A background investigation report quality inspection method and device and related training method

Abstract

This application discloses a quality inspection method, apparatus, and related training method for background investigation reports. The method includes: acquiring target background investigation materials and a target background investigation report regarding a target object; wherein the target background investigation report is generated based on the target background investigation materials; constructing a first target event relationship diagram corresponding to the target background investigation materials and a second target event relationship diagram corresponding to the target background investigation report, respectively, based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram; and performing quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the target quality inspection result of the target background investigation report. Through the above methods, this application can improve the accuracy of quality inspection of background investigation reports.

Description

Technical Field

[0001] This application relates to the field of background investigation technology, and in particular to a quality inspection method, apparatus and related training method for background investigation reports. Background Technology

[0002] Currently, the quality inspection or quality control of background investigation reports is usually carried out after the report is generated, mainly through manual inspection. Manual inspection is highly dependent on the experience level and subjective understanding of the inspectors, and the inspection conclusions lack unified quantitative standards. Therefore, the accuracy and efficiency of the inspection are low. Summary of the Invention

[0003] The main technical problem addressed by this application is to provide a quality inspection method, apparatus, and related training method for background investigation reports, which can improve the accuracy of quality inspection of background investigation reports.

[0004] The first aspect of this application provides a quality inspection method for a background investigation report. The method includes: obtaining target background investigation materials and a target background investigation report about a target object; wherein the target background investigation report is generated based on the target background investigation materials; constructing a first target event relationship diagram corresponding to the target background investigation materials and a second target event relationship diagram corresponding to the target background investigation report based on the target background investigation materials and the target background investigation report, respectively; and performing quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain a target quality inspection result for the target background investigation report.

[0005] The target quality inspection result includes sub-quality inspection results of at least one quality inspection dimension, wherein the at least one quality inspection dimension includes at least one of the following: completeness, consistency, reasonableness of conclusion, and objectivity of expression; and / or, the sub-quality inspection result includes quality inspection score.

[0006] The step of constructing a first target event relationship diagram corresponding to the target background investigation material and a second target event relationship diagram corresponding to the target background investigation report based on the target background investigation material and the target background investigation report, respectively, includes: determining at least one target background investigation event related to the target object from a reference text; wherein the reference text is the target background investigation material or the target background investigation report; for each target background investigation event, identifying target event entities related to the target background investigation event from the reference text; constructing a reference sub-relationship diagram corresponding to the target background investigation event using the relationships between the target event entities; and forming the first target event relationship diagram or the second target event relationship diagram using the reference sub-relationship diagrams corresponding to the target background investigation events.

[0007] The target quality inspection result includes at least one sub-quality inspection result of a quality inspection dimension and a target quality inspection evidence chain corresponding to each sub-quality inspection result. The target quality inspection evidence chain corresponding to each sub-quality inspection result is generated based on the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result. Each target quality inspection evidence corresponding to the sub-quality inspection result comes from the target background investigation materials.

[0008] The steps for determining the target quality inspection evidence corresponding to each of the sub-quality inspection results include: dividing the target background investigation materials to obtain several target material segments; merging at least two related target material segments as one of the initial quality inspection evidences; and using each unrelated target material segment as one of the initial quality inspection evidences; and finding at least one initial quality inspection evidence related to each of the initial quality inspection evidences as the target quality inspection evidence corresponding to each of the sub-quality inspection results.

[0009] The step of determining the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result includes: obtaining the target feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result, as the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result; wherein, the target fusion features are obtained by fusing the first target text semantic features of the target background investigation material, the second target text semantic features of the target background investigation report, the first target structured features of the first target event relationship diagram, and the second target structured features of the second target event relationship diagram.

[0010] The step of obtaining the target feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result includes: obtaining a first feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result; amplifying each first feature similarity to obtain each second feature similarity, which is used as the target feature similarity between the target evidence features and the target fusion features; or, using the ratio of each second feature similarity to the sum of similarities as the target feature similarity between the target evidence features and the target fusion features, wherein the sum of similarities is the sum of each second feature similarity.

[0011] The step of performing quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the target quality inspection result of the target background investigation report includes: extracting features from the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain a first target text semantic feature, a second target text semantic feature, a first target structured feature, and a second target structured feature; fusing the first target text semantic feature, the second target text semantic feature, the first target structured feature, and the second target structured feature to obtain a target fusion feature; and performing quality inspection based on the target fusion feature to obtain the target quality inspection result.

[0012] The quality inspection process, which involves performing quality checks based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the target quality inspection result of the target background investigation report, is executed using a quality inspection model. The training steps of the quality inspection model include: acquiring sample data; wherein the sample data includes sample background investigation materials and sample background investigation reports about the sample objects, the sample background investigation reports being generated based on the sample background investigation materials, and the sample background investigation reports annotating the actual quality inspection results; constructing a first sample event relationship diagram corresponding to the sample background investigation materials and a second sample event relationship diagram corresponding to the sample background investigation reports based on the sample background investigation materials and the sample background investigation reports, respectively; performing quality inspection using the quality inspection model based on the sample background investigation materials, the sample background investigation reports, the first sample event relationship diagram, and the second sample event relationship diagram to obtain the predicted quality inspection result of the sample background investigation report; and adjusting the model parameters of the quality inspection model based on the difference between the predicted quality inspection result and the actual quality inspection result.

[0013] The sample data includes a first type of sample data and a second type of sample data. The actual quality inspection result corresponding to the second type of sample data is determined based on business preferences. The model parameters of the quality inspection model are first adjusted based on the first type of sample data, and then the model parameters of the quality inspection model are adjusted based on the second type of sample data. And / or, before adjusting the model parameters of the quality inspection model using the difference between the predicted quality inspection result and the actual quality inspection result, the method further includes: freezing the current network parameters of the quality inspection model and adding a network layer to be trained to the target network layer of the quality inspection model. Adjusting the model parameters of the quality inspection model using the difference between the predicted quality inspection result and the actual quality inspection result includes: adjusting the network parameters of the target network layer using the difference between the predicted quality inspection result and the actual quality inspection result.

[0014] Wherein, the sample data includes first-class sample data, the predicted quality inspection result includes at least one sub-predicted quality inspection result of a quality inspection dimension, and the actual quality inspection result includes at least one sub-actual quality inspection result of a quality inspection dimension; adjusting the model parameters of the quality inspection model using the difference between the predicted quality inspection result and the actual quality inspection result includes: for each sub-predicted quality inspection result, determining a first loss based on the difference between the sub-predicted quality inspection result and each other sub-predicted quality inspection result; and determining a second loss based on the difference between each sub-predicted quality inspection result and the corresponding sub-actual quality inspection result; and adjusting the model parameters of the quality inspection model based on the first loss and the second loss.

[0015] The sample data includes a second type of sample data, the actual quality inspection results corresponding to the second type of sample data being determined based on business preferences. The predicted quality inspection results include at least one sub-predicted quality inspection results for a quality inspection dimension, and the actual quality inspection results include at least one sub-actual quality inspection result for a quality inspection dimension. Adjusting the model parameters of the quality inspection model using the difference between the predicted and actual quality inspection results includes: constructing a reward signal using the difference between the sub-predicted quality inspection results and the corresponding sub-actual quality inspection results; and adjusting the model parameters of the quality inspection model based on the reward signal.

[0016] A second aspect of this application provides a training method for a quality inspection model of a background investigation report. The method includes: acquiring sample data; wherein the sample data includes sample background investigation materials and sample background investigation reports concerning sample objects, the sample background investigation reports being generated based on the sample background investigation materials, and the sample background investigation reports indicating actual quality inspection results; constructing a first sample event relationship diagram corresponding to the sample background investigation materials and a second sample event relationship diagram corresponding to the sample background investigation reports based on the sample background investigation materials and the sample background investigation reports, respectively; performing quality inspection using the quality inspection model based on the sample background investigation materials, the sample background investigation reports, the first sample event relationship diagram, and the second sample event relationship diagram to obtain a predicted quality inspection result for the sample background investigation reports; and adjusting the model parameters of the quality inspection model based on the difference between the predicted quality inspection result and the actual quality inspection result.

[0017] A third aspect of this application provides an electronic device including a memory and a processor, wherein the memory is used to store program instructions and the processor is used to execute the program instructions to implement the above-described method.

[0018] A fourth aspect of this application provides a computer-readable storage medium for storing program instructions that can be executed to implement the above-described method.

[0019] The aforementioned technical solution, through target background investigation materials and reports, can capture the meaning, context, and potential information behind the text. Furthermore, through the first and second target event relationship diagrams, it can capture the logical relationships and hierarchical structure within the target background investigation materials and reports. Therefore, the target background investigation materials, reports, and the first and second target event relationship diagrams provide a more comprehensive quality inspection basis or a more comprehensive information perspective, reducing misjudgments caused by a single quality inspection basis or a single information perspective. Thus, quality inspection based on the target background investigation materials, reports, and the first and second target event relationship diagrams improves the accuracy of quality inspection of the target background investigation report.

[0020] In addition, the system automatically performs quality checks based on the target background investigation materials, target background investigation report, first target event relationship diagram, and second target event relationship diagram, thereby achieving automated quality checks on the background investigation report and improving the efficiency of the background investigation report quality checks. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating an embodiment of the quality inspection method for the background investigation report provided in this application; Figure 2 yes Figure 1 The flowchart of step S12 shown is a schematic diagram of one embodiment; Figure 3 yes Figure 1 The flowchart of step S13 shown is a schematic diagram of one embodiment. Figure 4 This is a flowchart illustrating an embodiment of the training method for the quality inspection model provided in this application; Figure 5 This is a schematic diagram of the framework structure of the quality inspection model provided in this application; Figure 6 This is a schematic diagram of an embodiment of the quality inspection device for the background investigation report provided in this application; Figure 7 This is a schematic diagram of the structure of an embodiment of the training device for the background investigation report quality inspection model provided in this application; Figure 8 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application; Figure 9 This is a schematic diagram of an embodiment of the computer-readable storage medium provided in this application. Detailed Implementation

[0022] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0023] In the following description, specific details such as particular system architectures, interfaces, and technologies are presented for illustrative purposes rather than for limiting purposes, in order to provide a thorough understanding of this application.

[0024] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, "many" in this document means two or more. Moreover, the term "at least one" in this document means any combination of at least two of any one or more of a plurality of objects. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0025] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the quality inspection method for the background investigation report provided in this application. It should be noted that if substantially the same result is obtained, this embodiment does not necessarily reflect that outcome. Figure 1 The illustrated process sequence is limited. For example... Figure 1 As shown, this embodiment includes: Step S11: Obtain background investigation materials and a background investigation report on the target object.

[0026] In this embodiment, background investigation materials and a background investigation report on the target object are obtained; wherein, the background investigation report is generated based on the background investigation materials.

[0027] In one embodiment, the background investigation materials and reports concerning the target object can be obtained from local storage or cloud storage. Of course, in other embodiments, the background investigation materials concerning the target object can be obtained in real time, and a background investigation report can be generated in real time based on the obtained materials; this is not limited to this method.

[0028] In one embodiment, background investigation materials for the target subject may include interview transcripts, manual verification records, etc., wherein the manual verification records are definitive records of information such as the target subject's educational background verified manually.

[0029] Step S12: Based on the target background investigation materials and the target background investigation report, construct the first target event relationship diagram corresponding to the target background investigation materials and the second target event relationship diagram corresponding to the target background investigation report, respectively.

[0030] In this embodiment, a first target event relationship diagram corresponding to the target background investigation materials and a second target event relationship diagram corresponding to the target background investigation report are constructed based on the target background investigation materials and the target background investigation report, respectively. The first target event relationship diagram corresponding to the target background investigation materials can reflect the logical relationships and hierarchical structure in the target background investigation materials, and the second target event relationship diagram corresponding to the target background investigation report can reflect the logical relationships and hierarchical structure in the target background investigation report.

[0031] Step S13: Conduct quality inspection based on the target background investigation materials, target background investigation report, first target event relationship diagram and second target event relationship diagram to obtain the target quality inspection results of the target background investigation report.

[0032] In this embodiment, quality inspection is performed based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the target quality inspection result of the target background investigation report. Through the target background investigation materials and the target background investigation report, the meaning, context, and potential information behind the text can be captured. Furthermore, through the first and second target event relationship diagrams, the logical relationships and hierarchical structure within the target background investigation materials and the target background investigation report can be captured. Therefore, the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram can provide a more comprehensive quality inspection basis or a more comprehensive information perspective, reducing misjudgments caused by a single quality inspection basis or a single information perspective. Therefore, quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram can improve the accuracy of the quality inspection of the target background investigation report.

[0033] In addition, the system automatically performs quality checks based on the target background investigation materials, target background investigation report, first target event relationship diagram, and second target event relationship diagram, thereby achieving automated quality checks on the background investigation report and improving the efficiency of the background investigation report quality checks.

[0034] In one embodiment, the target quality inspection result includes sub-quality inspection results of at least one quality inspection dimension, wherein the at least one quality inspection dimension includes at least one of the following: completeness, consistency, reasonableness of conclusion, and objectivity of expression. Since the quality inspection dimensions are not limited to specific dimensions and their number can be flexibly increased or decreased, they can be adjusted at any time according to the actual situation, offering high flexibility. Furthermore, by rationally selecting quality inspection dimensions based on the characteristics of the target object and the purpose of quality inspection, unnecessary inspection work can be avoided, improving quality inspection efficiency.

[0035] In cases where the target quality inspection results include sub-quality inspection results of at least two quality inspection dimensions, multi-dimensional quality inspection is conducted based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the multi-dimensional quality inspection results of the target background investigation report. These multi-dimensional quality inspection results provide a comprehensive evaluation of the target background investigation report from multiple perspectives, offering decision-makers a wealth of information.

[0036] It should be noted that completeness means the background investigation report should cover all relevant important information and aspects, without omitting any key content, providing users with comprehensive and systematic background information so they can have a holistic and in-depth understanding of the investigation target. Consistency means that the information within each part of the background investigation report is consistent, without contradictions or conflicts, ensuring the authenticity and reliability of the information. Reasonable conclusions mean that the conclusions drawn from the background investigation materials are logical, scientific, and objective, accurately reflecting the actual situation and characteristics of the target object, providing valuable reference for report users. Objectivity in expression means that the background investigation report should maintain a neutral and impartial attitude during the writing process, avoiding subjective bias and personal emotions, describing the situation and facts of the target object in objective and accurate language, without exaggeration, understatement, or distortion of information.

[0037] In one embodiment, the target quality inspection result includes sub-quality inspection results for at least one quality inspection dimension, wherein the sub-quality inspection result includes a quality inspection score. When the sub-quality inspection result is a quality inspection score, the quality inspection score is presented in a quantitative form on the performance of each quality inspection dimension, providing decision-makers with an intuitive and accurate reference.

[0038] In one embodiment, the target quality inspection result includes sub-quality inspection results for at least one quality inspection dimension and a target quality inspection evidence chain corresponding to each sub-quality inspection result. The target quality inspection evidence chain corresponding to each sub-quality inspection result is generated based on the contribution of each target quality inspection evidence to the sub-quality inspection result, and each target quality inspection evidence comes from the target background investigation materials. The target quality inspection evidence chain corresponding to each sub-quality inspection result provides detailed evidentiary support for the sub-quality inspection result, making the sub-quality inspection result an objectively based conclusion. In addition, because the target quality inspection evidence chain records the correspondence and contribution between the sub-quality inspection results and the target quality inspection evidence, the quality inspection process is traceable and reproducible.

[0039] In one specific implementation, the steps for determining the target quality inspection evidence corresponding to each sub-quality inspection result can be as follows: dividing the target background investigation materials into several target material segments; merging at least two related target material segments as one initial quality inspection evidence; and using each unrelated target material segment as a separate initial quality inspection evidence; and identifying at least one initial quality inspection evidence related to each sub-quality inspection result from the initial quality inspection evidence as the target quality inspection evidence corresponding to each sub-quality inspection result. By merging related target material segments, relevant information can be concentrated to form more targeted and coherent initial quality inspection evidence. This avoids omissions or misunderstandings caused by information dispersion when subsequently searching for target quality inspection evidence related to sub-quality inspection results, thereby ensuring a high degree of correlation between the determined target quality inspection evidence and the sub-quality inspection results and improving the accuracy of the evidence. In addition, each target quality inspection evidence originates from the initial quality inspection evidence that has been divided and merged. This initial quality inspection evidence clearly corresponds to a specific material segment in the target background investigation materials. This clear correspondence makes the source of the quality inspection results readily apparent, facilitating others' understanding and verification of the quality inspection process and conclusions.

[0040] One method is to calculate the semantic similarity of each target material segment to determine whether the target material segments are related.

[0041] Of course, in other specific implementations, each target material segment can be used as an initial quality inspection evidence, and then at least one initial quality inspection evidence related to each sub-quality inspection result can be found from each initial quality inspection evidence as the target quality inspection evidence corresponding to each sub-quality inspection result.

[0042] In one specific implementation, the step of determining the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result can be as follows: obtaining the target feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result, and using this as the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result. The target fusion features are obtained by fusing the first target text semantic features of the target background investigation materials, the second target text semantic features of the target background investigation report, the first target structured features of the first target event relationship diagram, and the second target structured features of the second target event relationship diagram.

[0043] The target fusion feature integrates the first target textual semantic features of the target background investigation materials, the second target textual semantic features of the target background investigation report, the first target structural features of the first target event relationship diagram, and the second target structural features of the second target event relationship diagram. This multi-dimensional feature fusion comprehensively and deeply characterizes the information in the target background investigation materials and report, covering multiple levels including textual semantics and structured relationships. When calculating the target feature similarity between target quality inspection evidence and the target fusion feature, it can consider multiple perspectives, avoiding the limitations of single features, and thus more accurately assessing the contribution of target quality inspection evidence to the sub-quality inspection results.

[0044] Furthermore, determining contribution by calculating the similarity between target evidence features and target fusion features allows for a more accurate measurement of the correlation between target quality inspection evidence and sub-quality inspection results. Higher target feature similarity indicates a greater semantic and structural match between the target quality inspection evidence and the sub-quality inspection results, resulting in stronger support for the sub-quality inspection results and a higher contribution. This precise matching method helps avoid the influence of subjective judgment and empiricism, making contribution assessment more objective and accurate.

[0045] In one specific implementation, obtaining the target feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result can specifically be as follows: obtaining the first feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result; amplifying each first feature similarity to obtain each second feature similarity; and using the ratio of each second feature similarity to the sum of similarities as the target feature similarity between the target evidence features and the target fusion features, wherein the sum of similarities is the sum of each second feature similarity.

[0046] By amplifying the similarity of each first feature to obtain the second feature similarity, the target quality inspection evidence that originally had a high similarity to the target fusion feature can be further strengthened. After amplification, the importance of this key evidence in quality inspection can be highlighted, allowing focus to be placed on important information. In addition, normalizing the amplified second feature similarity and calculating its ratio to the sum of similarities as the target feature similarity can further widen the gap in the contribution of different target quality inspection evidence to the sub-quality inspection results. Evidence that originally had similarity will have a more obvious distinction in contribution after amplification and normalization.

[0047] The specific formula is as follows:

[0048] In the formula, The contribution of the j-th target quality inspection evidence to the judgment of the k-th dimension sub-quality inspection result, i.e., the target feature similarity; Indicates target fusion features; The target evidence characteristic is represented by the j-th target quality inspection evidence.

[0049] Of course, in other specific implementations, the similarity of each second feature obtained by amplifying the similarity of each first feature can also be directly used as the target feature similarity between the target evidence feature and the target fusion feature, which is not limited here.

[0050] Please see Figure 2 , Figure 2 yes Figure 1 The flowchart shown is a schematic diagram of one embodiment of step S12. It should be noted that if substantially the same result is achieved, this embodiment does not necessarily follow the same pattern. Figure 2 The illustrated process sequence is limited. For example... Figure 2 As shown, this embodiment includes: Step S21: Identify at least one target background check event concerning the target object from the reference text.

[0051] In this embodiment, at least one background check event concerning the target object is determined from a reference text; wherein, the reference text is the target background investigation material or target background investigation report. It should be noted that background check events include, but are not limited to, educational events, work events, credit behavior events (such as credit behavior events, guarantee behavior events, etc.), etc., and are not limited here.

[0052] The obtained reference text may be in formats such as PDF, Word, Excel, or text. Therefore, in one embodiment, optical character recognition (OCR) is performed before determining at least one target background check event about the target object from the reference text.

[0053] Since the reference text may contain noise that is detrimental to subsequent processing, such as format-related noise (e.g., extra spaces, line breaks, tabs, etc.) and non-text content (images, QR codes, etc.), in one embodiment, the reference text is denoised before determining at least one target background check event about the target object from the reference text.

[0054] Step S22: For each target background check event, identify the target event entity related to the target background check event from the reference text.

[0055] In this embodiment, for each target background check event, the target event entities related to the target background check event are identified from the reference text. By determining the target background check events from the reference text and further identifying the target event entities related to each target background check event, the scattered information dispersed in the reference text can be comprehensively and systematically sorted out.

[0056] In one embodiment, for each target background check event, Named Entity Recognition (NER) technology is used to identify the target event entity related to the target background check event from the reference text.

[0057] In the reference text, the same target event entity related to the target background check event may be expressed in multiple different ways. Therefore, in one embodiment, after identifying the target event entity related to the target background check event from the reference text, entity standardization processing is performed on the target event entity related to the target background check event to unify these different expressions into a standard form, or to represent them according to a unified standard and specification, thereby eliminating the differences caused by different expressions and improving data consistency.

[0058] In the reference text, the formats and / or units of different time entities related to the target background check event may be inconsistent. Therefore, in one embodiment, after identifying the target event entity related to the target background check event from the reference text, the format and / or units of the time entities related to the target background check event are normalized to be represented in a uniform format and unit.

[0059] In the reference text, the units of different location entities related to the target background investigation event may be inconsistent. Therefore, in one embodiment, after identifying the target event entity related to the target background investigation event from the reference text, the units of the location entities related to the target background investigation event are normalized to be represented according to a unified unit, such as uniformly using administrative division codes or geographic codes to represent locations.

[0060] Step S23: Construct a reference sub-relationship graph corresponding to the target background investigation event by utilizing the relationships between the target event entities.

[0061] In this embodiment, a reference sub-relationship graph corresponding to the target background check event is constructed by utilizing the relationships between the target event entities. For example, taking the target event as a work event, the relationship between the target object, the company where the target works, and the time of employment, as well as the corresponding relationships, can be modeled as a reference sub-relationship graph corresponding to the work event.

[0062] In one implementation, dependency parsing is used to analyze the relationships between various target event entities.

[0063] Step S24: Use the reference sub-relationship diagrams corresponding to each target background investigation event to form a first target event relationship diagram or a second target event relationship diagram.

[0064] In this embodiment, a first target event relationship diagram or a second target event relationship diagram is formed using reference sub-relationship diagrams corresponding to each target background investigation event. Reference sub-relationship diagrams are constructed using the relationships between each target event entity, and then these reference sub-relationship diagrams are combined to form a complete target event relationship diagram. This graphically presents the complex connections between target-related events and entities, making it easier to understand and analyze than pure text descriptions. Furthermore, the target event relationship diagram clearly shows the associations between different target background investigation events and target event entities, helping to uncover deeper relationships and potential patterns hidden behind surface information.

[0065] Please see Figure 3 , Figure 3 yes Figure 1 The diagram shows a flowchart of one embodiment of step S13. It should be noted that if substantially the same result is achieved, this embodiment does not necessarily follow that approach. Figure 3 The illustrated process sequence is limited. For example... Figure 3 As shown, this embodiment includes: Step S31: Extract features from the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram respectively, and obtain the first target text semantic features, the second target text semantic features, the first target structured features, and the second target structured features.

[0066] In this embodiment, feature extraction is performed on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram, respectively, to obtain the first target text semantic features, the second target text semantic features, the first target structured features, and the second target structured features.

[0067] Step S32: Fuse the semantic features of the first target text, the semantic features of the second target text, the structured features of the first target, and the structured features of the second target to obtain the target fused features.

[0068] In this embodiment, the semantic features of the first target text, the semantic features of the second target text, the structural features of the first target, and the structural features of the second target are fused to obtain the target fusion feature. The semantic features of the target text can capture the meaning, context, and potential information behind the text, while the structural features of the target can reflect the logical relationships and hierarchical structure in the target background investigation report and materials. Therefore, the target fusion feature obtained by fusing the semantic features of the target text and the structural features of the target can provide a more comprehensive quality inspection basis or a more comprehensive information perspective.

[0069] In one embodiment, the fusion method for fusing the semantic features of the first target text, the semantic features of the second target text, the structured features of the first target, and the structured features of the second target can be feature concatenation, feature weighted summation, multilayer perceptron fusion, graph neural network fusion, etc., and is not limited here.

[0070] Step S33: Perform quality inspection based on target fusion features to obtain target quality inspection results.

[0071] In this embodiment, quality inspection is performed based on target fusion features to obtain target quality inspection results. Target text semantic features can capture the meaning, context, and potential information behind the text, while target structured features can reflect the logical relationships and hierarchical structure in the target background investigation report and materials. Target text semantic features and target structured features complement each other. Therefore, quality inspection based on the target fusion features obtained by fusing target text semantic features and target structured features provides a more comprehensive basis for quality inspection, reduces misjudgments caused by single features, and improves the accuracy of quality inspection.

[0072] Please see Figure 4 and Figure 5 , Figure 4 This is a flowchart illustrating an embodiment of the training method for the quality inspection model provided in this application. Figure 5 This is a schematic diagram of the framework structure of the quality inspection model provided in this application. It should be noted that if substantially the same result is obtained, this embodiment does not necessarily reflect that outcome. Figure 4 The illustrated process sequence is limited. For example... Figure 4 As shown, quality inspection is performed based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram. The target quality inspection results obtained from the target background investigation report are executed using a quality inspection model. This embodiment includes: Step S41: Obtain sample data.

[0073] In this embodiment, sample data is obtained; wherein, the sample data includes sample background investigation materials and sample background investigation reports about the sample objects, the sample background investigation reports are generated based on the sample background investigation materials, and the sample background investigation reports are marked with the actual quality inspection results.

[0074] In one embodiment, the background investigation materials and reports on the sample subjects can be obtained from local storage or cloud storage. Of course, in other embodiments, the background investigation materials on the sample subjects can be obtained in real time, and the background investigation report can be generated in real time based on the obtained background investigation materials; this is not limited here.

[0075] In one embodiment, the background investigation materials for the sample subjects may include interview transcripts, manual verification records, etc., wherein the manual verification records are definitive records of information such as the educational background of the sample subjects verified manually.

[0076] In one implementation, the quality control dimensions for the sample background investigation report include completeness, consistency, reasonableness of conclusions, and objectivity of expression. The specific sample data is as follows:

[0077] In the formula, Represents the sample dataset; This indicates a background investigation report for the sample. This indicates the background survey materials for the sample; This represents a multidimensional quality label; N represents the total number of samples.

[0078] Step S42: Based on the sample background survey materials and the sample background survey report, construct the first sample event relationship diagram corresponding to the sample background survey materials and the second sample event relationship diagram corresponding to the sample background survey report, respectively.

[0079] In this embodiment, a first sample event relationship diagram corresponding to the sample background survey materials and a second sample event relationship diagram corresponding to the sample background survey report are constructed based on the sample background survey materials and the sample background survey report, respectively. The first sample event relationship diagram corresponding to the sample background survey materials can reflect the logical relationships and hierarchical structure in the sample background survey materials, and the second sample event relationship diagram corresponding to the sample background survey report can reflect the logical relationships and hierarchical structure in the sample background survey report.

[0080] Step S43: Use the quality inspection model to perform quality inspection based on the sample background investigation materials, sample background investigation report, first sample event relationship diagram and second sample event relationship diagram to obtain the predicted quality inspection results of the sample background investigation report.

[0081] In this embodiment, a quality inspection model is used to perform quality inspection based on sample background investigation materials, sample background investigation reports, a first sample event relationship diagram, and a second sample event relationship diagram to obtain the predicted quality inspection results of the sample background investigation report. Through the sample background investigation materials and reports, the meaning, context, and potential information behind the text can be captured. Furthermore, through the first and second sample event relationship diagrams, the logical relationships and hierarchical structure within the sample background investigation materials and reports can be captured. Therefore, the sample background investigation materials, sample background investigation reports, first sample event relationship diagrams, and second sample event relationship diagrams provide a more comprehensive quality inspection basis or a more comprehensive information perspective, reducing misjudgments caused by a single quality inspection basis or a single information perspective. Therefore, quality inspection based on sample background investigation materials, sample background investigation reports, first sample event relationship diagrams, and second sample event relationship diagrams can improve the accuracy of the background investigation report's quality inspection.

[0082] In addition, the system automatically performs quality checks based on the sample background investigation materials, sample background investigation report, first sample event relationship diagram, and second sample event relationship diagram, thus achieving automated quality checks on the background investigation report and improving the efficiency of the quality checks.

[0083] In one implementation, the quality inspection model can be a pre-trained large language model. The pre-trained large language model serves as the foundational semantic representation and reasoning framework for the quality inspection model, enabling it to model complex long texts, multi-dimensional semantic relationships, and implicit logic.

[0084] It should be noted that the pre-trained large language model, used as a quality control model, has its initial parameters derived from training on a general corpus.

[0085] In one specific implementation, the pre-trained large language model can be Qwen3-8B.

[0086] In one embodiment, a quality inspection model is used to perform quality inspection based on sample background investigation materials, sample background investigation reports, a first sample event relationship diagram, and a second sample event relationship diagram to obtain a predicted quality inspection result for the sample background investigation report. Specifically, this can be achieved by: extracting features from the sample background investigation materials, sample background investigation reports, the first sample event relationship diagram, and the second sample event relationship diagram, respectively, to obtain first sample text semantic features, second sample text semantic features, first sample structured features, and second sample structured features; fusing these first sample text semantic features, second sample text semantic features, first sample structured features, and second sample structured features to obtain sample fusion features; and performing quality inspection based on these sample fusion features to obtain a predicted quality inspection result. Sample text semantic features can capture the meaning, context, and potential information behind the text, while sample structured features can reflect the logical relationships and hierarchical structure in the sample background investigation materials and sample background investigation reports. Sample text semantic features and sample structured features complement each other. Therefore, the sample fusion features obtained by fusing sample text semantic features and sample structured features provide a more comprehensive basis for quality inspection, reducing misjudgments caused by single features and improving the accuracy of quality inspection.

[0087] Specifically, the semantic features of the first sample text, the semantic features of the second sample text, the structured features of the first sample, and the structured features of the second sample are fused to obtain the sample fusion features, as shown in the following formula:

[0088] In the formula, Indicates sample fusion features; Indicates a feedforward network; This represents the feature obtained by fusing the semantic features of the first sample text and the semantic features of the second sample text. This represents the feature obtained by fusing the structured features of the first sample and the structured features of the second sample.

[0089] In one specific implementation, the predicted quality inspection result includes sub-predicted quality inspection results for at least one quality inspection dimension, wherein the sub-predicted quality inspection results are specifically as follows:

[0090] In the formula, This represents the predicted quality inspection result of the k-th dimension sub-sub ... Indicates the activation function; and This indicates a dimension-specific parameter.

[0091] Step S44: Adjust the model parameters of the quality inspection model by utilizing the difference between the predicted quality inspection results and the actual quality inspection results.

[0092] In this embodiment, the model parameters of the quality inspection model are adjusted by utilizing the difference between the predicted and actual quality inspection results. By adjusting the model parameters based on this difference, the predicted quality inspection results obtained by the quality inspection model based on sample background investigation materials, sample background investigation reports, first sample event relationship diagrams, and second sample event relationship diagrams can approximate the corresponding actual quality inspection results. In other words, it drives the quality inspection model to obtain the most accurate predicted quality inspection results based on sample background investigation materials, sample background investigation reports, first sample event relationship diagrams, and second sample event relationship diagrams, thereby improving the accuracy of the quality inspection model in inspecting background investigation reports.

[0093] In one embodiment, the quality inspection model can be a pre-trained large language model. Before adjusting the model parameters of the quality inspection model using the difference between the predicted and actual quality inspection results, the current network parameters of the quality inspection model are frozen, and a training layer is added to the target network layer of the quality inspection model. At this time, the model parameters of the quality inspection model are adjusted using the difference between the predicted and actual quality inspection results. Specifically, this can be done by adjusting the network parameters of the target network layer using the difference between the predicted and actual quality inspection results. Adding a training layer to each target network layer of the pre-trained large language model means adding trainable incremental parameters to each target network layer of the pre-trained large language model. In other words, without directly modifying the original parameters of the pre-trained large language model, additional trainable incremental parameters are introduced to make targeted adjustments to the trained large language model. On the one hand, this allows the adjusted large language model to retain its general language understanding capabilities; on the other hand, it enables the adjusted large language model to learn domain knowledge in the background investigation field, improving its adaptability to background investigation report quality inspection tasks and optimizing its background investigation report quality inspection performance—that is, improving the background investigation report quality inspection capability of the adjusted large language model. Furthermore, the network parameters of the untrained network layers are adjusted, i.e., the parameters of the added trainable incremental parameters are tuned, avoiding the high computational cost of full parameter tuning.

[0094] Specifically, after adding the target network layer to the target network layer, the network parameters of the target network layer are updated as follows:

[0095] In the formula, This represents the updated network parameters of the target network layer; This represents the current network parameters of the target network layer; Indicates the network parameters of the network layer to be trained; AB represents The low-rank adaptive weight matrix is, in other words, the two rank decomposition matrices corresponding to the network layer to be trained.

[0096] In one specific implementation, a network layer to be trained may be added to the attention layer and / or feedforward network layer of the quality inspection model.

[0097] In one embodiment, the sample data includes a first type of sample data, the predicted quality inspection results include sub-predicted quality inspection results of at least one quality inspection dimension, and the actual quality inspection results include sub-actual quality inspection results of at least one quality inspection dimension. The model parameters of the quality inspection model are adjusted using the difference between the predicted and actual quality inspection results. Specifically, for each sub-predicted quality inspection result, a first loss is determined based on the difference between the sub-predicted quality inspection result and each other sub-predicted quality inspection result; and a second loss is determined based on the difference between each sub-predicted quality inspection result and its corresponding sub-actual quality inspection result. The model parameters of the quality inspection model are then adjusted based on the first and second losses. The quality inspection task of the sample background investigation report is abstracted into a multi-dimensional quality judgment problem, and each quality inspection dimension is jointly modeled during the adjustment of the model parameters of the quality inspection model to simultaneously optimize the sub-quality inspection results of each quality inspection dimension in a single training iteration. Furthermore, by combining the first and second losses, the dependencies between different quality inspection dimensions are captured, preventing the quality inspection model from overfitting to a single quality inspection dimension. Furthermore, by unifying multiple quality inspection dimensions into the judgment target space of the same quality inspection model, the quality inspection model can simultaneously evaluate the performance of the background investigation report on multiple quality inspection dimensions during a single reasoning process.

[0098] In one specific implementation, when the sample data includes the first type of sample data, the loss is calculated using the following formula:

[0099] In the formula, CE represents the dimensional weights; The covariance constraint weights are represented by Cov; Cov represents the covariance operation. This represents the predicted quality inspection result of the k-th dimension sub-sub ... This represents the true quality inspection result of the k-th dimension. This indicates the predicted quality inspection results for other components.

[0100] In one embodiment, the sample data includes a second type of sample data. The actual quality inspection results corresponding to the second type of sample data are determined based on business preferences. The predicted quality inspection results include at least one sub-predicted quality inspection results for at least one quality inspection dimension, and the actual quality inspection results include at least one sub-actual quality inspection results for at least one quality inspection dimension. The model parameters of the quality inspection model are adjusted using the difference between the predicted and actual quality inspection results. Specifically, this can be achieved by constructing a reward signal based on the difference between the sub-predicted quality inspection results and their corresponding sub-actual quality inspection results; and adjusting the model parameters of the quality inspection model based on the reward signal. By introducing a reinforcement learning alignment mechanism based on business preference data, the judgment strategy of the quality inspection model is further optimized. During the reinforcement learning adjustment process, the sub-quality inspection results for different quality inspection dimensions given in the same background investigation report are considered as candidate actions. A reward signal is constructed using business preference data to guide the quality inspection model to gradually increase the output probability of quality inspection results that meet the business preference standards during the adjustment process, and reduce the output probability of judgment results with risks of omission, misjudgment, or over-inference. That is, through this reinforcement learning alignment process, the quality inspection results output by the quality inspection model are made consistent with the actual business preference standards.

[0101] In one specific implementation, the reward signal is constructed by utilizing the difference between the sub-predicted quality inspection results and the corresponding sub-true quality inspection results, as shown below:

[0102] In the formula, This indicates a reward signal; This represents the predicted quality inspection result of the k-th dimension sub-sub ... The true quality inspection result of the k-th dimension is determined based on business preferences; ; .

[0103] Furthermore, using the reward signal, the network parameters of the target network layer (which incorporates the network layer to be trained) of the quality inspection model are adjusted, as shown in the following formula:

[0104] In one implementation, the sample data includes a first type of sample data and a second type of sample data. The actual quality inspection results corresponding to the second type of sample data are determined based on business preferences. The model parameters of the quality inspection model are first adjusted based on the first type of sample data, and then adjusted again based on the second type of sample data. Adjusting the model parameters based on the first type of sample data first aims to build a quality inspection model with the capability to inspect background investigation reports, laying the foundation for the model to perform preliminary assessments of the quality of background investigation reports. Adjusting the model parameters based on the second type of sample data secondly, after the quality inspection model has acquired basic background investigation report quality inspection capabilities, further optimizes these capabilities through reinforcement learning. In other words, it is an enhancement and improvement upon the basic capabilities, using business preference data to construct reward signals to guide the quality inspection model in adjusting its output probabilities, making its output results more aligned with business preferences.

[0105] Please see Figure 6 , Figure 6 This is a schematic diagram of an embodiment of the background investigation report quality inspection device provided in this application. The background investigation report quality inspection device 60 includes an acquisition module 61, a construction module 62, and a quality inspection module 63; the acquisition module 61 is used to acquire target background investigation materials and a target background investigation report about the target object; wherein, the target background investigation report is generated based on the target background investigation materials; the construction module 62 is used to construct a first target event relationship diagram corresponding to the target background investigation materials and a second target event relationship diagram corresponding to the target background investigation report based on the target background investigation materials and the target background investigation report, respectively; the quality inspection module 63 is used to perform quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the target quality inspection result of the target background investigation report.

[0106] The aforementioned target quality inspection results include sub-quality inspection results of at least one quality inspection dimension, wherein at least one quality inspection dimension includes at least one of the following: completeness, consistency, reasonableness of conclusions, and objectivity of expression; and / or, the aforementioned sub-quality inspection results include quality inspection scores.

[0107] The construction module 62 is used to construct a first target event relationship diagram corresponding to the target background investigation material and a second target event relationship diagram corresponding to the target background investigation report, respectively, based on the target background investigation material and the target background investigation report. This includes: identifying at least one target background investigation event concerning the target object from the reference text; wherein the reference text is the target background investigation material or the target background investigation report; for each target background investigation event, identifying target event entities related to the target background investigation event from the reference text; constructing a reference sub-relationship diagram corresponding to the target background investigation event using the relationships between the target event entities; and forming the first target event relationship diagram or the second target event relationship diagram using the reference sub-relationship diagrams corresponding to each target background investigation event.

[0108] The aforementioned target quality inspection results include sub-quality inspection results of at least one quality inspection dimension and target quality inspection evidence chains corresponding to each sub-quality inspection result. The target quality inspection evidence chains corresponding to each sub-quality inspection result are generated based on the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result. Each target quality inspection evidence corresponding to the sub-quality inspection result comes from the target background investigation materials.

[0109] The quality inspection module 63 is used for the determination of each target quality inspection evidence corresponding to each sub-quality inspection result, including: dividing the target background investigation materials to obtain several target material segments; merging at least two related target material segments as an initial quality inspection evidence; and using each unrelated target material segment as an initial quality inspection evidence; and finding at least one initial quality inspection evidence related to each sub-quality inspection result from each initial quality inspection evidence as the target quality inspection evidence corresponding to each sub-quality inspection result.

[0110] The quality inspection module 63 is used to determine the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result. This includes: obtaining the target feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result, as the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result; wherein, the target fusion features are obtained by fusing the first target text semantic features of the target background investigation materials, the second target text semantic features of the target background investigation report, the first target structured features of the first target event relationship diagram, and the second target structured features of the second target event relationship diagram.

[0111] The quality inspection module 63 is used to obtain the target feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result, including: obtaining the first feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result; amplifying each first feature similarity to obtain each second feature similarity, which is used as the target feature similarity between the target evidence features and the target fusion features, or, using the ratio of each second feature similarity to the sum of similarities as the target feature similarity between the target evidence features and the target fusion features, wherein the sum of similarities is the sum of each second feature similarity.

[0112] The quality inspection module 63 is used to perform quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the target quality inspection result of the target background investigation report. This includes: extracting features from the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the first target text semantic features, the second target text semantic features, the first target structured features, and the second target structured features; fusing the first target text semantic features, the second target text semantic features, the first target structured features, and the second target structured features to obtain the target fusion features; and performing quality inspection based on the target fusion features to obtain the target quality inspection result.

[0113] The background investigation report quality inspection device 60 further includes a training module 64. The aforementioned quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram, to obtain the target quality inspection result of the target background investigation report, is performed using a quality inspection model. The training module 64 is used for the training steps of the quality inspection model, including: acquiring sample data; wherein, the sample data includes sample background investigation materials and sample background investigation reports about the sample objects, the sample background investigation reports are generated based on the sample background investigation materials, and the sample background investigation reports are marked with the actual quality inspection results; constructing the first sample event relationship diagram corresponding to the sample background investigation materials and the second sample event relationship diagram corresponding to the sample background investigation reports based on the sample background investigation materials and the sample background investigation reports, respectively; using the quality inspection model to perform quality inspection based on the sample background investigation materials, the sample background investigation reports, the first sample event relationship diagram, and the second sample event relationship diagram to obtain the predicted quality inspection result of the sample background investigation report; and adjusting the model parameters of the quality inspection model based on the difference between the predicted quality inspection result and the actual quality inspection result.

[0114] The aforementioned sample data includes a first type of sample data and a second type of sample data. The actual quality inspection results corresponding to the second type of sample data are determined based on business preferences. The model parameters of the quality inspection model are first adjusted based on the first type of sample data, and then the model parameters of the quality inspection model are adjusted based on the second type of sample data. And / or, the training module 64 is used to, before adjusting the model parameters of the quality inspection model by utilizing the difference between the predicted quality inspection results and the actual quality inspection results, include: freezing the current network parameters of the quality inspection model and adding a network layer to be trained to the target network layer of the quality inspection model; adjusting the model parameters of the quality inspection model by utilizing the difference between the predicted quality inspection results and the actual quality inspection results includes: adjusting the network parameters of the target network layer by utilizing the difference between the predicted quality inspection results and the actual quality inspection results.

[0115] The aforementioned sample data includes first-class sample data, the predicted quality inspection results include at least one sub-predicted quality inspection results of a quality inspection dimension, and the actual quality inspection results include at least one sub-actual quality inspection result of a quality inspection dimension. The training module 64 is used to adjust the model parameters of the quality inspection model by utilizing the difference between the predicted quality inspection results and the actual quality inspection results, including: for each sub-predicted quality inspection result, determining a first loss based on the difference between the sub-predicted quality inspection result and each other sub-predicted quality inspection result; and determining a second loss based on the difference between each sub-predicted quality inspection result and the corresponding sub-actual quality inspection result; and adjusting the model parameters of the quality inspection model based on the first loss and the second loss.

[0116] The sample data mentioned above includes second-type sample data, and the actual quality inspection results corresponding to the second-type sample data are determined based on business preferences. The predicted quality inspection results include at least one sub-predicted quality inspection results of the quality inspection dimension, and the actual quality inspection results include at least one sub-actual quality inspection results of the quality inspection dimension. The training module 64 is used to adjust the model parameters of the quality inspection model by utilizing the difference between the predicted quality inspection results and the actual quality inspection results, including: constructing a reward signal by utilizing the difference between the sub-predicted quality inspection results and the corresponding sub-actual quality inspection results; and adjusting the model parameters of the quality inspection model based on the reward signal.

[0117] Please see Figure 7 , Figure 7This is a schematic diagram of an embodiment of the training device for the background investigation report quality inspection model provided in this application. The training device 70 for the background investigation report quality inspection model includes an acquisition module 71, a construction module 72, a quality inspection module 73, and an adjustment module 74. The acquisition module 71 is used to acquire sample data. The sample data includes sample background investigation materials and sample background investigation reports about the sample objects. The sample background investigation reports are generated based on the sample background investigation materials and contain the actual quality inspection results. The construction module 72 is used to construct a first sample event relationship diagram corresponding to the sample background investigation materials and a second sample event relationship diagram corresponding to the sample background investigation reports, respectively, based on the sample background investigation materials and the sample background investigation reports. The quality inspection module 73 is used to perform quality inspection using the quality inspection model based on the sample background investigation materials, the sample background investigation reports, the first sample event relationship diagram, and the second sample event relationship diagram to obtain the predicted quality inspection results of the sample background investigation reports. The adjustment module 74 is used to adjust the model parameters of the quality inspection model based on the difference between the predicted quality inspection results and the actual quality inspection results.

[0118] Please see Figure 8 , Figure 8 This is a schematic diagram of an embodiment of the electronic device provided in this application. The electronic device 80 includes a memory 81 and a processor 82 coupled to each other. The processor 82 is used to execute program instructions stored in the memory 81 to implement the steps of any of the above-described background investigation report quality inspection methods and / or background investigation report quality inspection model training method embodiments. In a specific implementation scenario, the electronic device 80 may include, but is not limited to, a microcomputer, a server, etc. In addition, the electronic device 80 may also include mobile devices such as laptops and tablets, which are not limited here.

[0119] Specifically, processor 82 controls itself and memory 81 to implement the steps of any of the above-described background investigation report quality inspection methods and / or background investigation report quality inspection model training method embodiments. Processor 82 may also be referred to as a CPU (Central Processing Unit). Processor 82 may be an integrated circuit chip with signal processing capabilities. Processor 82 may also be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor. Furthermore, processor 82 may be implemented using integrated circuit chips.

[0120] Please see Figure 9 , Figure 9 This is a schematic diagram of an embodiment of the computer-readable storage medium provided in this application. The computer-readable storage medium 90 of this application embodiment stores program instructions 91. When executed, these program instructions 91 implement the methods provided by any embodiment and any non-conflicting combination of the quality inspection method and / or training method for the quality inspection model of the background investigation report of this application. The program instructions 91 can form a program file and be stored in the aforementioned computer-readable storage medium 90 in the form of a software product, so that a computer device (which may be a personal computer, server, or network device, etc.) can execute all or part of the steps of the methods of various embodiments of this application. The aforementioned computer-readable storage medium 90 includes various media capable of storing program code, such as a USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or terminal devices such as computers, servers, mobile phones, and tablets.

[0121] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

[0122] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A quality control method for background investigation reports, characterized in that, The method includes: Obtain background investigation materials and a background investigation report on the target object; wherein the background investigation report is generated based on the background investigation materials. Based on the target background investigation materials and the target background investigation report, respectively, construct a first target event relationship diagram corresponding to the target background investigation materials and a second target event relationship diagram corresponding to the target background investigation report; Quality inspection is performed based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the target quality inspection result of the target background investigation report.

2. The method according to claim 1, characterized in that, The target quality inspection result includes sub-quality inspection results for at least one quality inspection dimension, wherein, The at least one quality inspection dimension includes at least one of the following: completeness, consistency, reasonableness of conclusions, and objectivity of expression; And / or, the sub-quality inspection results include quality inspection scores.

3. The method according to claim 1, characterized in that, The step of constructing a first target event relationship diagram corresponding to the target background investigation materials and a second target event relationship diagram corresponding to the target background investigation report, respectively, based on the target background investigation materials and the target background investigation report, includes: From the reference text, at least one background check event concerning the target object is identified; wherein the reference text is the target background investigation material or the target background investigation report; For each of the target background investigation events, the target event entities related to the target background investigation event are identified from the reference text; By utilizing the relationships between the target event entities, a reference sub-relationship graph corresponding to the target background investigation event is constructed; The first target event relationship diagram or the second target event relationship diagram is formed by using the reference sub-relationship diagrams corresponding to each of the target background investigation events.

4. The method according to claim 1, characterized in that, The target quality inspection result includes at least one sub-quality inspection result of a quality inspection dimension and a target quality inspection evidence chain corresponding to each sub-quality inspection result. The target quality inspection evidence chain corresponding to each sub-quality inspection result is generated based on the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result. Each target quality inspection evidence corresponding to the sub-quality inspection result comes from the target background investigation materials.

5. The method according to claim 4, characterized in that, The steps for determining each target quality inspection evidence corresponding to each of the sub-quality inspection results include: The background investigation materials for the target are divided into several target material segments; Merge at least two relevant target material segments as one of the initial quality inspection evidences; Furthermore, each unrelated target material segment is used as one of the aforementioned initial quality inspection evidences; From each of the initial quality inspection evidences, at least one initial quality inspection evidence related to each of the sub-quality inspection results is identified and used as the target quality inspection evidence corresponding to each of the sub-quality inspection results.

6. The method according to claim 4, characterized in that, The steps for determining the contribution of each target quality inspection evidence corresponding to the sub-quality inspection result to the sub-quality inspection result include: The similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection result is obtained as the contribution of each target quality inspection evidence to the sub-quality inspection result; wherein, the target fusion features are obtained by fusing the first target text semantic features of the target background investigation material, the second target text semantic features of the target background investigation report, the first target structured features of the first target event relationship diagram, and the second target structured features of the second target event relationship diagram.

7. The method according to claim 6, characterized in that, The step of obtaining the target feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection results includes: Obtain the first feature similarity between the target evidence features and the target fusion features of each target quality inspection evidence corresponding to the sub-quality inspection results; The similarity of each of the first features is amplified to obtain the similarity of each of the second features, which is used as the target feature similarity between the target evidence feature and the target fusion feature. Alternatively, the ratio of the similarity of each of the second features to the sum of the similarities is used as the target feature similarity between the target evidence feature and the target fusion feature, wherein the sum of the similarities is the sum of the similarities of each of the second features.

8. The method according to claim 1, characterized in that, The quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram, to obtain the target quality inspection result of the target background investigation report, includes: Feature extraction is performed on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram to obtain the first target text semantic features, the second target text semantic features, the first target structured features, and the second target structured features. The semantic features of the first target text, the semantic features of the second target text, the structural features of the first target, and the structural features of the second target are fused to obtain the target fusion features; Quality inspection is performed based on the target fusion features to obtain the target quality inspection result.

9. The method according to claim 1, characterized in that, The quality inspection based on the target background investigation materials, the target background investigation report, the first target event relationship diagram, and the second target event relationship diagram, and obtaining the target quality inspection result of the target background investigation report, is performed using a quality inspection model. The training steps of the quality inspection model include: Obtain sample data; wherein, the sample data includes sample background investigation materials and sample background investigation reports about the sample objects, the sample background investigation reports are generated based on the sample background investigation materials, and the sample background investigation reports are marked with the actual quality inspection results; Based on the sample background survey materials and the sample background survey report, respectively, construct a first sample event relationship diagram corresponding to the sample background survey materials and a second sample event relationship diagram corresponding to the sample background survey report; The quality inspection model is used to perform quality inspection based on the sample background investigation materials, the sample background investigation report, the first sample event relationship diagram, and the second sample event relationship diagram to obtain the predicted quality inspection results of the sample background investigation report; The model parameters of the quality inspection model are adjusted based on the difference between the predicted quality inspection results and the actual quality inspection results.

10. The method according to claim 9, characterized in that, The sample data includes a first type of sample data and a second type of sample data. The actual quality inspection results corresponding to the second type of sample data are determined based on business preferences. The model parameters of the quality inspection model are first adjusted based on the first type of sample data, and then the model parameters of the quality inspection model are adjusted based on the second type of sample data. And / or, before adjusting the model parameters of the quality inspection model using the difference between the predicted quality inspection results and the actual quality inspection results, the method further includes: Freeze the current network parameters of the quality inspection model, and add the network layer to be trained to the target network layer of the quality inspection model; The step of adjusting the model parameters of the quality inspection model based on the difference between the predicted quality inspection results and the actual quality inspection results includes: The network parameters of the target network layer are adjusted based on the difference between the predicted quality inspection results and the actual quality inspection results.

11. The apparatus according to claim 9, characterized in that, The sample data includes first-class sample data, the predicted quality inspection results include at least one sub-predicted quality inspection results for at least one quality inspection dimension, and the actual quality inspection results include at least one sub-actual quality inspection results for at least one quality inspection dimension; adjusting the model parameters of the quality inspection model using the difference between the predicted quality inspection results and the actual quality inspection results includes: For each of the sub-prediction quality inspection results, a first loss is determined based on the difference between the sub-prediction quality inspection result and each of the other sub-prediction quality inspection results; Furthermore, a second loss is determined based on the difference between each of the sub-predicted quality inspection results and the corresponding sub-real quality inspection results; Based on the first loss and the second loss, the model parameters of the quality inspection model are adjusted.

12. The method according to claim 9, characterized in that, The sample data includes a second type of sample data, and the actual quality inspection results corresponding to the second type of sample data are determined based on business preferences. The predicted quality inspection results include at least one sub-predicted quality inspection results of a quality inspection dimension, and the actual quality inspection results include at least one sub-actual quality inspection results of a quality inspection dimension. The step of adjusting the model parameters of the quality inspection model based on the difference between the predicted quality inspection results and the actual quality inspection results includes: A reward signal is constructed by utilizing the difference between the predicted sub-quality inspection results and the corresponding actual sub-quality inspection results; Based on the reward signal, the model parameters of the quality inspection model are adjusted.

13. A training method for a background investigation report quality inspection model, characterized in that, The method includes: Obtain sample data; wherein, the sample data includes sample background investigation materials and sample background investigation reports about the sample objects, the sample background investigation reports are generated based on the sample background investigation materials, and the sample background investigation reports are marked with the actual quality inspection results; Based on the sample background survey materials and the sample background survey report, respectively, construct a first sample event relationship diagram corresponding to the sample background survey materials and a second sample event relationship diagram corresponding to the sample background survey report; The quality inspection model is used to perform quality inspection based on the sample background investigation materials, the sample background investigation report, the first sample event relationship diagram, and the second sample event relationship diagram to obtain the predicted quality inspection results of the sample background investigation report; The model parameters of the quality inspection model are adjusted based on the difference between the predicted quality inspection results and the actual quality inspection results.

14. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory being used to store program instructions, and the processor being used to execute the program instructions to implement the method as described in any one of claims 1-13.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program instructions that can be executed to implement the method as claimed in any one of claims 1-13.

Images