A knowledge grid-based content correction method and electronic device

Abstract

The application relates to a knowledge grid-based content correction method and electronic equipment, which comprises the following steps: receiving factual data and converting the factual data into structured triples; mapping the structured triples to a netted knowledge carrier to obtain a knowledge grid model; receiving to-be-corrected data and converting the to-be-corrected data into structured text data according to the type of the to-be-corrected data; performing factual comparison on the structured text data based on the knowledge grid model, and outputting a correction result. The beneficial effect is that the knowledge grid model obtained by mapping the factual data serves as a correction benchmark, the knowledge grid model performs global factual traversal on the to-be-corrected data, all factual points in the to-be-corrected data participate in verification, manual intervention is avoided, and the analysis efficiency is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge grid algorithm, in particular to a content correction method based on knowledge grid and electronic equipment. BACKGROUND

[0002] With the popularity of digital content production, there are often factual errors such as time, characters, numerical values, and cause-and-effect relationships in user inputted text, voice, tables and other multi-modal information. Traditional content proofreading methods mainly focus on spelling correction and grammar correction, and cannot judge the truth of facts at the semantic level. After the conversion of traditional technology, manual intervention is still needed to extract facts; the structured analysis of text data is low in efficiency and easy to miss. SUMMARY

[0003] The technical problem to be solved by the embodiments of the present application is to provide a content correction method based on knowledge grid and electronic equipment, which obtains a knowledge grid model as a correction reference through fact data mapping, and the knowledge grid model performs global fact traversal on the data to be corrected to ensure that all fact points in the data to be corrected participate in verification, avoiding manual intervention and improving analysis efficiency.

[0004] To solve the above technical problems, the present application provides a content correction method based on knowledge grid, comprising,

[0005] Receiving fact data and converting the fact data into structured triples;

[0006] Mapping the structured triples to a mesh knowledge carrier to obtain a knowledge grid model;

[0007] Receiving data to be corrected and converting it into structured text data according to the type of the data to be corrected;

[0008] Based on the knowledge grid model, the structured text data is subjected to fact comparison, and the correction result is outputted.

[0009] In a feasible embodiment, the receiving fact data and converting the fact data into structured triples comprises,

[0010] Extracting the basic data in the fact data;

[0011] Completing the basic data to obtain completed data;

[0012] Optimizing the basic data and the completed data through a large language model to output the structured triples.

[0013] In a feasible embodiment, the extracting the basic data in the fact data comprises,

[0014] An entity and a type in the fact data are labeled by a Chinese entity recognition model;

[0015] A binary relationship between entities is obtained by a relationship extraction model; wherein the basic data comprises the entities and the binary relationships.

[0016] In a feasible embodiment, the supplementing of the basic data to obtain the completed data comprises,

[0017] The missing data in the basic data is completed by using a preset regular template;

[0018] The missing data in the basic data is verified by a business dictionary to obtain the completed data.

[0019] In a feasible embodiment, the structured triplets comprise entities, binary relationships, attribute values, character position indexes, and query weights.

[0020] The mesh knowledge carrier comprises nodes, edges, and hyper-edges.

[0021] The nodes correspond to entities and attribute values of the structured triplets, the edges correspond to binary relationships of the structured triplets, and the hyper-edges connect at least three nodes.

[0022] In a feasible embodiment, the receiving of the to-be-corrected data and the conversion of the to-be-corrected data into structured text data according to a type of the to-be-corrected data comprises,

[0023] When the type of the to-be-corrected data is voice data, the voice data is converted into text data by voice recognition, a large language model is called to extract fact points from the text data, and the structured text data is obtained; or,

[0024] When the type of the to-be-corrected data is the text data, the large language model is directly called to extract fact points from the text data, and the structured text data is obtained.

[0025] In a feasible embodiment, the fact comparison of the structured text data based on the knowledge grid model to obtain a correction result and the outputting of the correction result comprise,

[0026] The fact comparison is performed by global multi-hop traversal based on the knowledge grid model, bidirectional breadth-first search is performed from the fact data as a starting point, wherein the fact weight is multiplied by a preset attenuation coefficient every time the fact comparison is performed outward by one hop, and the fact comparison is ended when all the nodes are traversed or the number of hops of the fact comparison reaches a maximum value.

[0027] In an example embodiment, the fact data comprises one or more of a portable document, a word processing software document, a presentation, a spreadsheet document, a plain text document, a markup language text document, and a simple information aggregation format document, and the fact data is transmitted via one or more of a hypertext transfer protocol, an Amazon cloud storage protocol, or a file transfer protocol.

[0028] In an example embodiment, after the fact comparison based on the knowledge grid model is performed on the structured text data to obtain a correction result, the knowledge grid-based content correction method further comprises scoring the to-be-corrected data by the following formula and outputting,

[0029] ;

[0030] wherein, represents scoring of the to-be-corrected data;

[0031] represents a scoring coefficient;

[0032] represents the fact weight;

[0033] represents a confidence decay coefficient, ;

[0034] represents a hop count;

[0035] represents the query weight.

[0036] Correspondingly, the present application also relates to an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the steps of the knowledge grid-based content correction method.

[0037] The present application has the following beneficial effects:

[0038] The knowledge grid model obtained by mapping the fact data serves as a correction reference, and the knowledge grid model performs global fact traversal on the to-be-corrected data, ensuring that all fact points in the to-be-corrected data participate in verification, avoiding manual intervention, and improving analysis efficiency.

[0039] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a step diagram of one embodiment of the present application;

[0041] Figure 2 is a step diagram of one embodiment of the present application;

[0042] Figure 3 is a structural diagram of content correction based on knowledge grid of the present application;

[0043] Figure 4 is a hardware structural schematic diagram of the electronic device of the present application. DETAILED DESCRIPTION

[0044] In order to make the above objectives, features and advantages of the present application more apparent, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be practiced in a number of different ways beyond the specific embodiments described and it is therefore contemplated to cover all such modifications as fall within the scope of the application. It should be noted that the use of particular verbiage or terminology is intended to convey a specific embodiment, but is not intended to limit the present application.

[0045] It should be noted that when an element is referred to as being "on" another element, it can be directly on the other element or intervening elements can also be present. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or intervening elements can also be present. As used herein the terms "vertical", "horizontal", "left", "right", and the like are merely used for the purpose of explanation.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be understood that the terms "and / or", as used herein, refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0047] A content correction method based on knowledge grid, with reference to Figure 1 , comprising,

[0048] Step S100, receiving fact data and converting the fact data into structured triples;

[0049] Step S200, mapping the structured triples to a mesh knowledge carrier to obtain a knowledge grid model;

[0050] Step S300, receiving to-be-corrected data and converting the to-be-corrected data into structured text data according to the type of the to-be-corrected data;

[0051] Step S400, performing fact comparison on the structured text data based on the knowledge grid model, and outputting a correction result.

[0052] Exemplarily, the factual data in step S100 includes one or more of a portable document PDF, a word processing software document Word, a presentation PPT, an electronic table document Excel, a pure text document TXT, a markup language text document Markdown, and a simple information aggregation format document RSS, and the factual data is transmitted through one or more of a hypertext transfer protocol HTTP or an Amazon cloud storage protocol S3 or a file transfer protocol FTP.

[0053] Exemplarily, receiving the to-be-corrected data and converting the to-be-corrected data into structured text data according to the type of the to-be-corrected data includes,

[0054] When the type of the to-be-corrected data is voice data, the voice data is converted into text data through voice recognition, a large language model is called to extract fact points from the text data, and structured text data is obtained; or,

[0055] When the type of the to-be-corrected data is text data, a large language model is directly called to extract fact points from the text data, and structured text data is obtained. Through the above conversion, the content correction method can be compatible with multi-modal user input data of voice data and text data.

[0056] Exemplarily, taking a PDF fact document as an example, the content of the fact document is “Company A was established on May 14, 2012, the legal representative is Zhang San, and the registered capital is 10 million yuan”, the file size is 8.2 KB, and after uploading through the POST / upload interface, doc_id=1001 is returned;

[0057] The user inputs a sentence: “Company A was established in February 2010, the legal representative is Zhang San, and the registered capital is 10 million yuan”, at this time, fact triples need to be extracted for subsequent comparison.

[0058] Through a three-level pipeline of entity, relationship preliminary screening, rule supplement, and model correction, the fact document is converted into a structured triple {e, r, v, idx, w_e} containing weights and position indexes, where e represents an entity, r represents a binary relationship, v represents an attribute value, idx represents a character position index, and w_e represents a query weight; the triple includes “e”: “Company A”, “r”: “establishment date”, “v”: “2012-05-14”, “idx”: “[0, 12]”, “w_e”: 1.0,

[0059] “e”: “Company A”, “r”: “legal representative”, “v”: “Zhang San”, “idx”: “[41, 43]”, “w_e”: 1.0,

[0060] {"e":"Company A","r":"registered capital","v":"100 million yuan","idx":"[45, 52]","w_e":0.8}

[0061] In one possible embodiment, step S100, receiving factual data and converting the factual data into structured triples, referring to Figure 2 , including,

[0062] Step S101, extracting basic data in the factual data;

[0063] Step S102, supplementing the basic data to obtain completed data;

[0064] Step S103, optimizing the basic data and the completed data through a large language model to output structured triples.

[0065] Illustratively, step S101, extracting basic data in the factual data, includes,

[0066] Annotating entities and types in the factual data through a Chinese entity recognition model, and obtaining binary relationships between entities through a relationship extraction model; wherein the basic data includes entities and binary relationships.

[0067] Step S102, supplementing the basic data to obtain completed data, includes,

[0068] Using a pre-set regular template to complete the missing data in the basic data, and verifying the missing data in the basic data through a business dictionary to obtain the completed data.

[0069] Illustratively, the specific process of the three-level pipeline of triples is as follows,

[0070] First level: use spaCy NER (entity recognition) model (such as zh_core_web_sm) and open source rel_model (relationship extraction model); identify "Company A" (label ORG, start position 0, end position 12), "May 14, 2012" (label DATE, start position 16, end position 27), "Zhang San" (label PERSON, start position 35, end position 37), and "100 million yuan" (label MONEY, start position 43, end position 49) from the factual document; then perform relationship extraction to extract binary relationships such as "Company A-Founded Date-May 14, 2012", "Company A-Legal Representative-Zhang San", but there is a missed detection (such as "registered capital-1000 million yuan" is not identified).

[0071] Second level: for the fixed format facts missed in the first level (date, numerical value, negative word), complete through a pre-set regular template:

[0072] Date template: (\d{4}) year (\d{1,2}) month (\d{1,2}) day, matches date information in the document;

[0073] Registered capital template: Registered capital (\s*) (\d+(?:\.\d+)?\s*[million billion]?) yuan, complete the missing relationship of "Company A-Registered Capital-1000 million yuan";

[0074] At the same time, combined with the business dictionary (key fact dictionary such as "establishment date", "registered capital", secondary fact dictionary such as "mailbox"), ensure that the domain-specific facts are not missed.

[0075] Level 3: Merge the results of the previous two levels, call the large language model to correct logical conflicts (such as time sequence contradictions), and supplement the character position index (idx) and query weight (w_e);

[0076] Weight rule: key facts (establishment date, legal representative) default w_e=1.0, secondary facts (registered capital) default w_e=0.8, support scenario adjustment;

[0077] Final output: Structured triples with source marks (source, such as "RE" for relationship extraction, "rule" for rule missing), examples are as follows: {"e":"Company A","r":"establishment date","v":"2012-05-14","idx":"[0,12]","w_e":1.0,"source":"RE"}, {"e":"Company A","r":"legal representative","v":"Zhang San","idx":"[35,37]","w_e":1.0,"source":"RE"}, {"e":"Company A","r":"registered capital","v":"1000 million yuan","idx":"[43,49]","w_e":0.8,"source":"rule"}

[0078] The source field marks the source, which is used for subsequent tracing.

[0079] In one possible embodiment, the structured triples include a network knowledge carrier including nodes, edges, and hyperedges;

[0080] The nodes correspond to entities and attribute values of structured triples, the edges correspond to binary relationships of structured triples, and the hyperedges connect at least three nodes.

[0081] Specifically, the nodes include entities, attribute values, and events; the edges include relationships, time sequences, and causality; and the hyperedges are cross-document anaphora aggregation. Each edge carries w_e (default key fact 1.0, secondary fact 0.8, and w_e is adjustable).

[0082] The super-edge definition can be understood as a large edge that can bind 3, 4 or even N nodes at a time, while the normal edge can only connect 2 nodes.

[0083] In a feasible embodiment, step S400, based on the knowledge grid model, the structured text data is subjected to fact comparison, and the error correction result is output, including,

[0084] Based on the knowledge grid model, the global multi-hop variable traversal fact comparison is performed, and bidirectional breadth-first search is performed from the fact data as the starting point, wherein the fact comparison is multiplied by a preset attenuation coefficient for each outward jump of the fact comparison, and the fact comparison is ended when all nodes are traversed or the number of jumps of the fact comparison reaches a maximum value. The maximum value can be set as 5 by default.

[0085] Illustratively, the output of step S400 is a sentence in which the incorrect segment is replaced with the correct value while keeping the rest of the text and tone unchanged.

[0086] In a feasible embodiment, after the structured text data is subjected to fact comparison based on the knowledge grid model and the error correction result is output, the content error correction method based on the knowledge grid further includes scoring and outputting the to-be-corrected data by the following formula,

[0087] ;

[0088] wherein, the to-be-corrected data is scored;

[0089] the scoring coefficient is represented by;

[0090] the fact weight is represented by;

[0091] the confidence attenuation coefficient is represented by, ;

[0092] the number of jumps is represented by;

[0093] the query weight is represented by.

[0094] Illustratively, the final output text supports three formats of JSON, Markdown and HTML highlighting, and is accompanied by a diff highlighting script.

[0095] Referring to Figure 3, the administrator inputs the above fact data, and the user inputs the multi-modal error correction data, and then the knowledge grid is used to compare the facts of the above embodiment method. In the knowledge grid, N represents a node, R represents a relationship, and V represents an attribute value.

[0096] The method in the above embodiment eliminates multi-modal differences from the source by receiving error correction data and converting it into structured text data according to the type:

[0097] For voice error correction data: first convert the voice signal to pure text through ASR (speech recognition) technology, then call a large language model to extract entity-relation-attribute value core fact points, and eliminate redundant information, and finally output standardized structured text;

[0098] For text error correction data (including PDF / Word / Excel formats): directly convert different formats of text into entity-relation-attribute value structured text through format analysis tools; regardless of the input modality, the error correction data is converted into a unified format that can be compared by machines through type-specific conversion and structured extraction, avoiding the compatibility limitations of traditional technologies such as the need for separate tools for voice and the inability to parse special format text.

[0099] The present application parses authoritative fact data into triples containing entities, relationships, attribute values, and weights, ensuring that each authoritative fact point is accurately extracted without missing sources; organizes triples into node-edge-superedge structures, where nodes correspond to entities / attribute values, edges correspond to binary relationships, and superedges correspond to cross-document / multi-entity facts, forming a global mesh network with no isolated nodes; based on the global connectivity of the mesh carrier, the structured text of the error correction data can start from any associated node and traverse all related authoritative facts, avoiding the random missing detection problem of traditional technologies that only recall Top-k fragments, and achieving 100% fact point participation in comparison. The high-dimensional association capability of the superedge and edge of the mesh carrier directly solves the limitations of traditional binary relationships.

[0100] In an exemplary embodiment, a computer device, which can be a terminal, is provided, and its internal structure diagram can be as shown in Figure 4The computer device shown in the figure includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be realized through WIFI, mobile cellular network, near field communication (NFC) or other technologies. The computer program is executed by the processor to realize a content correction method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0101] Those skilled in the art can understand that, Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0102] In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in each of the method embodiments described above.

[0103] In one embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program is executed by a processor to implement the steps in each of the method embodiments described above.

[0104] In one embodiment, a computer program product is provided, including a computer program, and the computer program is executed by a processor to implement the steps in each of the method embodiments described above.

[0105] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0106] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of each method. In the embodiments provided in the present application, any reference to memory, database or other medium can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0107] Any combination of the technical features in the above-described embodiments can be made, and for the sake of brevity, not all possible combinations are described, however, as long as there is no conflict, any combination of the technical features should be considered within the scope of the present disclosure.

[0108] The above-described embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A content errata correction method based on knowledge grids, characterized in that, include, Receive factual data and convert the factual data into structured triples; The structured triples are mapped to a network knowledge carrier to obtain a knowledge grid model; Receive the data to be corrected and convert it into structured text data according to the type of the data to be corrected; Based on the knowledge grid model, the structured text data is compared with facts, and the errata results are output. The structured triple includes an entity, a binary relation, and an attribute value; the network knowledge carrier includes nodes, edges, and hyperedges; the nodes correspond to the entity and attribute value of the structured triple, the edges correspond to the binary relation of the structured triple, and the hyperedges connect at least three of the nodes. The hyperedges are used to associate cross-document / multi-entity facts to form a global network structure without isolated nodes.

2. The content errata correction method based on knowledge grids according to claim 1, characterized in that, The process involves receiving factual data and converting it into structured triples. include, Extract the basic data from the factual data; The basic data is supplemented to obtain the complete data; The structured triples are output after optimizing the basic data and the completed data using a large language model.

3. The content errata correction method based on knowledge grids according to claim 2, characterized in that, The extraction of basic data from the factual data includes, The entities and types in the factual data are labeled using a Chinese entity recognition model, and the binary relations between entities are obtained using a relation extraction model; wherein, the basic data includes entities and binary relations.

4. The content errata correction method based on knowledge grids according to claim 2, characterized in that, The process of filling in the missing data in the basic data to obtain the complete data includes, The missing data in the basic data is completed by using a preset regular expression template, and the missing data in the basic data is verified by a business dictionary to obtain the completed data.

5. The content errata correction method based on knowledge grids according to claim 1, characterized in that, The structured triple also includes a character position index and a query weight.

6. The content errata correction method based on knowledge grids according to claim 1, characterized in that, The process of receiving the data to be corrected and converting it into structured text data according to its type includes: When the data to be corrected is speech data, it is converted into text data through speech recognition, and a large language model is used to extract fact points from the text data to obtain the structured text data; or... When the data to be corrected is text-based, the large language model is directly invoked to extract fact points from the text-based data to obtain the structured text data.

7. The content errata correction method based on knowledge grids according to claim 5, characterized in that, The step of performing fact comparison on the structured text data based on the knowledge grid model, and outputting the resulting corrections, includes: The fact comparison is performed based on the knowledge grid model with global multi-hop transformation. The fact data is used as the starting point for bidirectional breadth-first search. Each time the fact comparison goes outward, the fact weight is multiplied by a preset decay coefficient until all the nodes are traversed or the number of hops in the fact comparison reaches the maximum value, at which point the fact comparison ends.

8. The content errata correction method based on knowledge grids according to claim 1, characterized in that, The factual data includes one or more of portable documents, word processing software documents, presentations, spreadsheet documents, plain text documents, markup language text documents, and simplified information aggregation format documents, and the factual data is transmitted via one or more of Hypertext Transfer Protocol, Amazon Web Services Protocol, or File Transfer Protocol.

9. The content errata correction method based on knowledge grids according to claim 7, characterized in that, After performing fact comparison on the structured text data based on the knowledge grid model and outputting the resulting errata, the knowledge grid-based content errata method further includes scoring the data to be erratadated using the following formula and then outputting the score. ； in, This indicates that the data to be corrected will be scored; Indicates the rating coefficient; Indicates the weight of the stated fact; This represents the confidence decay coefficient. ; Indicates the number of jumps; This indicates the query weight.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor runs the computer program stored in the memory, the processor performs the steps of the content errata method based on the knowledge grid as described in any one of claims 1-9.

Images