject) and symbols. The event object (EventObject) consists of a series of characters and values. Among them, the nodes with gray background represent the content freely generated by the expected large model according to the context semantics (derived from the basic text unit table), the nodes with gray background represent the content generated by decoding according to the security grammar specification (derived from the supplementary text unit table), and the nodes with black background represent data with the same structure as the entity Json, which is a nested structure.

[0125] In order to obtain the desired output, in one possible implementation method, a large model is used to generate structured data that conforms to the information extraction structured framework based on the target text data, under the constraints of the text unit generation rules contained in the security grammar specification constructed according to the pre-defined information extraction structured framework. The process of obtaining structured data corresponding to the target text data may include: using the large model, based on the target text data and the text unit summary table, under the constraints of the text unit generation rules contained in the security grammar specification, generating text units one by one, the sequence composed of the generated text units is the structured data that conforms to the information extraction structured framework, that is, the structured data corresponding to the target text data.

[0126] The text unit summary table includes the above-mentioned basic text unit table (basic token table) and the above-mentioned supplementary text unit table (supplementary token table).

[0127] Next, we will further introduce the implementation process of "using the large model to generate text units one by one based on the target text data and the text unit table, under the constraints of the text unit generation rules contained in the security grammar specification."

[0128] In one possible implementation, Figure 9 As shown, using the large model, based on the target text data and the text unit table, under the constraints of the text unit generation rules contained in the security grammar specification, the process of generating text units one by one may include:

[0129] Step S901: extract features from target text data using a large model to obtain features of the target text data.

[0130] Step S902: Generate the following steps in each text unit using the large model:

[0131] Step S9021: Based on the characteristics of the target text data and the generated text units, predict the probability that the next text unit is each text unit in the text unit table, and obtain the probability distribution of the next text unit in the text unit table.

[0132] In the t-th text unit generation step, based on the characteristics of the target text data and the t-1 generated text units, the probability of the t-th text unit being each text unit in the text unit table is predicted, resulting in the probability distribution of the t-th text unit in the text unit table. It should be noted that for the first text unit generation step, since no text units have yet been generated, the probability of the first text unit being each text unit in the text unit table is predicted based on the characteristics of the target text data.

[0133] For example, Figure 10 As shown in the figure, text units such as {, "trigger" have been generated. In the current text unit generation step, the probability of the next text unit being each text unit in the text unit total table (including text units such as "hello", ...{, ..., "type", ..., "label", etc.) is predicted based on the characteristics of the target text data and the generated text units, and the probability distribution of the next text unit on the text unit total table is obtained.

[0134] Step S9022: Under the constraints of the text unit generation rules contained in the security grammar specification, the next text unit is determined according to the probability distribution of the next text unit in the text unit total table to obtain the text unit generated in the current step.

[0135] Specifically, under the constraints of the text unit generation rules included in the security grammar specification, the process of determining the next text unit according to the probability distribution of the next text unit in the text unit table and obtaining the text unit generated in the current step may include:

[0136] Step S9022-1: Filter text units that meet the text unit generation rules included in the security grammar specification from the text unit master table to obtain a plurality of first candidate text units.

[0137] Specifically, the process of screening text units that comply with the text unit generation rules contained in the security grammar specification from the text unit summary table to obtain several first candidate text units may include: for each text unit in the text unit summary table, judging whether the text unit can be generated after the last text unit that has been generated according to the text unit generation rules contained in the security grammar specification; if it is determined that the text unit can be generated after the last text unit that has been generated, determining that the text unit complies with the security grammar specification; if the text unit complies with the security grammar specification, determining the text unit as the first candidate text unit.

[0138] In the t-th text unit generation step, for each text unit in the text unit master table, according to the text unit generation rules contained in the security grammar specification, it is determined whether the text unit can be generated after the t-1-th text unit. If it is determined that the text unit can be generated after the t-1-th text unit, then it is determined that the text unit meets the security grammar specification and the text unit is determined as the first candidate text unit.

[0139] For example, the t-1th text unit is Figure 10 For a text unit "type" in the text unit table, according to the rule (7) in the above security grammar specification, the text unit <,> is forced to be normalized to generate the text unit <{>, <keywords>, <\n> or <\t>, since the text unit "type" is <keywords>(i.e., the text unit "type" is a keyword), therefore, it is determined that the text unit "type" can be generated after the t-1th text unit, and further, the text unit "type" complies with the security grammar specification. Figure 10 Text units with a medium-gray background are text units that comply with security grammar specifications.

[0140] Step S9022-2: Obtain the largest K probabilities from the probability distribution of the next text unit in the text unit total table, determine the text units corresponding to the largest K probabilities as second candidate text units, and obtain K second candidate text units.

[0141] Here, K is an integer greater than 0, and the specific value of K can be determined according to the actual application scenario.

[0142] Step S9022-3: Determine the next text unit from the plurality of first candidate text units and K second candidate text units to obtain the text unit generated in the current step.

[0143] Specifically, the process of determining the next text unit from several first candidate text units and K second candidate text units to obtain the text unit generated in the current step may include: determining the intersection of several first candidate text units and K second candidate text units to obtain several third candidate text units; obtaining the probabilities corresponding to several third candidate text units from the probability distribution of the next text unit in the text unit total table; determining the next text unit from several third candidate text units based on the probabilities corresponding to the several third candidate text units to obtain the text unit generated in the current step.

[0144] When determining the next text unit from several third candidate text units based on the probabilities corresponding to the several third candidate text units, the third candidate text unit corresponding to the maximum probability among the probabilities corresponding to the several third candidate text units can be determined as the next text unit, that is, the text unit generated in the current step.

[0145] The structured data generation method provided in this application optimizes the decoding process of the large model. During the decoding process of the large model, a security grammar specification constructed according to a predefined information extraction structured framework (which includes a structured framework for entity extraction, a structured framework for entity relationship extraction, and a structured framework for event extraction) is introduced. The introduction of the security grammar specification can constrain the large model to generate structured information that conforms to the information extraction structured framework. In order to generate structured information that conforms to the information extraction structured framework, in addition to introducing the security grammar specification, this application also introduces a supplementary text unit table based on the basic text unit table (the supplementary text unit table includes keywords and symbols involved in the information extraction structured framework). The introduction of the supplementary text unit table and the security grammar specification enables the large model to have a fully controllable and stable generation function, that is, it can stably generate correct structured information.

[0146] The above describes a structured data generation method provided by an embodiment of the present application. The following describes an apparatus for executing the structured data generation method.

[0147] See also Figure 11 , Figure 11 This is a schematic diagram of a structured data generating device provided in an embodiment of the present application. Figure 11 As shown, the structured data generating apparatus may include: a text data acquiring module 1101 and a text data structuring module 1102 .

[0148] The text data acquisition module 1101 is used to acquire target text data;

[0149] The text data structuring module 1102 is used to utilize a large model to generate structured data that conforms to the information extraction structuring framework based on the target text data, under the constraints of the text unit generation rules contained in the security grammar specification constructed according to the pre-defined information extraction structuring framework, and obtain structured data corresponding to the target text data.

[0150] The information extraction structured framework is used to indicate the data to be extracted and the organizational form of the data extraction results, and the text unit generation rules contained in the security grammar specification are text unit generation rules related to the structure indicated by the information extraction structured framework.

[0151] In a possible implementation, the information extraction structured framework includes: a structured framework for entity extraction, a structured framework for entity relationship extraction, and a structured framework for event extraction.

[0152] In a possible implementation, the structured framework of entity extraction is a first array object, the first array object includes a plurality of first dictionary objects, each first dictionary object represents structured information of an entity, and each first dictionary object includes keywords for entity extraction;

[0153] The structured framework of entity relationship extraction is a second array object, the second array object includes a plurality of second dictionary objects, each second dictionary object represents structured information of an entity relationship, and each second dictionary object includes keywords for entity relationship extraction;

[0154] The structured framework of event extraction is a third array object, which includes several third dictionary objects. Each third dictionary object represents the structured information of an event. Each third dictionary object includes keywords for event extraction, and each third dictionary object includes several sub-dictionary objects.

[0155] In one possible implementation, the text data structuring module 1102 generates structured data that conforms to the information extraction structuring framework using the large model based on the target text data and under the constraints of the text unit generation rules contained in the security grammar specification constructed based on the predefined information extraction structuring framework. Specifically, it is used to:

[0156] Using the large model, according to the target text data and the text unit table, under the constraints of the text unit generation rules contained in the security grammar specification, text units are generated one by one. The sequence composed of the generated text units is structured data that conforms to the information extraction structured framework.

[0157] Among them, the text unit summary table includes a basic text unit table and a supplementary text unit table. The supplementary text unit table is a structure-related text unit table constructed according to the information extraction structured framework. The supplementary text unit table includes keywords and symbols involved in the information extraction structured framework.

[0158] In one possible implementation, the text data structuring module 1102 utilizes the large model to generate text units one by one based on the target text data and the text unit table, under the constraints of the text unit generation rules included in the security grammar specification, specifically for:

[0159] Use the large model to extract features from the target text data to obtain the features of the target text data;

[0160] Use the large model to generate steps in each text unit:

[0161] Based on the characteristics of the target text data and the generated text units, predict the probability that the next text unit will be each text unit in the text unit table, and obtain the probability distribution of the next text unit on the text unit table;

[0162] Under the constraints of the text unit generation rules contained in the security grammar specification, the next text unit is determined according to the probability distribution of the next text unit on the text unit total table, and the text unit generated in the current step is obtained.

[0163] In one possible implementation, the text data structuring module 1102 utilizes the large model, under the constraints of the text unit generation rules included in the security grammar specification, and determines the next text unit based on the probability distribution of the next text unit in the text unit master table. When obtaining the text unit generated in the current step, the module specifically:

[0164] Screening text units that meet the text unit generation rules included in the security grammar specification from the text unit total table to obtain a plurality of first candidate text units;

[0165] Obtaining the largest K probabilities from the probability distribution of the next text unit on the text unit total table, determining the text units corresponding to the largest K probabilities as second candidate text units, and obtaining K second candidate text units, where K is an integer greater than 0;

[0166] From a number of first candidate text units and K second candidate text units, the next text unit is determined to obtain the text unit generated in the current step.

[0167] In one possible implementation, when the text data structuring module 1102 uses the large model to screen text units that meet the text unit generation rules included in the security grammar specification from the text unit table to obtain a plurality of first candidate text units, it is specifically configured to:

[0168] For each text unit in the text unit summary table:

[0169] Determining whether the text unit can be generated after the last generated text unit according to the text unit generation rules included in the security grammar specification;

[0170] If it is determined that the text unit can be generated after the last text unit that has been generated, then the text unit is determined to comply with the security grammar specification;

[0171] If the text unit complies with the security grammar specification, the text unit is determined as the first candidate text unit.

[0172] In one possible implementation, when the text data structuring module 1102 uses the large model to determine the next text unit from the plurality of first candidate text units and K second candidate text units and obtains the text unit generated in the current step, it is specifically configured to:

[0173] Determine the intersection of a plurality of first candidate text units and the K second candidate text units to obtain a plurality of third candidate text units;

[0174] Obtaining probabilities corresponding to a plurality of third candidate text units from the probability distribution of the next text unit on the text unit total table;

[0175] According to the respective probabilities corresponding to the plurality of third candidate text units, the next text unit is determined from the plurality of third candidate text units to obtain the text unit generated in the current step.

[0176] The structured data generation device provided by the embodiment of the present application can, after obtaining the target text data, use the big model to generate structured data that conforms to the information extraction structured framework based on the target text data under the constraints of the text unit generation rules contained in the security grammar specification constructed according to the predefined information extraction structured framework. The structured data generation device provided by the embodiment of the present application introduces the security grammar specification in the process of generating structured data using the big model. Since the security grammar specification is constructed according to the information extraction structured framework (the information extraction structured framework indicates the structure of the data to be generated), it contains text unit generation rules related to the structure indicated by the information extraction structured framework. Therefore, under the constraints of the text unit generation rules contained in the security grammar specification, it can ensure that the big model generates structured data that conforms to the information extraction structured framework. That is, the structured data generation device provided by the embodiment of the present application can stably generate accurate structured data.

[0177] An embodiment of the present application also provides an electronic device, which may include: at least one processor, at least one communication interface, at least one memory and at least one communication bus.

[0178] In the embodiment of the present application, the number of the processor, the communication interface, the memory, and the communication bus is at least one, and the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0179] The processor may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application;

[0180] The memory may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory;

[0181] The memory stores a program, and the processor can call the program stored in the memory, where the program is used to implement the steps of the structured data generation method provided in the above embodiment.

[0182] An embodiment of the present application also provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the steps of the structured data generation method provided in the above embodiment.

[0183] An embodiment of the present application also provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the steps of the structured data generation method provided in the above embodiment.

[0184] It should also be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided in this application, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines.

[0185] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course can also be implemented by special hardware including application-specific integrated circuits, special CPUs, special memories, special components, etc. In general, all functions performed by computer programs can be easily implemented with corresponding hardware, and the specific hardware structures used to implement the same function can also be diverse, such as analog circuits, digital circuits or special circuits, etc. However, for the present application, software program implementation is a better implementation method in most cases. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer's floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., and includes a number of instructions to enable a computer device (which can be a personal computer, training equipment, or network equipment, etc.) to execute the methods described in each embodiment of the present application.

[0186] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0187] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, a computer, a training device or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website, a computer, a training device or a data center. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium can be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).< / keywords> < / keywords> < / llm> < / keywords>

Claims

1. A method for generating structured data, characterized in that: include: Get target text data; Utilizing a large model, based on the target text data, and under the constraints of the text unit generation rules contained in the security grammar specification constructed based on a pre-defined information extraction structured framework, structured data that conforms to the information extraction structured framework is generated to obtain structured data corresponding to the target text data; wherein, the information extraction structured framework is used to indicate the data to be extracted and the organizational form of the data extraction results, and the text unit generation rules contained in the security grammar specification are text unit generation rules related to the structure indicated by the information extraction structured framework.

2. The structured data generation method according to claim 1, wherein: The information extraction structured framework includes: a structured framework for entity extraction, a structured framework for entity relationship extraction, and a structured framework for event extraction.

3. The structured data generation method according to claim 2, characterized in that: The structured framework of the entity extraction is a first array object, the first array object includes a plurality of first dictionary objects, each first dictionary object represents the structured information of an entity, and each first dictionary object includes keywords for entity extraction; The structured framework of the entity relationship extraction is a second array object, the second array object includes a plurality of second dictionary objects, each second dictionary object represents the structured information of an entity relationship, and each second dictionary object includes keywords for entity relationship extraction; The structured framework of the event extraction is a third array object, which includes several third dictionary objects. Each third dictionary object represents the structured information of an event. Each third dictionary object includes keywords for event extraction, and each third dictionary object includes several sub-dictionary objects.

4. The structured data generation method according to any one of claims 1 to 3, characterized in that: The method of utilizing the large model to generate structured data that conforms to the information extraction structured framework based on the target text data and under the constraints of text unit generation rules contained in the security grammar specification constructed based on the predefined information extraction structured framework includes: Using the large model, based on the target text data and the text unit table, and under the constraints of the text unit generation rules contained in the security grammar specification, text units are generated one by one, and the sequence composed of the generated text units is structured data that conforms to the information extraction structured framework; Among them, the text unit summary table includes a basic text unit table and a supplementary text unit table. The supplementary text unit table is a structure-related text unit table constructed according to the information extraction structured framework. The supplementary text unit table includes keywords and symbols involved in the information extraction structured framework.

5. The structured data generation method according to claim 4, characterized in that: The method of utilizing the large model to generate text units one by one based on the target text data and the text unit table and under the constraints of the text unit generation rules contained in the security grammar specification includes: Extracting features from the target text data using the large model to obtain features of the target text data; Use the large model to generate steps in each text unit: Based on the characteristics of the target text data and the generated text units, predicting the probability that the next text unit will be each text unit in the text unit table, and obtaining the probability distribution of the next text unit on the text unit table; Under the constraints of the text unit generation rules included in the security grammar specification, the next text unit is determined according to the probability distribution of the next text unit on the text unit master table to obtain the text unit generated in the current step.

6. The structured data generation method according to claim 5, characterized in that: The step of determining the next text unit according to the probability distribution of the next text unit on the text unit master table under the constraints of the text unit generation rules included in the security grammar specification to obtain the text unit generated in the current step includes: Screening text units that meet the text unit generation rules included in the security grammar specification from the text unit master table to obtain a plurality of first candidate text units; Obtaining K maximum probabilities from the probability distribution of the next text unit on the text unit total table, determining the text units corresponding to the K maximum probabilities as second candidate text units, and obtaining K second candidate text units, where K is an integer greater than 0; The next text unit is determined from the plurality of first candidate text units and the K second candidate text units to obtain the text unit generated in the current step.

7. The structured data generation method according to claim 6, characterized in that: The step of screening the text units that meet the text unit generation rules contained in the security grammar specification from the text unit master table to obtain a plurality of first candidate text units includes: For each text unit in the text unit summary table: Determining whether the text unit can be generated after the last generated text unit according to the text unit generation rules included in the security grammar specification; If it is determined that the text unit can be generated after the last text unit that has been generated, then the text unit is determined to comply with the security grammar specification; If the text unit complies with the security grammar specification, the text unit is determined as the first candidate text unit.

8. The method for generating structured data according to claim 6, wherein: Determining the next text unit from the plurality of first candidate text units and the K second candidate text units to obtain the text unit generated in the current step includes: Determine the intersection of the plurality of first candidate text units and the K second candidate text units to obtain a plurality of third candidate text units; Obtaining the probabilities corresponding to the plurality of third candidate text units respectively from the probability distribution of the next text unit on the text unit total table; According to the respective probabilities corresponding to the plurality of third candidate text units, the next text unit is determined from the plurality of third candidate text units to obtain the text unit generated in the current step.

9. A structured data generating device, characterized in that: include: Text data acquisition module and text data structuring module; The text data acquisition module is used to acquire target text data; The text data structuring module is used to use a large model to generate structured data that conforms to the information extraction structuring framework based on the target text data, under the constraints of the text unit generation rules contained in the security grammar specification constructed according to a pre-defined information extraction structuring framework, to obtain structured data corresponding to the target text data; wherein, the information extraction structuring framework is used to indicate the data to be extracted and the organizational form of the data extraction results, and the text unit generation rules contained in the security grammar specification are text unit generation rules related to the structure indicated by the information extraction structuring framework.

10. An electronic device, characterized in that: comprising at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is configured to execute the computer program so as to enable the electronic device to implement the steps of the structured data generating method according to any one of claims 1 to 8.

11. A computer storage medium, characterized in that The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the steps of the structured data generation method according to any one of claims 1 to 8.

12. A computer program product, characterized in that The method comprises computer-readable instructions, which, when executed on an electronic device, enable the electronic device to implement the steps of the structured data generating method according to any one of claims 1 to 8.

Images