Knowledge graph-based target reporting data processing method and device, and server

Abstract

The present specification provides a knowledge graph-based target reporting data processing method, device and server, which can be used in the field of artificial intelligence. When a user encounters a target problem related to target reporting while processing target reporting business, the user can initiate a target query request to the server. The server receives and responds to the target query request, first distinguishes different problem types, and according to the problem type, adopts a matching mode to process the target problem text carried by the target query request to obtain a corresponding target problem analysis result; then according to the target problem analysis result, a corresponding target query sentence matched with the target knowledge graph is constructed through semantic slot extraction; and then according to the target query sentence, the target knowledge graph is queried to timely feed back the corresponding target answer text to the user. Thus, the processing difficulty of the user in processing the target reporting business can be effectively reduced, and the user can be assisted to efficiently and accurately complete the related target reporting business.

Description

Technical Field

[0001] This specification belongs to the field of artificial intelligence technology, and in particular relates to methods, devices and servers for processing target reporting data based on knowledge graphs. Background Technology

[0002] In the financial services sector, to better protect user interests and effectively reduce the risks of financial service institutions, related institutions or internal entities often request the submission of relevant data files regularly or irregularly for evaluation purposes. For example, regulatory agencies may require financial service institutions to compile and submit a list of foreign exchange transactions handled in the previous month on the first day of each month, in accordance with established procedures.

[0003] Based on existing methods, staff responsible for target reporting in financial service institutions need to be familiar with the numerous and diverse target reporting rules and requirements of various other related institutions, as well as within their own institutions, in order to accurately prepare the corresponding target reporting documents and send them to the appropriate recipients in a standardized manner for review and processing. However, the target reporting rules and requirements of related institutions and within the financial service institutions themselves frequently change and are updated. It is difficult for staff to keep up with and master the latest target reporting rules and requirements in a timely and accurate manner, leading to frequent errors in the specific processing of target reporting tasks.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This specification provides a method, apparatus, and server for processing target reporting data based on knowledge graphs, which can effectively reduce the processing difficulty for users when handling target reporting business and assist users in completing related target reporting business efficiently and accurately.

[0006] This specification provides a knowledge graph-based method for processing target reporting data, applied to a server, including:

[0007] Receive a target query request; wherein the target query request carries at least the target question text related to the target report;

[0008] In response to the target query request, the target question text is processed using a preset classifier to determine the question type of the target question;

[0009] According to the problem type of the target problem, a preset problem analysis model matched with the target problem is determined from a plurality of preset problem analysis models as a target problem analysis model;

[0010] The target problem text is processed by using the target problem analysis model to obtain a corresponding target problem analysis result; wherein the target problem analysis result at least contains an entity object analysis result;

[0011] According to the target problem analysis result, semantic slot extraction is performed to obtain a corresponding semantic slot extraction result; and based on the semantic slot extraction result, a target query statement matched with the target knowledge graph is constructed; wherein the target knowledge graph is constructed by a server in advance according to a plurality of data source servers providing resource files related to the target report according to a preset protocol rule, in combination with a third-party knowledge database of a cooperation party through knowledge fusion;

[0012] The target knowledge graph is queried according to the target query statement; and a corresponding target answer text is fed back according to the query result.

[0013] In one embodiment, the problem type includes a problem involving a single entity object and a problem involving multiple entity objects;

[0014] Correspondingly, in the case where the problem type of the target problem includes a problem involving multiple entity objects, the target problem text is processed by using the target problem analysis model to obtain a corresponding target problem analysis result, including:

[0015] The target problem text is processed by using the target problem analysis model to obtain a corresponding target problem analysis result, including:

[0016] The part-of-speech of each word unit based on the word sequence is determined, and a corresponding part-of-speech label is set to obtain an annotated word sequence;

[0017] According to the part-of-speech label, a plurality of entity object representation words are determined in the annotated word sequence;

[0018] The relative position parameters between the other word units and the entity object representation words in the annotated word sequence except for the entity object representation words are determined;

[0019] According to the relative position parameters and the part-of-speech label, the other word units are processed to determine the association relationship representation words;

[0020] The entity object representation words and the association relationship representation words are combined to obtain the corresponding target problem analysis result.

[0021] In one embodiment, based on the semantic slot extraction result, the target query statement matched with the target knowledge graph is constructed, including:

[0022] According to the semantic slot extraction result, a matched query statement template is screened out; and an initial query statement is established based on the query statement template;

[0023] A target semantic link based on the target knowledge graph is obtained; wherein the target semantic link at least contains a url link of an entity object related to the target question;

[0024] The initial query statement is assembled by using the target semantic link, so as to obtain a corresponding target query statement.

[0025] In one embodiment, the target knowledge graph at least contains a plurality of nodes and connection edges between the nodes; wherein the nodes are used to represent entity objects in a target reporting business scenario; the connection edges are used to represent the association relationship between the connected entity objects based on the target reporting business; and the entity objects include at least one of the following: a data object to be reported, an initiating object of the data object to be reported, a receiving object of the data object to be reported, and a next processing flow of the data object to be reported.

[0026] In one embodiment, the target knowledge graph is queried according to the target query statement; and a corresponding target answer text is fed back according to the query result, including:

[0027] According to the url link of the entity object in the target query statement, an initial node is located in the target knowledge graph;

[0028] According to the target query statement, other nodes directly connected and indirectly connected with the initial node, and the connection edges between the initial node and the connected other nodes are searched, so as to obtain a corresponding query result;

[0029] The query result is processed by using a preset semantic processing model, so as to obtain a target answer text based on natural language; and the target answer text is fed back.

[0030] In one embodiment, the nodes in the target knowledge graph are also marked with attribute information related to the entity objects represented by the nodes; and the connection edges in the target knowledge graph are also marked with event conditions related to the association relationship represented by the connection edges.

[0031] Correspondingly, according to the target query statement, other nodes directly connected and indirectly connected with the initial node, and the connection edges between the initial node and the connected other nodes are searched, so as to obtain a corresponding query result, including:

[0032] According to the target query statement, other nodes directly connected and indirectly connected with the initial node, attribute information marked by the initial node and other nodes, and connection edges between the initial node and the connected other nodes, event conditions marked by the connection edges are retrieved to obtain corresponding query results.

[0033] In one embodiment, after feeding back the corresponding target answer text according to the query result, the method further comprises:

[0034] receiving an association query request; wherein the association query request carries at least an association question text related to an association question; the association query request is a query request initiated by a user terminal for the target answer text;

[0035] detecting whether the relevance between the association question and the target question is greater than a preset relevance threshold;

[0036] In the case where it is determined that the relevance between the association question and the target question is greater than the preset relevance threshold, generating an association question analysis result according to the target question analysis result;

[0037] According to the association question analysis result and the target query statement, an association query statement matched with the target knowledge graph is constructed;

[0038] According to the association query statement, the target knowledge graph is queried; and according to the query result and the target answer text, a corresponding association answer text is fed back.

[0039] In one embodiment, the method further comprises:

[0040] According to a preset protocol rule, resource files related to the target report provided by a plurality of data source servers are obtained;

[0041] The plurality of resource files are respectively subjected to semantic conversion processing to obtain a plurality of resource semantic texts corresponding to the resource files;

[0042] The text data type of the resource semantic text is determined; wherein the text data type of the resource semantic text includes at least one of the following: structured data text, unstructured data text, semi-structured data text;

[0043] According to the text data type of the resource semantic text, a matching knowledge extraction rule is determined and used to process the resource semantic text to extract corresponding knowledge element data;

[0044] According to a preset fusion rule, knowledge fusion is performed on the knowledge element data and a third-party knowledge database to obtain fused knowledge element data;

[0045] According to the fused knowledge element data, a corresponding target knowledge graph is constructed.

[0046] In one embodiment, in the case that the text data type of the resource semantic text comprises structured data text, the resource semantic text is processed by using a matching knowledge extraction rule, including:

[0047] According to a preset field template, the resource semantic text is retrieved, and a text field associated with a preset identification item field is extracted from the resource semantic text to be combined to obtain knowledge element data corresponding to the resource semantic text; the preset field template contains a plurality of preset identification item fields; the knowledge element data at least includes an entity object representation word and an associated relationship representation word.

[0048] In one embodiment, in the case that the text data type of the resource semantic text comprises unstructured data text or semi-structured data text, the resource semantic text is processed by using a matching knowledge extraction rule, including:

[0049] The resource semantic text is subjected to semantic recognition by using a preset semantic recognition model to obtain a corresponding resource semantic text recognition result;

[0050] The resource semantic text recognition result is subjected to word segmentation processing to obtain a plurality of word segmentation units, and a grammatical relationship feature of the word segmentation units based on the resource semantic text recognition result is determined;

[0051] According to a preset field template and the grammatical relationship feature of the word segmentation units based on the resource semantic text recognition result, knowledge element data corresponding to the resource semantic text is identified and extracted from the plurality of word segmentation units.

[0052] In one embodiment, according to a preset fusion rule, knowledge fusion is performed by using the knowledge element data and a third-party knowledge database to obtain fused knowledge element data, including:

[0053] According to a preset fusion rule, third-party knowledge element data is obtained from the third-party knowledge database;

[0054] The vector distance between semantic feature vectors of entity object representation words in different knowledge element data is calculated;

[0055] According to the vector distance between the semantic feature vectors of the entity object representation words and the third-party knowledge element data, an entity object representation word indicating a same entity object is determined, and knowledge element data containing the entity object representation word indicating the same entity object is subjected to first fusion processing to obtain corresponding first fused knowledge element data;

[0056] The first fused knowledge element data and the third-party knowledge element data are subjected to a second fusion processing to obtain corresponding second fused knowledge element data as the fused knowledge element data.

[0057] In one embodiment, according to the fused knowledge element data, a corresponding target knowledge graph is constructed, including:

[0058] According to the fused knowledge element data, an initial knowledge graph is constructed;

[0059] According to the initial knowledge graph, knowledge processing is performed to obtain extension data of the knowledge graph; wherein the knowledge processing includes at least one of the following: reasoning and extension based on existing nodes and existing connection edges in the initial knowledge graph; prediction and extension based on attribute information of existing nodes and existing connection edges in the initial knowledge graph; prediction and extension based on existing graph structure and existing connection edges in the initial knowledge graph;

[0060] According to the extension data of the knowledge graph, the initial knowledge graph is adjusted to obtain a corresponding target knowledge graph.

[0061] The present specification also provides a target report data processing device based on a knowledge graph, applied to a server, including:

[0062] A receiving module is configured to receive a target query request; wherein the target query request carries at least a target problem text of a target problem related to a target report;

[0063] A classification module is configured to respond to the target query request, process the target problem text by using a preset classifier, and determine a problem type of the target problem;

[0064] A determination module is configured to determine a matching preset problem analysis model from a plurality of preset problem analysis models as a target problem analysis model according to the problem type of the target problem;

[0065] A processing module is configured to process the target problem text by using the target problem analysis model to obtain a corresponding target problem analysis result; wherein the target problem analysis result at least includes an entity object analysis result;

[0066] A construction module is configured to perform semantic slot extraction according to the target problem analysis result to obtain a corresponding semantic slot extraction result; and construct a target query statement matched with a target knowledge graph based on the semantic slot extraction result; wherein the target knowledge graph is constructed by a server according to a plurality of data source servers providing resource files related to the target report according to a preset protocol rule, in combination with a third-party knowledge database of a cooperation party, through knowledge fusion;

[0067] The query module is configured to query the target knowledge graph according to the target query statement, and feed back the corresponding target answer text according to the query result.

[0068] The present specification also provides a server comprising a processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the related steps of the target reporting data processing method based on a knowledge graph.

[0069] The present specification also provides a computer-readable storage medium having computer instructions stored thereon, wherein the instructions are executed by a processor to implement the following steps: receiving a target query request, wherein the target query request carries at least a target question text of a target question related to a target report; in response to the target query request, processing the target question text by using a preset classifier to determine a question type of the target question; according to the question type of the target question, determining a matching preset question analysis model from a plurality of preset question analysis models as a target question analysis model; processing the target question text by using the target question analysis model to obtain a corresponding target question analysis result, wherein the target question analysis result at least includes an entity object analysis result; performing semantic slot extraction according to the target question analysis result to obtain a corresponding semantic slot extraction result; and constructing a target query statement matched with a target knowledge graph based on the semantic slot extraction result, wherein the target knowledge graph is constructed by a server in advance according to a plurality of data source servers providing resource files related to the target report according to a preset protocol rule, in combination with a third-party knowledge database of a cooperation party through knowledge fusion; querying the target knowledge graph according to the target query statement; and feeding back a corresponding target answer text according to a query result.

[0070] Based on the knowledge graph-based target submission data processing method, device and server provided in the specification, before implementation, the server can obtain and provide resources related to target submission according to the preset protocol rules by interacting with multiple resource parties and cooperation parties, and at the same time, combined with the third-party knowledge database of the cooperation party, a relatively comprehensive and accurate target knowledge graph for the target submission business scenario is constructed through knowledge fusion. When a user encounters an unclear target problem related to target submission when processing target submission business (or target submission task), a target query request can be initiated to the server. The server receives and responds to the target query request, first distinguishes different problem types, and according to the problem type, adopts a matching mode to process the target problem text carried by the target query request, and obtains the corresponding target problem analysis result; then according to the target problem analysis result, the corresponding target query statement matched with the target knowledge graph is constructed through semantic slot extraction; and then according to the target query statement, the target knowledge graph is queried, and the corresponding target answer text is fed back to the user in time. Thus, the processing difficulty of the user when processing the target submission business can be effectively reduced, the user can be assisted to efficiently and accurately complete the related target submission business, and the error in target submission can be reduced. BRIEF DESCRIPTION OF DRAWINGS

[0071] In order to more clearly illustrate the embodiments of the specification, the drawings needed in the embodiments will be briefly introduced as follows. The drawings in the following description are only some embodiments described in the specification, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0072] Figure 1 is a flowchart of a knowledge graph-based target submission data processing method provided by an embodiment of the specification;

[0073] Figure 2 is a schematic diagram of one embodiment of the knowledge graph-based target submission data processing method provided by the embodiment of the specification in one scene example;

[0074] Figure 3 is a schematic diagram of one embodiment of the knowledge graph-based target submission data processing method provided by the embodiment of the specification in one scene example;

[0075] Figure 4 is a schematic diagram of one embodiment of the knowledge graph-based target submission data processing method provided by the embodiment of the specification in one scene example;

[0076] Figure 5is a schematic diagram of an embodiment of the knowledge graph-based target reporting data processing method provided by the embodiments of the present specification in one scenario example;

[0077] Figure 6 is a schematic diagram of an embodiment of the knowledge graph-based target reporting data processing method provided by the embodiments of the present specification in one scenario example;

[0078] Figure 7 is a schematic diagram of an embodiment of the knowledge graph-based target reporting data processing method provided by the embodiments of the present specification in one scenario example;

[0079] Figure 8 is a schematic diagram of an embodiment of the knowledge graph-based target reporting data processing method provided by the embodiments of the present specification in one scenario example;

[0080] Figure 9 is a structural composition schematic diagram of a server provided by an embodiment of the present specification;

[0081] Figure 10 is a structural composition schematic diagram of a knowledge graph-based target reporting data processing apparatus provided by an embodiment of the present specification;

[0082] Figure 11 is a schematic diagram of an embodiment of the knowledge graph-based target reporting data processing method provided by the embodiments of the present specification in one scenario example;

[0083] Figure 12 is a schematic diagram of an embodiment of the knowledge graph-based target reporting data processing method provided by the embodiments of the present specification in one scenario example;

[0084] Figure 13 is a schematic diagram of an embodiment of the knowledge graph-based target reporting data processing method provided by the embodiments of the present specification in one scenario example. DETAILED DESCRIPTION

[0085] In order for those skilled in the art to better understand the technical solutions in the present specification, the technical solutions in the embodiments of the present specification will be described clearly and completely below in conjunction with the drawings in the embodiments of the present specification. Obviously, the described embodiments are only part of the embodiments of the present specification, not all the embodiments. Based on the embodiments in the present specification, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present specification.

[0086] Reference Figure 1As shown, the embodiments of the present specification provide a target reporting data processing method based on a knowledge graph, wherein the method is specifically applied to the server side. In specific implementation, the method can include the following contents:

[0087] S101: receiving a target query request; wherein the target query request carries at least a target problem text of a target problem related to target reporting;

[0088] S102: in response to the target query request, processing the target problem text by using a preset classifier to determine the problem type of the target problem;

[0089] S103: according to the problem type of the target problem, determining a matched preset problem analysis model from a plurality of preset problem analysis models as a target problem analysis model;

[0090] S104: processing the target problem text by using the target problem analysis model to obtain a corresponding target problem analysis result; wherein the target problem analysis result at least includes an entity object analysis result;

[0091] S105: according to the target problem analysis result, performing semantic slot extraction to obtain a corresponding semantic slot extraction result; and based on the semantic slot extraction result, constructing a target query sentence matched with the target knowledge graph; wherein the target knowledge graph is constructed by a server according to a plurality of data source servers providing resource files related to target reporting, in combination with a third-party knowledge database of a cooperation party, through knowledge fusion according to a preset protocol rule;

[0092] S106: querying the target knowledge graph according to the target query sentence; and feeding back a corresponding target answer text according to the query result.

[0093] Among them, the above-mentioned target knowledge graph can be specifically understood as a knowledge graph designed for a target reporting scene, which is fused with knowledge data of a plurality of data source servers (resource parties) and contains almost all the required target reporting rules and requirements. The construction method of the above-mentioned target knowledge graph will be described later. The above-mentioned target reporting data can be understood as a related business data file to be reported. The above-mentioned target reporting can be, for example, a regulatory reporting, a summary reporting, an original material reporting, etc.

[0094] Based on the above embodiment, when the user responsible for the target submission business encounters a target problem related to the target submission that he does not know when processing the target submission business, he can initiate a target query request to the server through the user terminal he holds; wherein the target query request carries at least a target problem text. The server receives and responds to the target query request, first distinguishes different problem types, and according to the problem type, adopts a matching mode to process the target problem text carried by the target query request, and obtains the corresponding target problem analysis result; then according to the target problem analysis result, through semantic slot extraction, a corresponding target query statement matching the target knowledge graph is constructed; then according to the target query statement, the target knowledge graph is queried to efficiently obtain the corresponding query result; and based on the above query result, the corresponding target answer text is fed back to the user in time. Thus, the processing difficulty of the user when processing the target submission business can be effectively reduced, the user can be assisted to efficiently and accurately complete the related target submission business, and the error during processing can be reduced.

[0095] In some embodiments, referring to Figure 2 The above knowledge graph-based target submission data processing method can be specifically applied to the server side. The server can be specifically understood as a system server responsible for related data processing deployed in a financial service institution. The financial service institution can be an institution that provides financial services to users, such as a bank, a shopping website, a payment platform, and the like.

[0096] Referring to Figure 2 The server can be connected to a resource server. The resource server can be deployed on the side of a related institution such as a regulatory agency, other related institutions (i.e., data sources) associated with the financial service institution, and the like. The resource server regularly or irregularly publishes resource files related to target submission, such as announcements, target submission specifications, video speeches, news messages, and the like.

[0097] The regulatory agency can be an agency responsible for monitoring and managing the financial service institution, such as the State Administration of Financial Supervision and Management, the Audit Office, the Foreign Exchange Administration, or other related management units. The other related institutions associated with the financial service institution can include other related institutions having a subordinate relationship with the financial service institution (e.g., the parent company of the financial service institution), other related institutions having a cooperative relationship with the financial service institution, and the like.

[0098] Specifically, the resource servers connected with the server can include resource server 1 deployed on the side of institution 1, resource server 2 deployed on the side of institution 2, and resource server N deployed on the side of institution N. The resource servers connected with the server can also include resource server a deployed on the side of associated financial service institution a, resource server b deployed on the side of associated company b, and resource server c deployed on the side of financial service institution c in cooperation. Among them, the financial service institution c also needs to regularly or irregularly perform target submission. In addition, the resource servers connected with the server also include a server of a financial service institution to which the server belongs, which is responsible for sorting and publishing internal target submission rules and requirements.

[0099] In addition, referring to Figure 2 According to the corresponding cooperation agreement, the financial service institution to which the server belongs can share the knowledge database for target submission with the financial service institution c. Correspondingly, the server can be connected with the knowledge database (denoted as a third-party knowledge database) of the financial service institution c. The third-party knowledge database stores third-party knowledge element data collected and sorted by the financial service institution c in cooperation when performing target submission business.

[0100] Before implementation, the server can first obtain a plurality of resource files related to target submission by interacting with a plurality of data source servers according to preset agreement rules; then extract corresponding knowledge element data by processing the resource files; further, the server can first perform first fusion processing by using the extracted knowledge element data according to preset fusion rules, to obtain first fused knowledge element data; then perform second fusion between the first fused knowledge element data and the third knowledge database, to obtain more comprehensive fused knowledge element data; finally, the server can construct a more comprehensive and accurate target knowledge graph according to the fused knowledge element data.

[0101] In implementation, when a staff of a financial service institution to which the server belongs (hereinafter referred to as a user) encounters a target problem related to a target submission (including an external target submission and an internal target submission) that the user does not know when processing a specific target submission business, the user can use a user terminal held by the user to initiate a target query request to the server; wherein the target query request carries at least a target problem text. The server receives and responds to the target query request, first distinguishes different problem types, and according to the problem type, adopts a matching mode to efficiently and specifically process the target problem text carried by the target query request, to obtain a corresponding target problem analysis result; then according to the target problem analysis result, a target query sentence corresponding to and matching the target knowledge graph is constructed through semantic slot extraction; then according to the target query sentence, a corresponding query result can be quickly obtained by querying the target knowledge graph; and based on the query result, a target answer text based on natural language suitable for human users to understand is converted; and the target answer text is sent to the user terminal. In this way, the user can quickly solve the user's own problem based on the target answer text displayed on the user terminal, and then can smoothly and accurately complete the related target submission business.

[0102] The server and the resource server can specifically include a background server capable of realizing data transmission and data processing. Specifically, the server and the resource server can be an electronic device with data operation, storage and network interaction functions. Alternatively, the server and the resource server can be a software program running in the electronic device to provide support for data processing, storage and network interaction. In the embodiment, the number of servers included in the server and the resource server is not specifically limited. The server and the resource server can be one server, or several servers, or a server cluster formed by several servers.

[0103] The user terminal can specifically include a front end capable of realizing data acquisition and data transmission. Specifically, the user terminal can be a desktop computer, a tablet computer, a notebook computer, a smart phone or the like. Alternatively, the user terminal can be a software application capable of running in the electronic device.

[0104] In some embodiments, during the above implementation, when a user (e.g., a staff member of a financial service institution handling target reporting) encounters a specific target question related to target reporting, they can generate a corresponding target query request using their user terminal and send the target query request to the server. This target query request carries at least a target question text. Specifically, the target question text can describe the target question the user hopes to answer. For example, what data items should be included in the target reporting data for Institution 2 regarding its WE business? Or, for Institution 3, what format should the target reporting data for its ER business be in, and by what deadline?

[0105] In some embodiments, after receiving a target query request, the server can first obtain the target question text carried in the target query request through data parsing; then respond to the target query request by using a preset classifier to process the target question text in order to determine the question type of the target question described by the target question text.

[0106] Specifically, the aforementioned pre-defined classifier can be understood as a binary classification model trained in advance using labeled sample question texts with marked question types, capable of identifying the question type described by the output question text based on the input question text.

[0107] In some embodiments, the problem type may specifically include: a problem involving a single entity object, a problem involving multiple entity objects, etc.

[0108] Accordingly, when the target problem type includes problems involving multiple entity objects, refer to Figure 3 As shown, the above-described target problem analysis model is used to process the target problem text to obtain the corresponding target problem analysis results. In specific implementation, this may include the following:

[0109] S1: Use the target question analysis model to segment the target question text to obtain a word sequence containing multiple word segments ordered in sequence;

[0110] S2: Determine the part-of-speech tag for each word segmentation unit based on the word sequence, and set the corresponding part-of-speech tag to obtain the annotated word sequence;

[0111] S3: Based on the part-of-speech tags, multiple entity object representation words are identified in the annotated word sequence (the entity object analysis results are obtained);

[0112] S4: Determine the relative position parameters between the other word segmentation units (excluding entity object representation words) and entity object representation words in the annotated word sequence;

[0113] S5: According to the relative position parameters and the part-of-speech tags, other word segmentation units are processed to determine the correlation relationship representation words (the correlation relationship analysis result is obtained);

[0114] S6: The entity object representation words and the correlation relationship representation words are combined to obtain the corresponding target problem analysis result.

[0115] Based on the above embodiment, the server can distinguish different problem types, select and use a matching preset problem analysis model for different problem types to perform targeted processing on the target problem text, so that a target problem analysis result with relatively high precision and relatively small error can be obtained.

[0116] The above entity object can include at least one of the following: a data object to be reported (for example, a data report of a related institution to be uploaded), an initiating object of the data object to be reported, a receiving object of the data object to be reported, a next processing flow of the data object to be reported, and the like.

[0117] The above problem related to a single entity object can include an attribute problem related to a single entity object, and often only involves a single entity object. For example, what data items need to be included in the foreign exchange transfer target reporting data.

[0118] The above problem related to multiple entity objects can include a correlation relationship problem between multiple entity objects, and often involves multiple entity objects at the same time. For example, how to send the foreign exchange transfer target reporting data (corresponding to the first entity object) to institution 1 (corresponding to the second entity object).

[0119] For the above problem text of different problem types, the content to be focused on and the specific analysis processing logic will be different when performing problem analysis processing. Therefore, before implementation, two different preset problem analysis models for different problem types can be pre-trained, which can be denoted as: a preset first problem analysis model and a preset second problem analysis model. The preset first problem analysis model is obtained by pre-training using sample problem text related to a problem involving a single entity object, and is an algorithm model suitable for analyzing and processing problem text related to a problem involving a single entity object. The preset second problem analysis model is obtained by pre-training using sample problem text related to a problem involving multiple entity objects, and is an algorithm model suitable for analyzing and processing problem text related to a problem involving multiple entity objects.

[0120] Correspondingly, in a case where the question type of the target question includes a question involving multiple entity objects, the above-described preset second question analysis model can be determined as the matched target question analysis model. In a case where the question type of the target question includes a question involving a single entity object, the above-described preset first question analysis model can be determined as the matched target question analysis model.

[0121] In specific implementation, for the target question text of a question involving multiple entity objects, the above-described manner can be referred to, and the target question text can be analyzed and processed by using the matched target question analysis model to obtain a target question analysis result with higher reference value.

[0122] For the target question text of a question involving a single entity object, the preset first question analysis model can be used as the matched target question analysis model, and the target question text can be processed by using the target question analysis model in the following manner to obtain a corresponding target question analysis result: the target question text is processed by using the target question analysis model to obtain a word sequence including multiple word units sorted in sequence; the parts of speech of each word unit based on the word sequence are determined, and corresponding part-of-speech labels are set to obtain an annotated word sequence; an entity object representation word is determined in the annotated word sequence according to the part-of-speech labels; relative position parameters between other word units and the entity object representation word in the annotated word sequence except for the entity object representation word are determined; other word units related to the entity object represented by the entity object representation word and involving attribute information description are selected from the other word units as attribute information representation words according to the relative position parameters and the part-of-speech labels; and finally, the entity object representation word and the attribute information representation words are combined to obtain the corresponding target question analysis result.

[0123] In some embodiments, the above-described semantic slot can be specifically understood as an intent keyword for representing a user query intent. In specific implementation, the above-described target question analysis result can be retrieved based on a preset intent keyword table to extract words identical to or semantically similar to the preset intent keyword as a corresponding semantic slot extraction result; the preset intent keyword table includes multiple preset intent keywords.

[0124] The preset intent keyword table can be specifically constructed in the following manner: a large amount of historical data is clustered to obtain multiple preset intent keywords that can better cover all common query intent types; and the multiple preset intent keywords are combined to obtain the corresponding preset intent keyword table.

[0125] In some embodiments, the above-described semantic slot extraction result can be obtained by referring to the following manner: Figure 4As shown, based on the semantic slot extraction result, a target query statement matching the target knowledge graph is constructed, and in specific implementation, the following content can be included:

[0126] S1: According to the semantic slot extraction result, a matching query statement template is selected; and an initial query statement is established based on the query statement template;

[0127] S2: Obtain a target semantic link based on the target knowledge graph; wherein the target semantic link at least contains a url link of an entity object related to the target problem;

[0128] S3: Assemble the initial query statement using the target semantic link to obtain the corresponding target query statement.

[0129] Based on the above embodiments, a target query statement matching the target knowledge graph can be constructed, and the computer can efficiently and accurately identify and process the target query statement; and then the server can use the target query statement to efficiently and accurately query the target knowledge graph to obtain a query result meeting the user's query intention.

[0130] In some embodiments, the query statement template can be an SQPARQL-based statement template. The SQPARQL (SPARQL Protocol and RDF Query Language) can be an efficient query language and data acquisition protocol.

[0131] Before specific implementation, a large number of historical query statements of different query intention types can be grouped, learned and sorted to determine the statement features (including fixed fields in the statement and pending fields in the statement) of the query statements of different query intention types; and then according to the statement features of the query statements of different query intention types, a plurality of query statement templates corresponding to different query intention types are constructed. Accordingly, in specific implementation, according to the semantic slot extraction result, a query statement template matching the query intention type indicated by the semantic slot extraction result can be selected from the plurality of query statement templates; and the target problem analysis result is used to fill part of the pending fields in the query statement template to obtain an initial query statement.

[0132] Before implementation, the URL link of the entity object in the target knowledge graph can be obtained according to the currently established target knowledge graph, and the URL link of the entity object is stored in a data table; and the corresponding relationship between the URL link of each entity object and the node identifier (or the object name of the entity object) of the entity object in the target knowledge graph is stored in the data table, so as to obtain the entity object link table associated with the target knowledge graph. Correspondingly, during implementation, the URL link of the entity object related to the target problem can be obtained as the target semantic link according to the entity object representation word in the target problem analysis result and the entity object link table is retrieved.

[0133] In some embodiments, the target knowledge graph can at least include a plurality of nodes and connection edges between the nodes; wherein the nodes can be used to represent entity objects in the target reporting business scenario; the connection edges can be used to represent the association relationship between the connected entity objects based on the target reporting business; and the entity objects can include at least one of the following: a data object to be reported, an initiating object of the data object to be reported, a receiving object of the data object to be reported, a next processing flow of the data object to be reported, etc. In the target knowledge graph, the above nodes are at least marked with node identifiers. During implementation, the object name or object identifier of the entity object indicated by the node can be used as the node identifier of the node.

[0134] Based on the above embodiments, the target knowledge graph based on the above structure can comprehensively and finely represent the full amount of rules and requirements related to the target reporting based on the related resource files provided by different resource servers.

[0135] In some embodiments, referring to Figure 5 According to the target query statement, the target knowledge graph is queried, and the corresponding target answer text is fed back according to the query result. During implementation, the following contents can be included:

[0136] S1: According to the URL link of the entity object in the target query statement, the initial node in the target knowledge graph is located;

[0137] S2: According to the target query statement, other nodes directly connected and indirectly connected with the initial node, and the connection edges between the initial node and the connected other nodes are retrieved, and the corresponding query result is obtained;

[0138] S3: processing the query result by using a preset semantic processing model to obtain a target answer text based on natural language; and feeding back the target answer text.

[0139] Based on the above embodiments, the server can efficiently and accurately query the target knowledge graph by using the target query statement to obtain the corresponding query result. Furthermore, the server can obtain the target answer text based on natural language and suitable for human user reading comprehension by using the preset semantic processing model based on the above query result, and then feed back the target answer text to the user, so that the user can obtain a better interactive experience.

[0140] The other nodes directly connected with the initial node specifically refer to the nodes connected with the initial node in the target knowledge graph by only one connection edge without other third-party nodes. The other nodes indirectly connected with the initial node specifically refer to the nodes connected with the initial node in the target knowledge graph by at least two connection edges and through other third-party nodes.

[0141] The preset semantic processing model can be specifically understood as a pre-trained NLP model capable of converting the query result based on machine language into content text based on natural language and suitable for human user reading comprehension. The NLP (Natural Language Processing) is natural language processing.

[0142] In some embodiments, the nodes in the target knowledge graph can also be marked with attribute information related to the entity object represented by the node; and the connection edge in the target knowledge graph can also be marked with event conditions related to the association relationship represented by the connection edge.

[0143] Accordingly, the retrieval of the other nodes directly and indirectly connected with the initial node, the connection edge between the initial node and the connected other nodes, and the corresponding query result based on the target query statement can specifically include:

[0144] Retrieval of the other nodes directly and indirectly connected with the initial node, the attribute information marked on the initial node and the other nodes, the connection edge between the initial node and the connected other nodes, and the event conditions marked on the connection edge based on the target query statement to obtain the corresponding query result.

[0145] Based on the above embodiments, according to the target query statement, the target knowledge graph can be queried to obtain relatively rich and detailed query results including object identifiers of entity objects, attribute information of entity objects, association relationships between entity objects, and event conditions of association relationships, and then more rich and detailed target answer texts can be generated and fed back to the user based on the above query results.

[0146] The attribute information of the entity object can specifically include the category of the entity object, the source of the entity object, the content of the entity object, and the data item identifier contained in the entity object.

[0147] The event condition can specifically include a triggered condition of the association relationship, an association event triggered when the association relationship is implemented, and an event that will have a mutual influence on the association relationship.

[0148] In addition, the connection edge can also mark the attribute information of the association relationship, such as the transmission deadline of the association relationship, the transmission medium of the association relationship, and the transmission format of the association relationship.

[0149] In some embodiments, after the corresponding target answer text is fed back according to the query result, the method can specifically include the following content when implemented:

[0150] S1: receiving an association query request; wherein the association query request carries at least an association question text related to an association question; the association query request is a query request initiated by a user terminal for the target answer text;

[0151] S2: detecting whether the relevance between the association question and the target question is greater than a preset relevance threshold;

[0152] S3: generating an association question analysis result according to the target question analysis result in the case where it is determined that the relevance between the association question and the target question is greater than the preset relevance threshold;

[0153] S4: constructing an association query statement matched with the target knowledge graph according to the association question analysis result and the target query statement;

[0154] S5: querying the target knowledge graph according to the association query statement; and feeding back a corresponding association answer text according to the query result and the target answer text.

[0155] In many cases, the user can be inspired to generate related new questions that need to be further inquired and answered after reading the target answer text and solving the current target question. In this case, the user can use the user terminal to initiate the association query request immediately after the target answer text is displayed.

[0156] Based on the above embodiments, the server determines that the user's newly asked associated question is indeed a question closely related to the previously asked target question when determining that the relevance between the associated question and the target question is greater than the preset relevance threshold. In this case, the server can reuse the relevant information contained in the intermediate data generated when processing the target question, such as the target question analysis result and the target query statement, to assist in processing the current associated question, so as to reduce the data processing amount while more accurately completing the processing of the associated question and feeding back an associated answer text that better meets the user's needs.

[0157] In specific implementation, the server stores relatively key intermediate data such as question analysis results and query statements generated during processing into a cache each time when processing a query request, so that when an associated query request involving an associated question related to a previously processed question is received within a short period of time, the intermediate data obtained when processing other query requests can be effectively and fully utilized to more efficiently and accurately complete the processing of the associated question. The data in the cache can be set to be automatically cleared when not used again for more than an effective time period from the time of last reuse, or automatically cleared when not used again for more than an effective time period from the time of storage.

[0158] In some embodiments, referring to FIG. 1, the method can further include the following content in specific implementation: Figure 6

[0159] S1: According to a preset protocol rule, obtain resource files related to a target report provided by a plurality of data source servers;

[0160] S2: Perform semantic conversion processing on the plurality of resource files respectively to obtain a plurality of resource semantic texts corresponding to the resource files;

[0161] S3: Determine the text data type of the resource semantic text; wherein the text data type of the resource semantic text includes at least one of the following: structured data text, unstructured data text, and semi-structured data text;

[0162] S4: According to the text data type of the resource semantic text, determine and use a matching knowledge extraction rule to process the resource semantic text to extract corresponding knowledge element data;

[0163] S5: According to a preset fusion rule, perform knowledge fusion using the knowledge element data and a third-party knowledge database to obtain fused knowledge element data;

[0164] ​S6: constructing a corresponding target knowledge graph according to the fused knowledge element data.

[0165] Based on the above embodiments, the server can cooperate with multiple resource parties according to preset protocol rules, obtain and effectively utilize and fuse knowledge data from different data source servers, and construct a relatively comprehensive and effective target knowledge graph.

[0166] In specific implementation, the server can obtain resource files actively provided by multiple data source servers according to preset protocol rules; or can query databases responsible for target reporting by data source servers using corresponding collection credentials to collect the resource files according to preset protocol rules; or can collect and obtain information data related to target reporting published by regulatory agencies or other related agencies through various ways such as official websites and emails to generate the resource files according to preset protocol rules.

[0167] In some embodiments, the above semantic recognition processing of multiple resource files can specifically include the following:

[0168] S1: determining the file type of the current resource file; wherein the file type includes at least one of the following: text file, picture file, video file, audio file;

[0169] S2: determining a target semantic conversion model matched from multiple preset semantic conversion models according to the file type of the current resource file;

[0170] S3: performing semantic recognition on the current resource file using the target semantic conversion model to obtain a current resource semantic text corresponding to the current resource file.

[0171] Before specific implementation, multiple preset semantic conversion models can be trained using sample resource files of different file types. Each preset semantic conversion model corresponds to a file type.

[0172] The above preset semantic conversion model can be used for coarse-grained semantic coarse recognition of resource files of the corresponding file type to quickly find semantic content related to target reporting in the resource files; and convert the part of semantic content into text data convenient for subsequent semantic fine recognition as a resource semantic text corresponding to the resource file.

[0173] In some embodiments, in actual implementation, the resource semantic text can be divided into resource semantic texts of different text data types, such as structured data text (containing structured data), unstructured data text (containing unstructured data), semi-structured data text (containing semi-structured data), and the like, according to the structural features of the data contained in the resource semantic text. Then, different text data types are distinguished, and the resource semantic text is processed by using the matching knowledge extraction rule for targeted extraction, so that the required knowledge element data (or knowledge data) can be extracted more efficiently and accurately.

[0174] The knowledge element data specifically includes one or more combinations of elements. For example, a knowledge element data can include two entity object representation words and an association relationship representation word between the two entity objects.

[0175] In some embodiments, when the text data type of the resource semantic text includes structured data text, the processing of the resource semantic text by using the matching knowledge extraction rule in actual implementation can include the following contents:

[0176] According to the preset field template, the resource semantic text is retrieved, and the text fields associated with the preset identification item fields are extracted from the resource semantic text to be combined to obtain the knowledge element data corresponding to the resource semantic text. The preset field template includes a plurality of preset identification item fields. The knowledge element data at least includes entity object representation words and association relationship representation words.

[0177] Based on the above embodiments, the resource semantic text with structured data text can be processed by using the matching knowledge extraction rule to efficiently and accurately extract the required knowledge element data from the resource semantic text.

[0178] In some embodiments, when the text data type of the resource semantic text includes unstructured data text or semi-structured data text, referring to FIG. 8, Figure 7 The processing of the resource semantic text by using the matching knowledge extraction rule in actual implementation can include the following contents:

[0179] S1: The resource semantic text is subjected to semantic recognition by using a preset semantic recognition model to obtain a corresponding resource semantic text recognition result;

[0180] S2: The resource semantic text recognition result is subjected to word segmentation processing to obtain a plurality of word segmentation units, and the grammatical relationship features of the word segmentation units based on the resource semantic text recognition result are determined;

[0181] S3: identifying and extracting, from the plurality of segmented units, knowledge element data corresponding to the resource semantic text based on the syntax relationship features of the resource semantic text recognition result and the preset field template.

[0182] The preset semantic recognition model can be understood as an algorithm model that is trained in advance using a large amount of sample resource semantic text of unstructured data text and semi-structured data text, and can perform fine-grained semantic recognition on input resource semantic text to obtain and output the required semantic recognition result related to the target report.

[0183] Based on the above embodiments, the resource semantic text of unstructured data text or semi-structured data text can be processed by using matching knowledge rules to efficiently and accurately extract the required knowledge element data from the resource semantic text.

[0184] In some embodiments, the knowledge element data and the third-party knowledge database are fused according to the preset fusion rule to obtain fused knowledge element data, as shown in Figure 8 The specific implementation can include the following contents:

[0185] S1: acquiring third-party knowledge element data from the third-party knowledge database according to the preset fusion rule;

[0186] S2: calculating the vector distance between the semantic feature vectors of the entity object representation words in different knowledge element data;

[0187] S3: determining the entity object representation words indicating the same entity object according to the vector distance between the semantic feature vectors of the entity object representation words and the third-party knowledge element data; and performing first fusion processing on the knowledge element data containing the entity object representation words indicating the same entity object to obtain corresponding first fused knowledge element data;

[0188] S4: performing second fusion processing on the first fused knowledge element data and the third-party knowledge element data to obtain corresponding second fused knowledge element data as the fused knowledge element data.

[0189] Based on the above embodiments, the server can first perform first fusion on the extracted knowledge element data to obtain first fused knowledge element data; then perform second fusion on the first fused knowledge element data and the third-party knowledge element data in the third-party database held by the cooperative party to obtain second fused knowledge element data, so as to obtain more comprehensive and effective fused knowledge element data through multiple fusions.

[0190] In specific implementation, the entity alignment can be performed on the entity object representation words according to the semantic feature vectors of the entity object representation words. Then, according to the alignment result, the entity object representation words with the same semantic feature vectors can be filtered out as the entity object representation words indicating the same entity object. Further, according to the alignment result, the entity object representation words with similar semantic feature vectors can be filtered out as potential entity object representation words. Then, taking the third-party knowledge element data as a reference, the entity object representation words with the most similar semantic feature vectors among the potential entity object representation words can be filtered out as the entity object representation words indicating the same entity object according to the vectors between the semantic feature vectors of the potential entity object representation words. Then, the knowledge element data containing the entity object representation words indicating the same entity object can be subjected to the first fusion processing.

[0191] In a similar manner, the first fused knowledge element data and the third-party knowledge element data are processed to find the entity object representation words indicating the same entity object, and the first fused knowledge element data and the third-party knowledge element data containing the entity object representation words indicating the same entity object are subjected to the second fusion processing.

[0192] After obtaining the fused knowledge element data, the fused knowledge element data can be stored in the knowledge database of the party for backup. Meanwhile, according to the corresponding cooperation agreement, the knowledge database can be shared with the cooperative party.

[0193] The first fusion processing and the second fusion processing at least include data merging, anaphora resolution and other related processing on the knowledge element data containing the entity object representation words indicating the same entity object.

[0194] In some embodiments, considering that the target submission in the financial service field often has high precision requirements, after obtaining the fused knowledge element data, the method can further include the following contents in specific implementation: sending the fused knowledge element data to a target submission expert for manual detection; or, performing machine detection on the fused knowledge element data according to a preset experience rule. The preset experience rule is generated based on the detection experience and historical detection records of the staff. Thus, the fused knowledge element data can have high precision.

[0195] In some embodiments, the construction of the corresponding target knowledge graph according to the fused knowledge element data can include the following contents in specific implementation:

[0196] S1: constructing an initial knowledge graph according to the fused knowledge element data;

[0197] S2: obtaining extended data of the knowledge graph through knowledge processing based on the initial knowledge graph; wherein the knowledge processing comprises at least one of the following: reasoning and extension based on existing nodes and existing connection edges in the initial knowledge graph; prediction and extension based on attribute information of existing nodes and existing connection edges in the initial knowledge graph; prediction and extension based on existing graph structure and existing connection edges in the initial knowledge graph.

[0198] S3: adjusting the initial knowledge graph based on the extended data of the knowledge graph to obtain a corresponding target knowledge graph.

[0199] Based on the above embodiments, a relatively more comprehensive and detailed target knowledge graph can be constructed through relevant knowledge processing based on the fused knowledge element data.

[0200] In some embodiments, in specific implementation, first, nodes corresponding to entity objects can be marked in the graph based on entity object representation words in the fused knowledge element data; and corresponding node identifiers can be marked at the nodes based on the entity object representation words. Then, two nodes having an association relationship can be connected by a connection edge based on association relationship representation words in the fused knowledge element data. In addition, attribute information of an entity object corresponding to a node can be marked at the node based on the attribute information of the entity object. Attribute information of a connection edge corresponding to the connection edge can be marked on the connection edge based on the attribute information of the connection edge; and / or, a corresponding connection edge can be determined in the graph based on an event condition; and the corresponding event condition can be marked on the connection edge. Thus, the initial knowledge graph can be obtained.

[0201] After the initial knowledge graph is obtained, the initial knowledge graph can be further processed. Specifically, on the one hand, based on reasoning and extension of existing nodes and existing connection edges in the initial knowledge graph, based on existing information of the initial knowledge graph, data information that can be explicitly determined to exist but is not directly represented in the knowledge graph can be extended and completed accordingly. On the other hand, based on prediction and extension of attribute information of existing nodes and existing connection edges in the initial knowledge graph; and / or, based on prediction and extension of existing graph structure and existing connection edges in the initial knowledge graph, potential data information that has a great probability of existing in the knowledge graph can be predicted and extended accordingly.

[0202] In the specific knowledge processing process, after the extended data is obtained through the knowledge processing, the method further includes: performing quality evaluation on the extended data and the knowledge processing process according to the initial knowledge graph; determining whether to adopt the extended data according to the evaluation result; or whether to further process the extended data. Thus, the quality of the extended data can be effectively improved, and the quality of the constructed target knowledge graph can be improved.

[0203] In some embodiments, after the corresponding target knowledge graph is constructed according to the fused knowledge element data, the method can further include the following content when implemented:

[0204] S1: Obtain the resource file updated in the current preset time period and the knowledge element data updated in the current preset time period of the third-party knowledge database every preset time period;

[0205] S2: Update the target knowledge graph according to the resource file updated in the current preset time period and the knowledge element data updated in the current preset time period of the third-party knowledge database.

[0206] Based on the above embodiments, the target knowledge graph can be updated regularly to ensure that the target knowledge graph maintains high precision and wide coverage.

[0207] In some embodiments, after the corresponding target answer text is fed back according to the query result, the method can further include: the user uses the user terminal to process the corresponding target submission service according to the target answer text.

[0208] As can be seen from the above, based on the target report data processing method based on the knowledge graph provided by the embodiments of the present specification, before the specific implementation, the server can interact with the resource party and the cooperative party according to the preset protocol rules, obtain the resource files related to the target report provided by the plurality of data source servers, and combine the third-party knowledge database of the cooperative party to construct a relatively comprehensive and accurate target knowledge graph for the target report business scene through knowledge fusion. When the user is processing the target report business and encounters a target problem related to the target report, the server can initiate a target query request. The server receives and responds to the target query request, first distinguishes different problem types, and according to the problem type, adopts a matching mode to process the target problem text carried by the target query request, and obtains the corresponding target problem analysis result; then according to the target problem analysis result, the corresponding target query statement matched with the target knowledge graph is constructed through semantic slot extraction; and then according to the target query statement, the target knowledge graph is queried to feed back the corresponding target answer text to the user in time. Thus, the processing difficulty of the user when processing the target report business can be effectively reduced, the user can be assisted to efficiently and accurately complete the related target report business, and the error in the target report can be reduced.

[0209] The embodiments of the present specification also provide a server, which is shown in Figure 9 The server includes a network communication port 901, a processor 902, and a memory 903, which are connected by internal cables so that each structure can perform specific data interaction.

[0210] The network communication port 901 can be specifically used to receive a target query request; wherein the target query request carries at least a target problem text of a target problem related to the target report.

[0211] The processor 902 can be specifically configured to respond to the target query request, process the target question text by using a preset classifier, determine a question type of the target question, determine a matched preset question analysis model as a target question analysis model from a plurality of preset question analysis models according to the question type of the target question, process the target question text by using the target question analysis model to obtain a corresponding target question analysis result, wherein the target question analysis result at least includes an entity object analysis result, perform semantic slot extraction according to the target question analysis result to obtain a corresponding semantic slot extraction result, construct a target query sentence matched with a target knowledge graph based on the semantic slot extraction result, wherein the target knowledge graph is constructed by a server according to a plurality of data source servers providing resource files related to the target report according to a preset protocol rule, in combination with a third-party knowledge database of a cooperative party through knowledge fusion, query the target knowledge graph according to the target query sentence, and feed back a corresponding target answer text according to a query result.

[0212] The memory 903 can be specifically configured to store corresponding instruction programs.

[0213] In the embodiment, the network communication port 901 can be a virtual port that is bound with different communication protocols, so as to send or receive different data. For example, the network communication port can be a port responsible for web data communication, can be a port responsible for FTP data communication, and can be a port responsible for mail data communication. In addition, the network communication port can also be an entity communication interface or a communication chip. For example, it can be a wireless mobile network communication chip such as GSM, CDMA, etc.; it can also be a Wifi chip; and it can also be a Bluetooth chip.

[0214] In the embodiment, the processor 902 can be implemented in any appropriate manner. For example, the processor can take the form of, for example, a microprocessor or processor and a computer readable medium storing computer readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. The present specification is not limited thereto.

[0215] In the embodiment, the memory 903 can include a plurality of levels. In a digital system, as long as it can save binary data, it can be a memory; in an integrated circuit, a circuit without a physical form and having a storage function is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, a TF card, etc.

[0216] The embodiment of the present specification also provides a computer readable storage medium based on the above knowledge graph-based target reporting data processing method, the computer readable storage medium stores computer program instructions, and the following steps are implemented when the computer program instructions are executed: receiving a target query request; wherein the target query request carries at least a target problem text of a target problem related to a target report; in response to the target query request, processing the target problem text by using a preset classifier to determine a problem type of the target problem; according to the problem type of the target problem, determining a matched preset problem analysis model from a plurality of preset problem analysis models as a target problem analysis model; processing the target problem text by using the target problem analysis model to obtain a corresponding target problem analysis result; wherein the target problem analysis result at least includes an entity object analysis result; according to the target problem analysis result, performing semantic slot extraction to obtain a corresponding semantic slot extraction result; and based on the semantic slot extraction result, constructing a target query sentence matched with a target knowledge graph; wherein the target knowledge graph is constructed by a server according to a plurality of data source servers providing resource files related to the target report according to a preset protocol rule, in combination with a third-party knowledge database of a cooperation party through knowledge fusion; querying the target knowledge graph according to the target query sentence; and feeding back a corresponding target answer text according to a query result.

[0217] In the embodiment, the storage medium includes but is not limited to a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk drive (HDD) or a memory card. The memory can be used to store computer program instructions. The network communication unit can be an interface set according to a standard specified by a communication protocol, used for network connection communication.

[0218] In the embodiment, the functions and effects of the program instructions stored in the computer readable storage medium can be explained by comparing with other embodiments, and will not be described here.

[0219] Referring to Figure 10 In the software layer, the embodiment of the present specification also provides a knowledge graph-based target reporting data processing device, which specifically can include the following structure modules:

[0220] The receiving module 1001 can be specifically used for receiving a target query request; wherein the target query request carries at least a target problem text of a target problem related to a target report;

[0221] The classification module 1002 can be specifically configured to respond to the target query request, process the target question text by using a preset classifier, and determine a question type of the target question.

[0222] The determination module 1003 can be specifically configured to determine a preset question analysis model matched from a plurality of preset question analysis models according to the question type of the target question, as a target question analysis model.

[0223] The processing module 1004 can be specifically configured to process the target question text by using the target question analysis model to obtain a corresponding target question analysis result, wherein the target question analysis result at least includes an entity object analysis result.

[0224] The construction module 1005 can be specifically configured to perform semantic slot extraction according to the target question analysis result to obtain a corresponding semantic slot extraction result, and construct a target query sentence matched with a target knowledge graph based on the semantic slot extraction result, wherein the target knowledge graph is constructed by a server according to a plurality of data source servers providing resource files related to the target report according to a preset protocol rule, in combination with a third-party knowledge database of a cooperation party through knowledge fusion.

[0225] The query module 1006 can be specifically configured to query the target knowledge graph according to the target query sentence, and feed back a corresponding target answer text according to a query result.

[0226] In some embodiments, the question type can specifically include a question related to a single entity object, a question related to a plurality of entity objects, and the like.

[0227] Correspondingly, in the case where the question type of the target question includes a question related to a plurality of entity objects, the device can process the target question text by using the target question analysis model to obtain a corresponding target question analysis result in the following manner: performing word segmentation processing on the target question text by using the target question analysis model to obtain a word group sequence including a plurality of word segmentation units sorted in sequence; determining a part of speech of each word segmentation unit based on the word group sequence, and setting a corresponding part of speech label to obtain an annotated word group sequence; determining a plurality of entity object representation words in the annotated word group sequence according to the part of speech label; determining a relative position parameter between other word segmentation units and the entity object representation words in the annotated word group sequence except for the entity object representation words; processing the other word segmentation units according to the relative position parameter and the part of speech label to determine association relationship representation words; and combining the entity object representation words and the association relationship representation words to obtain the corresponding target question analysis result.

[0228] In some embodiments, the device, when specifically implemented, can construct a target query statement matching the target knowledge graph based on the semantic slot extraction result in the following manner: filtering out a matching query statement template according to the semantic slot extraction result; and establishing an initial query statement based on the query statement template; obtaining a target semantic link based on the target knowledge graph; wherein the target semantic link at least contains a url link of an entity object related to the target question; and assembling the initial query statement using the target semantic link to obtain a corresponding target query statement.

[0229] In some embodiments, the target knowledge graph can at least contain a plurality of nodes and connection edges between the nodes; wherein the nodes are used to represent entity objects in the target reporting business scenario; the connection edges are used to represent the association relationship between the connected entity objects based on the target reporting business; and the entity objects include at least one of the following: a data object to be reported, an initiating object of the data object to be reported, a receiving object of the data object to be reported, and a next processing flow of the data object to be reported.

[0230] In some embodiments, the device, when specifically implemented, can query the target knowledge graph according to the target query statement in the following manner; and feed back the corresponding target answer text according to the query result: locating an initial node in the target knowledge graph according to the url link of the entity object in the target query statement; retrieving other nodes directly and indirectly connected with the initial node and the connection edges between the initial node and the connected other nodes according to the target query statement to obtain a corresponding query result; processing the query result using a preset semantic processing model to obtain a target answer text based on natural language; and feeding back the target answer text.

[0231] In some embodiments, the nodes in the target knowledge graph are also marked with attribute information related to the entity objects represented by the nodes; and the connection edges in the target knowledge graph are also marked with event conditions related to the association relationship represented by the connection edges.

[0232] Correspondingly, the device, when specifically implemented, can retrieve other nodes directly and indirectly connected with the initial node and the connection edges between the initial node and the connected other nodes according to the target query statement to obtain a corresponding query result in the following manner: retrieving other nodes directly and indirectly connected with the initial node, attribute information marked by the initial node and the other nodes, and the connection edges between the initial node and the connected other nodes, the event conditions marked by the connection edges according to the target query statement to obtain a corresponding query result.

[0233] In some embodiments, after feeding back the corresponding target answer text according to the query result, the device, when specifically implemented, can also be used to: receive an associated query request; wherein the associated query request carries at least an associated question text related to the associated question; the associated query request is a query request initiated by the user terminal for the target answer text; detect whether the relevance between the associated question and the target question is greater than a preset relevance threshold; in the case where it is determined that the relevance between the associated question and the target question is greater than the preset relevance threshold, obtain and generate an associated question analysis result according to the target question analysis result; construct an associated query statement matched with the target knowledge graph according to the associated question analysis result and the target query statement; query the target knowledge graph according to the associated query statement; and feed back the corresponding associated answer text according to the query result and the target answer text.

[0234] In some embodiments, the device, when specifically implemented, can also be used to: obtain resource files related to the target report according to a preset protocol rule and provided by a plurality of data source servers; respectively perform semantic conversion processing on the plurality of resource files to obtain a plurality of resource semantic texts corresponding to the resource files; determine the text data type of the resource semantic text; wherein the text data type of the resource semantic text includes at least one of the following: structured data text, unstructured data text, and semi-structured data text; determine and use a matched knowledge extraction rule to process the resource semantic text according to the text data type of the resource semantic text, to extract corresponding knowledge element data; perform knowledge fusion using the knowledge element data and a third-party knowledge database according to a preset fusion rule to obtain fused knowledge element data; and construct a corresponding target knowledge graph according to the fused knowledge element data.

[0235] In some embodiments, in the case where the text data type of the resource semantic text includes structured data text, the device, when specifically implemented, can use a matched knowledge extraction rule to process the resource semantic text in the following manner: retrieve the resource semantic text according to a preset field template; and combine the text fields associated with the preset identification item fields extracted from the resource semantic text to obtain the knowledge element data corresponding to the resource semantic text; wherein the preset field template contains a plurality of preset identification item fields; and the knowledge element data at least includes an entity object representation word and an associated relationship representation word.

[0236] In some embodiments, when the text data type of the resource semantic text comprises unstructured data text or semi-structured data text, the device, when specifically implemented, can process the resource semantic text according to the matched knowledge extraction rule in the following manner: performing semantic recognition on the resource semantic text by using a preset semantic recognition model to obtain a corresponding resource semantic text recognition result; performing word segmentation processing on the resource semantic text recognition result to obtain a plurality of word segmentation units; determining the grammatical relationship features of the word segmentation units based on the resource semantic text recognition result; and according to the preset field template and the grammatical relationship features of the word segmentation units based on the resource semantic text recognition result, identifying and extracting knowledge element data corresponding to the resource semantic text from the plurality of word segmentation units.

[0237] In some embodiments, the device, when specifically implemented, can perform knowledge fusion according to the preset fusion rule by using the knowledge element data and the third-party knowledge database to obtain fused knowledge element data in the following manner: acquiring third-party knowledge element data from the third-party knowledge database according to the preset fusion rule; calculating the vector distance between semantic feature vectors of entity object representation words in different knowledge element data; determining entity object representation words indicating the same entity object according to the vector distance between the semantic feature vectors of the entity object representation words and the third-party knowledge element data; and performing first fusion processing on the knowledge element data containing the entity object representation words indicating the same entity object to obtain corresponding first fused knowledge element data; and performing second fusion processing on the first fused knowledge element data and the third-party knowledge element data to obtain corresponding second fused knowledge element data as the fused knowledge element data.

[0238] In some embodiments, the device, when specifically implemented, can construct a corresponding target knowledge graph according to the fused knowledge element data in the following manner: constructing an initial knowledge graph according to the fused knowledge element data; obtaining extension data of the knowledge graph through knowledge processing based on the initial knowledge graph; wherein the knowledge processing comprises at least one of the following: reasoning and extension based on existing nodes and existing connection edges in the initial knowledge graph; prediction and extension based on attribute information of existing nodes and existing connection edges in the initial knowledge graph; prediction and extension based on existing graph structures and existing connection edges in the initial knowledge graph; and adjusting the initial knowledge graph according to the extension data of the knowledge graph to obtain a corresponding target knowledge graph.

[0239] It should be noted that the units, devices or modules and the like illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described as various modules respectively described in terms of functions. Of course, in the implementation of the present specification, the functions of each module can be implemented in the same or more software and / or hardware, or the modules implementing the same function can be implemented by a combination of multiple sub-modules or sub-units, etc. The above described device embodiments are only illustrative, for example, the division of the units is only a logical function division, and in actual implementation, there can be another division mode, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the coupling or direct coupling or communication connection between the units or devices shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0240] As can be seen from the above, the knowledge graph-based target reporting data processing device provided in the embodiments of the present specification can effectively reduce the processing difficulty of the user when processing the target reporting business, assist the user to efficiently and accurately complete the related target reporting business, and reduce the errors in target reporting.

[0241] In a specific scene example, the knowledge graph-based target reporting data processing method provided in the present specification can be applied to construct and apply a knowledge graph-based intelligent question and answer system in the target reporting scene. The specific implementation process can be referred to the following content.

[0242] Among them, the above-mentioned knowledge graph-based question and answer system (Knowledge-Based Question Answering system, KBQA) can specifically refer to an intelligent system that can parse and convert a question into a query in a knowledge graph, filter the results after querying, and translate the results into a result convenient for human understanding.

[0243] In the present scene example, the constructed knowledge graph-based intelligent question and answer system in the target reporting field can help the target reporting related personnel (for example, the user) to continuously track the relevant rule change situation and internalize it into the industry system in a timely manner through the knowledge graph as data support in the form of question and answer; and find the defects of the system in order to carry out system management and ensure the effective implementation of the relevant requirements (complete the target reporting business responsible). The above-mentioned knowledge graph-based intelligent question and answer system in the target reporting field, as an auxiliary system for bank (a kind of financial service institution) data reporting, can be independent or nested in the original reporting process, used for bank target reporting knowledge retrieval, helping the user to quickly understand the target reporting related information, avoiding the data reporting errors caused by unclear understanding of the relevant requirements and system, and causing adverse effects on the reputation of the bank.

[0244] For specific implementation, please refer to Figure 11 As shown, the bank server can connect to multiple different regulatory reporting systems (e.g., resource servers) to acquire and construct a target reporting knowledge graph (e.g., a target knowledge graph) and an intelligent question-answering system based on a large amount of relevant data (e.g., resource files related to the target reporting). The target reporting knowledge graph, as core data, serves as the support for the intelligent question-answering system and can also be directly accessed and used by other target reporting systems. The intelligent question-answering system can be provided as an independent system or functional module for rapid integration by various target reporting systems.

[0245] When constructing the knowledge graph for the aforementioned objectives, please refer to [reference needed]. Figure 12 As shown. Specifically, knowledge sources can include: 1. Documents such as regulations, reporting requirements, and reporting standards issued by regulatory agencies (e.g., the State Financial Supervision and Administration Bureau, the National Audit Office, the State Administration of Foreign Exchange, etc.); 2. Internal management systems and data tables from the bank's business systems.

[0246] When performing knowledge extraction, one can first use natural language processing to automatically extract semi-structured or unstructured data, and then manually extract entities, relationships, and attributes. Alternatively, structured data (such as standard data reporting tables) can be directly imported into a graph database for knowledge extraction.

[0247] When performing knowledge fusion, knowledge input can be obtained from third-party knowledge base products or existing structured data when constructing a knowledge graph.

[0248] When performing specific knowledge processing, knowledge graph (initial knowledge graph) completion and reasoning can be carried out in the following ways: completion based on ontology reasoning; link prediction based on representation and knowledge graph embedding; and completion of the graph based on graph structure and relation path features.

[0249] Once the aforementioned knowledge graph is established, it can be used for retrieval, enabling applications such as intelligent question answering. Knowledge graphs and semantic technologies can also be used to assist in data analysis and decision-making. For details, please refer to [link to relevant documentation]. Figure 13 As shown.

[0250] The problem classifier can be used for problem classification (determining the problem type) of input question text (e.g., target question text). The problem analysis can include, in sequence, word segmentation, part-of-speech tagging, entity recognition, and entity disambiguation (to obtain a corresponding target question analysis result). Based on the semantic slot extraction, category discrimination (e.g., determining the query intent type) can be performed. Based on the PARQL extraction, an SQPARQL template can be constructed (to obtain an initial query statement). Based on the semantic linking, the natural language expression to be linked in the semantic slot can be linked to the URL in the knowledge graph corresponding to the <category, source, entity> (to obtain the URL link of the entity object). Based on the SPARQL query generation, a standard SPARQL query can be constructed (to obtain a target query statement) using the query generation module, the problem category, and the connected entity. The result generation can include obtaining the corresponding result (query result) returned by the SPARQL query and converting it into a form (target answer text) that is easy for the user to understand.

[0251] Based on the above scenario example, it is verified that the target reporting data processing method based on the knowledge graph provided in the specification can significantly improve user experience by performing knowledge retrieval in a question and answer manner, helping users understand target reporting related knowledge in a question and answer manner; through interactive exploration analysis, it can help ordinary users to discover, verify and reason, and they can complete the whole process by themselves without the assistance of professionals; through the use of interactive machine learning technology, new knowledge can be continuously integrated with old knowledge to actively discover the similarities and differences of the requirements of different fields of target reporting lines; by using the graphic database storage method, compared with the traditional storage method, the data retrieval speed is faster, the real-time response of human-computer interaction is realized, and the user can make instant decisions.

[0252] Although the description has been provided with reference to the method operations that are illustrated in the examples or flowcharts, additional or fewer operations can be included based on the conventional or non-creative means. The order of the steps listed in the examples is only one of the many ways of executing the steps, and does not represent the only way of executing the steps. In actual implementation of the device or client product, the method steps can be executed in sequence or in parallel (for example, in a parallel processor or a multi-threaded processing environment, or even in a distributed data processing environment). The terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, product or device that includes a list of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, product or device. Without more limitations, it does not exclude the presence of other same or equivalent elements in the process, method, product or device that includes the elements. The terms "first", "second" and the like are used to indicate names, not any particular order.

[0253] Those skilled in the art also know that, in addition to implementing the controller in the form of pure computer readable program code, the same function can be achieved by logically programming the method steps in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers. Therefore, such a controller can be considered as a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0254] The description can be described in the general context of computer-executable instructions, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, and the like that perform particular tasks or implement particular abstract data types. The description can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are connected through a communication network. In a distributed computing environment, program modules can be located in both local and remote computer-readable storage media, including storage devices.

[0255] Those skilled in the art can clearly understand the present specification can be implemented by means of software and necessary general hardware platforms through the above description of the embodiments. Based on such understanding, the technical solutions of the present specification can essentially be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) execute the methods described in the various embodiments or some parts of the embodiments of the present specification.

[0256] The various embodiments in the present specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. The present specification can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, small computers, large computers, distributed computing environments including any of the above systems or devices, etc.

[0257] Although the present specification is described through the embodiments, those skilled in the art know that the present specification has many modifications and changes without departing from the spirit of the present specification, and it is intended that the appended claims include these modifications and changes without departing from the spirit of the present specification.

Claims

1. A knowledge graph-based target submission data processing method, characterized in that, The application is applied to a server, comprising: receiving a target query request; wherein the target query request carries at least a target question text of a target question related to a target report; in response to the target query request, processing the target question text by using a preset classifier to determine a question type of the target question; determining a matching preset question analysis model from a plurality of preset question analysis models as a target question analysis model according to the question type of the target question; processing the target question text by using the target question analysis model to obtain a corresponding target question analysis result; wherein the target question analysis result at least includes an entity object analysis result; performing semantic slot extraction according to the target question analysis result to obtain a corresponding semantic slot extraction result; and constructing a target query sentence matched with a target knowledge graph based on the semantic slot extraction result; wherein the target knowledge graph is constructed by a server according to a plurality of data source servers providing resource files related to the target report, in combination with a third-party knowledge database of a partner, through knowledge fusion according to a preset protocol rule; querying the target knowledge graph according to the target query sentence; and feeding back a corresponding target answer text according to a query result; after feeding back the corresponding target answer text according to the query result, the method further comprises: receiving an associated query request; wherein the associated query request carries at least an associated question text related to an associated question; the associated query request is a query request initiated by a user terminal for the target answer text; detecting whether a correlation degree between the associated question and the target question is greater than a preset correlation degree threshold; in a case where it is determined that the correlation degree between the associated question and the target question is greater than the preset correlation degree threshold, obtaining and generating an associated question analysis result according to the target question analysis result; constructing an associated query sentence matched with the target knowledge graph according to the associated question analysis result and the target query sentence; querying the target knowledge graph according to the associated query sentence; and feeding back a corresponding associated answer text according to a query result and the target answer text.

2. The method of claim 1, wherein, The question type includes a question related to a single entity object and a question related to a plurality of entity objects; correspondingly, in a case where the question type of the target question includes the question related to the plurality of entity objects, processing the target question text by using the target question analysis model to obtain the corresponding target question analysis result includes: performing word segmentation processing on the target question text by using the target question analysis model to obtain a word group sequence containing a plurality of word segmentation units sorted in sequence; determining a part of speech of each word segmentation unit based on the word group sequence, and setting a corresponding part of speech label to obtain an annotated word group sequence; determining a plurality of entity object representation words in the annotated word group sequence according to the part of speech label; determining a relative position parameter between other word segmentation units and the entity object representation words in the annotated word group sequence except for the entity object representation words; processing the other word segmentation units according to the relative position parameter and the part of speech label to determine associated relationship representation words; Combine the entity object representation word and the association relationship representation word to obtain a corresponding target question analysis result.

3. The method of claim 1, wherein, Based on the semantic slot extraction result, a target query sentence matching the target knowledge graph is constructed, including: According to the semantic slot extraction result, a matching query sentence template is screened out; and an initial query sentence is established based on the query sentence template; Obtain a target semantic link based on the target knowledge graph; wherein the target semantic link at least contains a url link of an entity object related to the target question; Assemble the initial query sentence using the target semantic link to obtain a corresponding target query sentence.

4. The method of claim 3, wherein, The target knowledge graph at least contains a plurality of nodes and connection edges between the nodes; wherein the nodes are used to represent entity objects in a target reporting business scenario; the connection edges are used to represent the association relationship between the connected entity objects based on the target reporting business; the entity objects include at least one of the following: a data object to be reported, an initiating object of the data object to be reported, a receiving object of the data object to be reported, and a next processing flow of the data object to be reported.

5. The method of claim 4, wherein, According to the target query sentence, the target knowledge graph is queried; and according to the query result, a corresponding target answer text is fed back, including: According to the url link of the entity object in the target query sentence, an initial node is located in the target knowledge graph; According to the target query sentence, other nodes directly connected and indirectly connected with the initial node, and the connection edges between the initial node and the connected other nodes are retrieved to obtain a corresponding query result; Using a preset semantic processing model to process the query result, a target answer text based on natural language is obtained; and the target answer text is fed back.

6. The method of claim 4, wherein, The nodes in the target knowledge graph are also marked with attribute information related to the entity objects represented by the nodes; and the connection edges in the target knowledge graph are also marked with event conditions related to the association relationship represented by the connection edges. Accordingly, according to the target query sentence, other nodes directly connected and indirectly connected with the initial node, and the connection edges between the initial node and the connected other nodes are retrieved to obtain a corresponding query result, including: According to the target query sentence, other nodes directly connected and indirectly connected with the initial node, the attribute information marked by the initial node and the other nodes, and the connection edges between the initial node and the connected other nodes, and the event conditions marked by the connection edges are retrieved to obtain a corresponding query result.

7. The method of claim 1, wherein, The method further comprises: According to a preset protocol rule, a plurality of resource files related to the target reporting provided by a plurality of data source servers are obtained; The plurality of resource files are respectively subjected to semantic conversion processing to obtain a plurality of resource semantic texts corresponding to the resource files; Determine the text data type of the resource semantic text; wherein the text data type of the resource semantic text includes at least one of the following: structured data text, unstructured data text, and semi-structured data text; According to the text data type of the resource semantic text, a matching knowledge extraction rule is determined and used to process the resource semantic text to extract corresponding knowledge element data; According to the preset fusion rule, knowledge fusion is performed on the knowledge element data and the third-party knowledge database to obtain fused knowledge element data. According to the fused knowledge element data, a corresponding target knowledge graph is constructed.

8. The method of claim 7, wherein, In the case where the text data type of the resource semantic text includes structured data text, the resource semantic text is processed by using a matching knowledge extraction rule, including: According to the preset field template, the resource semantic text is retrieved, and text fields associated with the preset identification item fields are extracted from the resource semantic text to obtain knowledge element data corresponding to the resource semantic text; wherein the preset field template includes a plurality of preset identification item fields; the knowledge element data at least includes entity object representation words and associated relationship representation words.

9. The method of claim 7, wherein, In the case where the text data type of the resource semantic text includes unstructured data text or semi-structured data text, the resource semantic text is processed by using a matching knowledge extraction rule, including: The resource semantic text is subjected to semantic recognition by using a preset semantic recognition model to obtain a corresponding resource semantic text recognition result. The resource semantic text recognition result is subjected to word segmentation processing to obtain a plurality of word segmentation units, and the syntax relationship features of the word segmentation units based on the resource semantic text recognition result are determined. According to the preset field template and the syntax relationship features of the word segmentation units based on the resource semantic text recognition result, knowledge element data corresponding to the resource semantic text is identified and extracted from the plurality of word segmentation units.

10. The method of claim 7, wherein, According to the preset fusion rule, knowledge fusion is performed on the knowledge element data and the third-party knowledge database to obtain fused knowledge element data, including: According to the preset fusion rule, third-party knowledge element data is obtained from the third-party knowledge database; The vector distance between the semantic feature vectors of the entity object representation words in different knowledge element data is calculated. According to the vector distance between the semantic feature vectors of the entity object representation words and the third-party knowledge element data, entity object representation words indicating the same entity object are determined, and knowledge element data containing the entity object representation words indicating the same entity object are subjected to first fusion processing to obtain corresponding first fused knowledge element data. The first fused knowledge element data and the third-party knowledge element data are subjected to second fusion processing to obtain corresponding second fused knowledge element data as the fused knowledge element data.

11. The method of claim 7, wherein, According to the fused knowledge element data, a corresponding target knowledge graph is constructed, including: According to the fused knowledge element data, an initial knowledge graph is constructed. According to the initial knowledge graph, knowledge processing is performed to obtain extension data of the knowledge graph; wherein the knowledge processing includes at least one of the following: reasoning and extension based on existing nodes and existing connection edges in the initial knowledge graph; prediction and extension based on attribute information of existing nodes and existing connection edges in the initial knowledge graph; prediction and extension based on existing graph structure and existing connection edges in the initial knowledge graph; According to the extended data of the knowledge graph, the initial knowledge graph is adjusted to obtain a corresponding target knowledge graph. 12.A knowledge graph-based target submission data processing apparatus, characterized by comprising: Applied to a server, comprising: A receiving module is configured to receive a target query request, wherein the target query request carries at least a target problem text of a target problem related to a target report; A classification module is configured to respond to the target query request, process the target problem text by using a preset classifier, and determine a problem type of the target problem; A determination module is configured to determine a matching preset problem analysis model from a plurality of preset problem analysis models as a target problem analysis model according to the problem type of the target problem; A processing module is configured to process the target problem text by using the target problem analysis model to obtain a corresponding target problem analysis result, wherein the target problem analysis result at least includes an entity object analysis result; A construction module is configured to perform semantic slot extraction according to the target problem analysis result to obtain a corresponding semantic slot extraction result, and construct a target query sentence matched with a target knowledge graph based on the semantic slot extraction result, wherein the target knowledge graph is constructed by a server according to a plurality of data source servers providing resource files related to the target report, in combination with a third-party knowledge database of a partner, through knowledge fusion according to a preset protocol rule; A query module is configured to query the target knowledge graph according to the target query sentence, and feed back a corresponding target answer text according to a query result; after feeding back the corresponding target answer text according to the query result, receive an associated query request, wherein the associated query request carries at least an associated problem text related to an associated problem; the associated query request is a query request initiated by a user terminal for the target answer text; detect whether a correlation degree between the associated problem and the target problem is greater than a preset correlation degree threshold; in a case where it is determined that the correlation degree between the associated problem and the target problem is greater than the preset correlation degree threshold, obtain and generate an associated problem analysis result according to the target problem analysis result; construct an associated query sentence matched with the target knowledge graph according to the associated problem analysis result and the target query sentence; query the target knowledge graph according to the associated query sentence; and feed back a corresponding associated answer text according to a query result and the target answer text.

13. A server, characterized by A processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the steps of the method of any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, A computer instruction is stored thereon, and the instruction is executed by a processor to implement the steps of the method of any one of claims 1 to 11.

Images