Object information reply method, apparatus, device, medium, and program product

Abstract

Embodiments of the present disclosure disclose object information reply methods, devices, equipment, media and program products. A specific implementation of the method includes: extracting search information from query information to obtain search information; querying a first candidate object information set corresponding to the search information from an object information knowledge graph; using a large model-based graph search enhancement generation technology, searching for a second candidate object information set with a semantic similarity satisfying a target similarity condition; and using a large language model to generate object information reply content for the query information according to the first candidate object information set and the second candidate object information set. This implementation is related to information retrieval, can break through the matching limitations by using the semantic understanding ability of the large model, and can improve the matching accuracy and efficiency by using the structured association data of the knowledge graph, to accurately obtain each candidate object information. Thus, the reply content corresponding to the object information can be accurately generated.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to object information response methods, apparatus, devices, media, and program products. Background Technology

[0002] Currently, with the continuous development of artificial intelligence, AI technology is being increasingly applied to various industries. For object lookup methods within a target industry, the common approaches are: searching through a pre-defined static object directory or conducting object searches via online consultation.

[0003] However, when using the above method, the following technical problems often arise: Static object directories require manual, periodic entry and updates. When significant changes occur to object information, timely synchronization of updates becomes difficult, and maintenance becomes challenging. Furthermore, online consultations are also inefficient and costly. Summary of the Invention

[0004] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of this disclosure provide object information response methods, apparatuses, electronic devices, computer-readable media, and program products to address the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide an object information response method, comprising: extracting retrieval information from query information representing the object information to obtain retrieval information; querying a first candidate object information set corresponding to the retrieval information from an object information knowledge graph; retrieving a second candidate object information set whose semantic similarity to the query information meets the target similarity condition using a graph retrieval enhancement generation technique based on a large model; and generating object information response content for the query information using a pre-trained large language model based on the first candidate object information set and the second candidate object information set.

[0007] Optionally, the aforementioned retrieval information includes: query intent and query keywords; and the aforementioned querying of the first candidate object information set corresponding to the aforementioned retrieval information from the object information knowledge graph includes: using the aforementioned query intent and the aforementioned query keywords as query conditions to generate a graph query statement; and executing the aforementioned graph query statement to query the first candidate object information set corresponding to the aforementioned retrieval information from the aforementioned object information knowledge graph.

[0008] Optionally, the above-mentioned method of generating object information response content for the query information using a pre-trained large language model based on the first candidate object information set and the second candidate object information set includes: sorting and filtering the first candidate object information set and the second candidate object information set according to their relevance to obtain a third candidate object information set; generating prompt words for the third candidate object information set and the query information; and inputting the prompt words into the large language model to obtain the object information response content.

[0009] Optionally, the above-mentioned object information knowledge graph is constructed through the following steps: obtaining an object information corpus for the target domain; constructing an object information knowledge graph based on the above-mentioned object information corpus, wherein the graph data corresponding to the above-mentioned object information knowledge graph is stored in a graph database through graph query statements.

[0010] Optionally, the target domain is the government affairs domain, and the object information corpus is constructed through the following steps: using a pre-trained government affairs object information extraction model, extracting multi-dimensional object information from each government affairs text in the acquired government affairs text set, wherein the object information includes: government affairs responsibility information and confidence level; extracting at least one government affairs text from the above government affairs text set whose corresponding confidence level is less than the target confidence level; sending the above at least one government affairs text to the verification terminal to generate at least one corresponding object information on the verification terminal; and constructing the object information corpus based on the above at least one object information and the obtained object information set.

[0011] Optionally, the aforementioned government affairs text set is generated through the following steps: obtaining a government affairs dataset; for each piece of government affairs data in the aforementioned government affairs dataset, performing the following generation steps: performing at least one of the following on the aforementioned government affairs data: content filtering and sensitive information desensitization to obtain government affairs processing data; integrating object information into the aforementioned government affairs processing data to obtain object information; and performing standard formatting conversion on the aforementioned government affairs processing data and the aforementioned object information to obtain government affairs text.

[0012] Secondly, some embodiments of this disclosure provide an object information response apparatus, comprising: an extraction unit configured to extract retrieval information from query information representing object information to obtain retrieval information; a query unit configured to query a first candidate object information set corresponding to the retrieval information from an object information knowledge graph; a retrieval unit configured to retrieve a second candidate object information set whose semantic similarity to the query information meets a target similarity condition using a graph retrieval enhancement generation technique based on a large model; and a generation unit configured to generate object information response content for the query information using a pre-trained large language model based on the first candidate object information set and the second candidate object information set.

[0013] Optionally, the above-mentioned retrieval information includes: query intent and query keywords; and the query unit can be configured to: use the above-mentioned query intent and the above-mentioned query keywords as query conditions to generate a graph query statement; execute the above-mentioned graph query statement to query the first candidate object information set corresponding to the above-mentioned retrieval information from the above-mentioned object information knowledge graph.

[0014] Optionally, the generation unit can be configured to: sort and filter the first candidate object information set and the second candidate object information set according to their relevance to obtain a third candidate object information set; generate prompt words for the third candidate object information set and the query information; and input the prompt words into the large language model to obtain the object information response content.

[0015] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, such that when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0016] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method as described in any implementation of the first aspect.

[0017] Fifthly, some embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0018] The above-described embodiments of this disclosure have the following beneficial effects: The object information response method of some embodiments of this disclosure can overcome matching limitations by utilizing the semantic understanding capabilities of large models, and simultaneously improve matching accuracy and efficiency by leveraging the structured association data of knowledge graphs, thereby accurately obtaining information about each candidate object. Thus, the response content corresponding to the object information can be accurately generated. Specifically, the reason for the low efficiency of querying related object information is that static object directories require manual periodic entry and updates. When a large number of object information changes, there is a problem that the updated object information cannot be synchronized in a timely manner and maintenance is difficult. In addition, online consultation methods also suffer from low efficiency and high cost. Based on this, the object information response method of some embodiments of this disclosure first extracts retrieval information from the query information representing the object information used for querying, which can accurately obtain retrieval information to determine the key retrieval content for object retrieval, facilitating subsequent object matching. Then, it queries the first candidate object information set corresponding to the above retrieval information from the object information knowledge graph. Here, through the association reasoning capabilities corresponding to the object information knowledge graph, the first candidate object information set can be initially and accurately retrieved. Furthermore, by leveraging graph retrieval enhancement generation techniques based on large models and utilizing semantic matching capabilities to overcome the limitations of keyword matching, a second set of candidate object information that semantically similar to the query information can be accurately retrieved. Finally, based on the first and second candidate object information sets, a pre-trained large language model can accurately and efficiently generate response content for the query information. In summary, the semantic understanding capabilities of large models can overcome matching limitations, while the structured relational data of knowledge graphs can improve matching accuracy and efficiency, thus accurately obtaining information for each candidate object. Therefore, response content corresponding to the object information can be generated accurately and efficiently. Attached Figure Description

[0019] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0020] Figure 1 This is a schematic diagram of an application scenario of the object information response method according to some embodiments of the present disclosure; Figure 2 These are flowcharts of some embodiments of the object information response method according to this disclosure; Figure 3 These are flowcharts of some other embodiments of the object information response method according to this disclosure; Figure 4These are schematic diagrams illustrating the structure of some embodiments of the object information response device according to this disclosure; Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0021] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0022] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0023] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0024] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0025] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0026] Before performing any of the operations involving the collection, storage, or use of user personal information (such as query information) disclosed in this disclosure, the relevant organizations or individuals shall fulfill their obligations, including conducting personal information security impact assessments, informing personal information subjects, and obtaining prior authorization and consent from personal information subjects.

[0027] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0028] Figure 1 This is a schematic diagram illustrating an application scenario of an object information response method according to some embodiments of this disclosure. Figure 1 In the application scenario, firstly, the electronic device 101 can extract retrieval information from the query information 102, which represents information about the object being queried, to obtain retrieval information 103. In Figure 1In the application scenario, query information 102 can be: querying the person in charge of "Responsibility A". Retrieval information 103 can include: "Responsibility A" and "person". Then, electronic device 101 can query the first candidate object information set 105 corresponding to the above-mentioned retrieval information 103 from the object information knowledge graph 104. Then, electronic device 101 can use graph retrieval enhancement generation technology based on large models to retrieve the second candidate object information set 106 whose semantic similarity with the above-mentioned query information 102 meets the target similarity condition. Finally, based on the above-mentioned first candidate object information set 105 and the above-mentioned second candidate object information set 106, using the pre-trained large language model 107, object information response content 108 is generated for the above-mentioned query information 102. Figure 1 In the application scenario, the object information response content 108 could be: "The person in charge of "Responsibility A" is "Zhang**".

[0029] It should be noted that the aforementioned electronic device 101 can be either hardware or software. When the electronic device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the electronic device is software, it can be installed in the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0030] It should be understood that Figure 1 The number of electronic devices shown is merely illustrative. Any number of electronic devices can be used depending on the implementation requirements.

[0031] Continue to refer to Figure 2 The flowchart 200 illustrates some embodiments of the object information response method according to the present disclosure. The object information response method includes the following steps: Step 201: Extract retrieval information from the query information representing the information of the queried object to obtain retrieval information.

[0032] In some embodiments, the executing entity of the above-described object information response method (e.g.) Figure 1The electronic device 101 shown can extract retrieval information from query information representing information used to query objects, thereby obtaining retrieval information. The query object information can be content related to the query object. For example, the query object information can be the responsibilities of the querying user, and the corresponding query information could be "the responsibilities of the querying user." Another example is the query object information being the office address of the querying user, and the corresponding query information could be "the office address of the querying user." The retrieval information can be the retrieval content used for subsequent retrieval of corresponding object information. That is, based on the retrieval information, corresponding matching candidate objects are retrieved. For example, the retrieval information can be key fields in the query information. For example, for the query information "the responsibilities of the querying user," the corresponding retrieval information could be "responsibilities" and "personnel." Another example is for the query information being "who is in position A," and the corresponding retrieval information could be "position A" and "personnel."

[0033] As an example, the aforementioned executing entity can utilize a large language model to determine the query keywords corresponding to the query information, which can then be used as retrieval information.

[0034] In some optional implementations of certain embodiments, the retrieval information includes: query intent and query keywords. The query intent can be the purpose of querying the content corresponding to the query information. For example, for the query information "the responsibilities of a user", the corresponding query intent could be: the intent to query responsibilities. The query keywords can be key query terms in the query information. For example, for the query information "the responsibilities of a user", the corresponding query keywords could include: "responsibilities" and "personnel".

[0035] As an example, firstly, the aforementioned execution entity can generate a definitive prompt message representing the query intent and query keywords for the query information. Then, the definitive prompt message is input into the large language model to obtain the query intent and query keywords.

[0036] Optionally, the aforementioned executing entity can query the first candidate object information set corresponding to the retrieved information from the object information knowledge graph, including the following steps: The first step is to generate a graph query statement by using the above query intent and keywords as query conditions. The query conditions can be the conditions within the graph query statement. The graph query statement can be a statement that queries graph content. In practice, the graph query statement can be a Cypher statement.

[0037] The second step is to execute the above graph query statement to query the first candidate object information set corresponding to the above search information from the above object information knowledge graph.

[0038] Step 202: Query the first candidate object information set corresponding to the above-mentioned retrieval information from the object information knowledge graph.

[0039] In some embodiments, the aforementioned executing entity can query a first candidate object information set corresponding to the retrieved information from an object information knowledge graph. The object information knowledge graph can be a knowledge graph generated based on the object information corresponding to each object. In practice, nodes in the object information knowledge graph can be object identifiers and object attributes. Edges in the object information knowledge graph can include: associations between objects, and correspondences between objects and object attributes. For example, object attributes can include: object responsibilities, object contact information, object gender, and object office address. The candidate object information in the first candidate object information set can be candidate object information that matches the content to be retrieved corresponding to the retrieved information. For example, the first candidate object information can be a candidate object identifier.

[0040] As an example, firstly, retrieve all search nodes from the aforementioned object information knowledge graph that share the same keywords as the searched information. Then, for each search node, identify the node groups connected to it as object identifiers. Finally, perform node deduplication on the obtained node groups to obtain deduplicated node groups, which serve as the first candidate object information set.

[0041] In some optional implementations of certain embodiments, the above-mentioned object information knowledge graph is constructed through the following steps: The first step is to acquire a corpus of object information for the target domain. The target domain can be the domain from which the knowledge graph is to be built. For example, the target domain could be logistics or warehousing. The corresponding object information corpus could be the corpus content of various logistics objects within the logistics domain. Similarly, the corresponding object information corpus could be the corpus content of various warehousing processing objects within the warehousing domain. In short, the object information corpus can be a corpus storing object information for various objects within the target domain.

[0042] The second step is to construct an object information knowledge graph based on the aforementioned object information corpus. The graph data corresponding to this object information knowledge graph is stored in a graph database using graph query statements. In practice, the graph query statements can be Cypher statements.

[0043] As an example, firstly, the aforementioned execution entity can extract various objects from the object information corpus as entities. Then, based on the object information corpus, it determines the entity relationships between these entities. Finally, it generates an object information knowledge graph by using each entity as a graph node and the entity relationships as graph edges.

[0044] In some optional implementations of certain embodiments, the target domain described above is the government affairs domain. The corresponding object information corpus can be a proprietary corpus of the government affairs domain. The object information corpus may include: over 100,000 sample data from weekly government reports, departmental responsibility descriptions, government affairs handling guides, and publicly available government internet corpus data.

[0045] The above-mentioned object information corpus was constructed through the following steps: The first step involves using a pre-trained government affairs object information extraction model to extract multi-dimensional object information from each government affairs text in the acquired government affairs text set. This object information includes: government affairs responsibility information and confidence level. The government affairs object information extraction model can be a model that extracts object information related to government affairs objects. In practice, the government affairs object information extraction model can be a BERT-BILSTM-CRF model. This model can be pre-trained based on various corpora from a proprietary corpus of government affairs. The government affairs object information extraction model can accurately identify professional terms, item names, department names, etc., in government affairs scenarios. The confidence level represents the accuracy of government affairs responsibility information. Government affairs texts can be text content within the government affairs domain.

[0046] Specifically, the information dimensions that the government information extraction model supports for extraction can include: (1) Work items: A complete description of the items (e.g., “Handling the filing of the 2025 corporate income tax settlement and payment”, “Coordinating the cross-departmental project approval items of the target unit”).

[0047] (2) Key terms for the item: Extract the core terms of the item (e.g., "corporate income tax settlement" and "cross-departmental project approval").

[0048] (3) Person in charge: The person directly in charge of the matter (e.g., name, employee number).

[0049] (4) Department / Office: The first-level department and second-level office where the person in charge is located.

[0050] (5) Collaboration relationship: Collaborating personnel and departments involved in the matter.

[0051] (6) Processing period: The processing time or time range of the matter (e.g., "3 working days", "October-November 2025").

[0052] As an example, the government information extraction model can be trained through the following steps: First, based on the various corpora in the proprietary corpus of the government affairs field, the initial government affairs object information extraction model is trained to obtain the model after the first training.

[0053] Then, based on the 50,000+ labeled sample data, the parameters of the BILSTM-CRF layer in the model after the first training were fine-tuned to obtain the government affairs object information extraction model.

[0054] The second step is to extract at least one government document from the aforementioned set of government documents whose corresponding confidence level is lower than the target confidence level. The target confidence level can be a pre-set confidence threshold. For example, the target confidence level could be "0.8". When the confidence level of a government document is lower than the target confidence level, the government responsibility information extracted by the representation model is not accurate enough, and further confirmation of government responsibilities is required.

[0055] The third step involves sending at least one of the aforementioned government documents to a verification terminal, where at least one corresponding object information is generated. This verification terminal can be the terminal corresponding to the verification interface used for manual confirmation of government responsibilities. There is a one-to-one correspondence between the government documents in the at least one government document and the object information in the at least one object information. The object information can be the result of manual supplementation and annotation of government information.

[0056] The fourth step is to construct an object information corpus based on at least one object information and the obtained object information set.

[0057] As an example, the aforementioned executing entity can combine the above-mentioned at least one object information and the obtained object information set to obtain an object information corpus.

[0058] In some optional implementations of certain embodiments, the aforementioned government document set is generated through the following steps: The first step is to obtain the government data set. This government data can be data from the government sector.

[0059] As an example, by connecting to the online cloud documents of the government collaborative office platform through the API interface, a scheduled task can be designed (for example, starting at 2:00 AM every Monday) to automatically retrieve the weekly report documents submitted by all users in the previous week. It supports various formats of weekly report data such as Word, PDF, collaborative documents, platform forms, and rich text, so that users do not need to manually upload them to achieve automatic acquisition of government datasets.

[0060] The second step involves performing the following generation steps for each piece of government data in the aforementioned government data set: Sub-step 1 involves performing at least one of the following on the aforementioned government data: content filtering and sensitive information desensitization, to obtain government processing data. Content filtering can involve initial screening using rules, followed by parsing the content of various documents and then using a large language model for secondary filtering to obtain valid weekly report data. Sensitive information desensitization can employ a hybrid desensitization algorithm based on rules and Named Entity Recognition (NER). Specifically, first, explicitly sensitive information (e.g., matters marked "Confidential," "Secret," or "Top Secret," ID numbers, mobile phone numbers, and classified document numbers) is filtered using a rule base. Then, the NER model identifies and masks non-public sensitive information such as names and departments (e.g., permissions can be further configured for public display), ensuring the data complies with government data security regulations.

[0061] Sub-step 2 involves integrating the aforementioned government processing data to obtain object information. This object information integration can involve combining various pieces of information corresponding to each object.

[0062] As an example, by calling the platform's user information interface, personal information such as department, position, full department path, and userId of each object in the government affairs processing data is assembled to obtain object information.

[0063] Sub-step 3 involves standardizing and formatting the aforementioned government processing data and object information to obtain government documents.

[0064] As an example, the Apache Tika tool can be used to parse weekly reports in different formats, extract the plain text content, convert it into standardized text with UTF-8 encoding, remove redundant format tags (such as font styles in Word and headers and footers in PDF), and encapsulate it according to the structure of "submitter-submission time-department-report body" to obtain government affairs text, which is convenient for subsequent information extraction.

[0065] Step 203: Using graph retrieval enhancement generation technology based on large models, retrieve a second set of candidate object information whose semantic similarity with the above query information meets the target similarity condition.

[0066] In some embodiments, the aforementioned executing entity can utilize a large-model-based graph retrieval enhancement generation technique to retrieve a set of second candidate object information whose semantic similarity to the query information meets the target similarity condition. The large-model-based graph retrieval enhancement generation technique can be a technique based on the large-model GraphRAG to achieve semantic matching. Semantic similarity can be the degree of semantic matching between the query content corresponding to the query information and the object information. In practice, semantic similarity can be a value between 0 and 1; the higher the value, the higher the degree of semantic matching, and the more accurate the object corresponding to the retrieved second candidate object information. The target similarity condition can be retrieving object information with a semantic similarity higher than the target similarity, or it can be retrieving object information whose corresponding semantic similarity is among the top target number. The target number can be a pre-set number. For example, the target number can be 10. The second candidate object information can be object information that has semantic similarity to the query information and is used as a candidate object. For example, the second candidate object information can be the object identifier corresponding to the candidate object.

[0067] Step 204: Based on the first candidate object information set and the second candidate object information set, use a pre-trained large language model to generate object information response content for the above query information.

[0068] In some embodiments, the executing entity can generate object information response content for the query information based on the first candidate object information set and the second candidate object information set, using a pre-trained large language model. The large language model can be a large language model that supports various processing of object information. In practice, the large language model can be a large language model pre-trained based on object information and knowledge from various domains. Specifically, the large language model can support object information response.

[0069] As an example, firstly, the aforementioned execution entity can generate query suggestions that most closely match the query information from the first and second candidate object information sets. Then, the query suggestions are input into the large language model to obtain the object information response.

[0070] The above-described embodiments of this disclosure have the following beneficial effects: The object information response method of some embodiments of this disclosure can overcome matching limitations by utilizing the semantic understanding capabilities of large models, and simultaneously improve matching accuracy and efficiency by leveraging the structured association data of knowledge graphs, thereby accurately obtaining information about each candidate object. Thus, the response content corresponding to the object information can be accurately generated. Specifically, the reason for the low efficiency of querying related object information is that static object directories require manual periodic entry and updates. When a large number of object information changes, there is a problem that the updated object information cannot be synchronized in a timely manner and maintenance is difficult. In addition, online consultation methods also suffer from low efficiency and high cost. Based on this, the object information response method of some embodiments of this disclosure first extracts retrieval information from the query information representing the object information used for querying, which can accurately obtain retrieval information to determine the key retrieval content for object retrieval, facilitating subsequent object matching. Then, it queries the first candidate object information set corresponding to the above retrieval information from the object information knowledge graph. Here, through the association reasoning capabilities corresponding to the object information knowledge graph, the first candidate object information set can be initially and accurately retrieved. Furthermore, by leveraging graph retrieval enhancement generation techniques based on large models and utilizing semantic matching capabilities to overcome the limitations of keyword matching, a second set of candidate object information that semantically similar to the query information can be accurately retrieved. Finally, based on the first and second candidate object information sets, a pre-trained large language model can accurately and efficiently generate response content for the query information. In summary, the semantic understanding capabilities of large models can overcome matching limitations, while the structured relational data of knowledge graphs can improve matching accuracy and efficiency, thus accurately obtaining information for each candidate object. Therefore, response content corresponding to the object information can be generated accurately and efficiently.

[0071] Further reference Figure 3 The flowchart 300 illustrates some other embodiments of the object information response method according to the present disclosure. This object information response method includes the following steps: Step 301: Extract retrieval information from the query information representing the information of the queried object to obtain retrieval information.

[0072] Step 302: Query the first candidate object information set corresponding to the above-mentioned retrieval information from the object information knowledge graph.

[0073] Step 303: Using graph retrieval enhancement generation technology based on large models, retrieve a second candidate object information set whose semantic similarity with the above query information meets the target similarity condition.

[0074] In some embodiments, the specific implementation of steps 301-303 and the resulting technical effects can be found in [reference needed]. Figure 2 Steps 201-203 in the corresponding embodiments will not be repeated here.

[0075] Step 304: Sort and filter the first candidate object information set and the second candidate object information set according to their relevance to obtain the third candidate object information set.

[0076] In some embodiments, the execution entity (e.g., Figure 1 The electronic device 101 shown can perform correlation sorting and filtering on the first candidate object information set and the second candidate object information set to obtain a third candidate object information set.

[0077] As an example, firstly, the aforementioned executing entity can deduplicate the candidate object information sets of the first and second candidate object information sets to obtain a third candidate object information set. Then, it can select the top three candidate object information sets with the highest confidence from the candidate object information set to obtain a third candidate object information set.

[0078] Step 305: Generate prompt words for the above-mentioned third candidate object information set and the above-mentioned query information.

[0079] In some embodiments, the executing entity may generate prompt words for the third candidate object information set and the query information. These prompt words may be prompt words used by the prompt language model to generate response content based on the third candidate object information set and the query information.

[0080] Step 306: Input the above prompt words into the above large language model to obtain the above object information response content.

[0081] In some embodiments, the execution entity may input the prompt words into the large language model to obtain the object information response content.

[0082] from Figure 3 It can be seen from this that, with Figure 2 Compared to the description of some corresponding embodiments, Figure 3 In some corresponding embodiments, the object information response method process 300, by sorting and filtering the first candidate object information set and the second candidate object information set based on relevance, can accurately generate object information response content based on a large language model.

[0083] Further reference Figure 4 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of an object information response device, which are similar to... Figure 2Corresponding to the method embodiments shown, this object information response device can be specifically applied to various electronic devices.

[0084] like Figure 4 As shown, an object information response device 400 includes: an extraction unit 401, a query unit 402, a retrieval unit 403, and a generation unit 404. The extraction unit 401 is configured to extract retrieval information from query information representing object information, obtaining retrieval information; the query unit 402 is configured to query a first candidate object information set corresponding to the retrieval information from an object information knowledge graph; the retrieval unit 403 is configured to retrieve a second candidate object information set whose semantic similarity to the query information meets the target similarity condition using a graph retrieval enhancement generation technique based on a large model; and the generation unit 404 is configured to generate object information response content for the query information based on the first and second candidate object information sets using a pre-trained large language model.

[0085] In some optional implementations of some embodiments, the above-mentioned retrieval information includes: query intent and query keywords; and the query unit 402 can be further configured to: use the above-mentioned query intent and the above-mentioned query keywords as query conditions to generate a graph query statement; execute the above-mentioned graph query statement to query the first candidate object information set corresponding to the above-mentioned retrieval information from the above-mentioned object information knowledge graph.

[0086] In some optional implementations of certain embodiments, the generation unit 404 may be further configured to: sort and filter the first candidate object information set and the second candidate object information set according to their relevance to obtain a third candidate object information set; generate prompt words for the third candidate object information set and the query information; and input the prompt words into the large language model to obtain the object information response content.

[0087] It is understandable that the units described in the object information response device 400 are related to the reference. Figure 2 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the object information response device 400 and the units contained therein, and will not be repeated here.

[0088] The following is for reference. Figure 5 It illustrates electronic devices suitable for implementing some embodiments of this disclosure (e.g., Figure 1 A schematic diagram of the structure of electronic device 101)500. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0089] like Figure 5 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory 502 or a program loaded from a storage device 508 into a random access memory 503. The random access memory 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, the read-only memory 502, and the random access memory 503 are interconnected via a bus 504. An input / output interface 505 is also connected to the bus 504.

[0090] Typically, the following devices can be connected to the input / output interface 505: input devices 506 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 508 including, for example, magnetic tape, hard disk, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0091] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a read-only memory 502. When the computer program is executed by the processing device 501, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0092] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0093] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0094] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: extract retrieval information from query information representing information about the queried object, obtaining retrieval information; query a first set of candidate object information corresponding to the retrieval information from an object information knowledge graph; utilize a large-model-based graph retrieval enhancement generation technique to retrieve a second set of candidate object information whose semantic similarity to the query information meets the target similarity condition; and, based on the first and second candidate object information sets, generate an object information response to the query information using a pre-trained large language model.

[0095] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0096] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0097] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an extraction unit, a query unit, a retrieval unit, and a generation unit. The names of these units do not necessarily limit the unit itself; for example, an extraction unit may also be described as "a unit that extracts retrieval information from query information representing information used for querying objects, thereby obtaining retrieval information."

[0098] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0099] Some embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements any of the object information response methods described above.

[0100] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for responding to object information, comprising: The retrieval information is obtained by extracting retrieval information from the query information that represents the information of the queried object. Query the first candidate object information set corresponding to the retrieved information from the object information knowledge graph; Using graph retrieval enhancement generation technology based on large models, a second set of candidate object information that has semantic similarity to the query information and meets the target similarity condition is retrieved; Based on the first candidate object information set and the second candidate object information set, a pre-trained large language model is used to generate object information response content for the query information.

2. The method according to claim 1, wherein, The retrieval information includes: query intent and query keywords; and The step of querying the first candidate object information set corresponding to the retrieved information from the object information knowledge graph includes: The query intent and the query keywords are used as query conditions to generate a graph query statement; Execute the graph query statement to query the first candidate object information set corresponding to the retrieval information from the object information knowledge graph.

3. The method according to claim 1, wherein, The step of generating object information response content for the query information based on the first candidate object information set and the second candidate object information set, using a pre-trained large language model, includes: The first candidate object information set and the second candidate object information set are sorted and filtered based on relevance to obtain the third candidate object information set. Generate prompt words for the third candidate object information set and the query information; The prompt words are input into the large language model to obtain the object information response content.

4. The method according to claim 1, wherein, The object information knowledge graph is constructed through the following steps: Acquire a corpus of object information for the target domain; Based on the object information corpus, an object information knowledge graph is constructed, wherein the graph data corresponding to the object information knowledge graph is stored in a graph database through graph query statements.

5. The method according to claim 4, wherein, The target domain is the government affairs domain, and the object information corpus is constructed through the following steps: Using a pre-trained government affairs object information extraction model, multi-dimensional object information is extracted from each government affairs text in the acquired government affairs text set. The object information includes: government affairs responsibility information and confidence level. Extract at least one government document from the set of government documents with a confidence level lower than the target confidence level; The at least one government document is sent to the verification terminal to generate at least one corresponding object information on the verification terminal; Based on the at least one object information and the obtained object information set, construct an object information corpus.

6. The method according to claim 5, wherein, The set of government documents is generated through the following steps: Obtain government data sets; For each piece of government data in the aforementioned government data set, the following generation steps are performed: Perform at least one of the following on the government data: content filtering and sensitive information desensitization, to obtain government processed data; The government processing data is integrated with object information to obtain object information; The government processing data and the object information are converted into government documents using a standard format.

7. An object information response device, comprising: The extraction unit is configured to extract retrieval information from the query information that represents information about the query object, and obtain retrieval information. The query unit is configured to query the first candidate object information set corresponding to the retrieval information from the object information knowledge graph; The retrieval unit is configured to use graph retrieval enhancement generation technology based on large models to retrieve a set of second candidate object information whose semantic similarity with the query information meets the target similarity condition; The generation unit is configured to generate object information response content for the query information based on the first candidate object information set and the second candidate object information set, using a pre-trained large language model.

8. An electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

9. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

10. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-6.

Images