Content moderation methods, electronic devices and software products

Abstract

This disclosure provides a content moderation method, an electronic device, and a computer program product. The content moderation method includes: obtaining screenshots of communication related to the content to be reviewed; extracting text content, background color, text location information, and identity information of the communication participants from the screenshots; determining the temporal order of the text communication information based on the location information, and determining the sender information of the text communication information based on the background color and the identity information of the communication participants, and then generating structured information; performing semantic parsing on the structured information to obtain response intent information; if the response intent information indicates that the content moderator has approved the content to be reviewed, the review result of the content to be reviewed is determined as approved; if the response intent information indicates that the content moderator has disapproved the content to be reviewed, the review result of the content to be reviewed is determined as disapproved.

Description

Technical Field

[0001] This disclosure relates to a content moderation method, electronic device, storage medium, and program product. Background Technology

[0002] In the field of digital service platforms, with the rapid development of online transactions and services, various industry platforms (such as e-commerce, financial services, and medical appointments) have become core infrastructure connecting users, suppliers, and operators. The business review process, as a critical node in the process, directly impacts the platform's compliance, operational efficiency, and user satisfaction.

[0003] Existing digital service platforms typically establish a strict set of business rules to ensure the standardization of transactions or services and the controllability of risks. Due to the complexity of business scenarios and the frequent occurrence of extreme situations, if existing business rules are not updated or adjusted in a timely manner, or if the existing business rules are not perfect, the platform's review system may fail to approve the business content according to the business rules, requiring manual updates or supplements to the business rules.

[0004] In the above scenario, supplier personnel and operations staff need to communicate and report business details to their superiors via instant messaging software, creating a review and communication record. These personnel can then upload screenshots of the review and communication record to the platform's review system and initiate a review request. The platform's review system personnel will review the uploaded screenshots and explanations, and based on the superior's response (such as approval), manually approve the review, allowing the business to proceed.

[0005] However, existing technical solutions rely on manual methods to identify and review communication records. Reviewers must read each screenshot and manually compare it with relevant business information, resulting in a lengthy and inefficient review process. Furthermore, the traceability, completeness, and consistency of communication records are difficult to guarantee, thus failing to meet the needs of efficient and compliant digital business management. In addition, manual review is easily affected by subjective factors, making it difficult to achieve standardized processing across scenarios and industries. Summary of the Invention

[0006] This disclosure provides a content moderation method, electronic device, storage medium, and program product.

[0007] According to one aspect of this disclosure, a content moderation method is provided, comprising: Obtain screenshots of communication related to the content to be reviewed; Extract the text content, background color, text location information, and identity information of the communication participants from the screenshot of the review communication. The communication participants include the content submitter and the content reviewer. After determining the time sequence of the text communication information based on the location information, and determining the sender information of the text communication information based on the background color and the identity information of the communication participants, structured information is generated. The structured information contains text communication information that identifies the sender information, and the text communication information is arranged in order of sending time. Semantic parsing is performed on the structured information to obtain the response intent information of the content reviewer in the text communication information; If the response intent information indicates that the content reviewer has approved the content to be reviewed, the review result of the content to be reviewed is determined to be approved; If the response intent information indicates that the content reviewer does not approve the review of the content to be reviewed, the review result of the content to be reviewed will be determined as a failure to pass the review.

[0008] The above-disclosed technical solution decomposes the content review process into automated steps such as screenshot acquisition, information extraction, message sequence reconstruction, and semantic parsing, thereby transforming the manual review process into intelligent decision-making, thus shortening the review time, improving consistency, and ensuring the compliance and efficiency of digital business.

[0009] According to at least one embodiment of the content moderation method of this disclosure, after determining the temporal order of the text communication information based on the location information and determining the sender information of the text communication information based on the background color and the identity information of the communication participants, structured information is generated, including: The text communication information is sorted based on the location information to obtain the sorted content; The background color is matched with a preset color role mapping rule to determine the role of the sender of the text communication information in the communication process; Based on the identity information of the communication participants, determine the sender information corresponding to the role; The sorted content and the corresponding sender information are combined in chronological order to obtain the structured information.

[0010] According to the technical solution of this embodiment, it is possible to sequence and assign roles to scattered and unstructured chat information in the review communication screenshots, thereby improving the machine understandability of communication records and the degree of review automation, and significantly improving the system's processing efficiency and compliance traceability.

[0011] According to at least one embodiment of the content moderation method of this disclosure, after determining the temporal order of the text communication information based on the location information and determining the sender information of the text communication information based on the background color and the identity information of the communication participants, structured information is generated, including: Based on the content to be reviewed, the text communication information and the identity information of the communication participants are verified to obtain the verification result; In response to the verification result being successful, the time sequence of the text communication information is determined based on the location information, and the sender information of the text communication information is determined based on the background color and the identity information of the communication participants, and then the structured information is generated.

[0012] According to the technical solution of this embodiment, the above information can be guaranteed to be accurate and reliable, the efficiency and accuracy of automatic review can be improved, and the traceability and compliance of data can be enhanced.

[0013] According to at least one embodiment of the content review method of this disclosure, the step of verifying the text communication information and the identity information of the communication participants based on the content to be reviewed, and obtaining the verification result, includes: Based on the content to be reviewed, one or more of the following checks are performed on the text communication information and the identity information of the communication participants: semantic consistency check, business authenticity check, and identity consistency check, to obtain the check result.

[0014] According to the technical solution of this implementation method, abnormal or non-compliant information can be eliminated before reconstruction, thereby improving the accuracy, efficiency and business compliance of the audit. At the same time, a traceable and reliable audit data chain is formed, further ensuring business compliance and security.

[0015] According to at least one embodiment of the content moderation method of this disclosure, before obtaining the screenshots of the moderation communication related to the content to be moderated, the content moderation method further includes: Determine whether the content to be reviewed conforms to the preset review rules; In response to the fact that the content to be reviewed does not conform to the preset review rules, obtain the review communication screenshots related to the content to be reviewed.

[0016] According to the technical solution of this embodiment, unnecessary acquisition of screenshots for review communication can be reduced, system resources can be saved, and the review process can be made more efficient and intelligent.

[0017] According to at least one embodiment of the content review method of this disclosure, determining whether the content to be reviewed conforms to preset review rules includes: The content to be reviewed is formatted and cleaned to obtain the processed content. Determine whether the processed content conforms to the preset review rules.

[0018] The technical solution of this embodiment can ensure the accuracy and efficiency of rule judgment, while reducing manual intervention and improving the automation of the review process.

[0019] According to at least one embodiment of the content review method of this disclosure, determining whether the content to be reviewed conforms to preset review rules includes: Retrieve the review rules corresponding to the content to be reviewed from the preset rule base; Determine whether the content to be reviewed conforms to the review rules.

[0020] According to the technical solution of this implementation method, the compliance of the content to be reviewed can be determined quickly, accurately and automatically, thereby improving review efficiency, reducing labor costs, and ensuring the standardization and traceability of business operations.

[0021] According to at least one embodiment of the content moderation method of this disclosure, the step of determining the review result of the content to be reviewed as "review failed" if the response intent information indicates that the content moderator does not approve the review of the content to be reviewed includes: If the response intent information indicates that the content reviewer does not approve the content to be reviewed, the content to be reviewed and the screenshot of the review communication will be pushed to the reviewer. In response to receiving a notification from the reviewer that the content to be reviewed has failed the review, the review result of the content to be reviewed is determined as "review failed".

[0022] According to the technical solution of this embodiment, it is possible to handle manual intervention in abnormal situations, while improving the compliance, traceability and overall audit efficiency of the process.

[0023] According to at least one embodiment of the content moderation method of this disclosure, in response to the semantic parsing being performed using a large language model, after the content to be reviewed and the screenshot of the review communication are pushed to the reviewers, the method further includes: Obtain the manual review results input by the reviewer for the content to be reviewed; In response to the manual review result indicating approval, the large language model is optimized based on the manual review result.

[0024] According to the technical solution of this embodiment, the large language model can learn and continuously improve the logic of human judgment, thereby improving the accuracy and consistency of automatic review results, reducing the burden of human review, and building an iteratively optimized intelligent review closed-loop mechanism.

[0025] According to at least one embodiment of the content moderation method of this disclosure, the step of semantically parsing the structured information to obtain the content moderator's response intent information in the text communication information includes: The structured information is input into multiple artificial intelligence models to obtain multiple intent information; Based on the multiple intent information, the results are fused and weighted to obtain the response intent information.

[0026] According to the technical solution of this embodiment, the differences and complementary advantages of semantic understanding among multiple artificial intelligence models can be fully utilized to achieve high-precision semantic recognition and robust judgment, thereby improving the system's automatic review and response decision-making capabilities in complex communication scenarios.

[0027] According to at least one embodiment of the content moderation method of this disclosure, the step of semantically parsing the structured information to obtain the content moderator's response intent information in the text communication information includes: Obtain the business context information of the content to be reviewed; Based on the structured information and the business context information, semantic parsing is performed to obtain the content reviewer's response intent information in the text communication information.

[0028] The technical solution of this embodiment can significantly improve the accuracy of semantic parsing and business matching, ensuring that the review and judgment are more intelligent and consistent.

[0029] According to at least one embodiment of the content review method of this disclosure, after determining that the review result of the content to be reviewed is approved, the method further includes: Generate review log information for the content to be reviewed.

[0030] According to the technical solution of this implementation method, the entire process of content review can be recorded and traced, ensuring that the review decision-making process is transparent and the results are verifiable. At the same time, it provides data support for anomaly review, compliance audit and model optimization, thereby improving the credibility of the system and the compliance of business management.

[0031] According to another aspect of this disclosure, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, causing the processor to perform a content moderation method according to any embodiment of this disclosure.

[0032] According to another aspect of this disclosure, a readable storage medium is provided, wherein executable instructions are stored therein, which, when executed by a processor, are used to implement the content moderation method of any embodiment of this disclosure.

[0033] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the content moderation method of any embodiment of this disclosure. Attached Figure Description

[0034] The accompanying drawings illustrate exemplary embodiments of the present disclosure and, together with the description thereof, serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0035] Figure 1 This is a schematic diagram illustrating an application scenario of a content moderation method according to one embodiment of the present disclosure.

[0036] Figure 2 This is an illustrative interactive flow of a content moderation method according to one embodiment of the present disclosure. Figure 1 .

[0037] Figure 3 yes Figure 2 The illustrated interactive flow of the reconstruction method in the content moderation approach is shown. Figure 1 .

[0038] Figure 4 yes Figure 2 The illustrated interactive flow of the reconstruction method in the content moderation approach is shown. Figure 2 .

[0039] Figure 5 yes Figure 2 The illustrated interactive flow of the reconstruction method in the content moderation approach is shown. Figure 3 .

[0040] Figure 6 yes Figure 2 The diagram illustrates the interactive process of obtaining screenshots in the content moderation method.

[0041] Figure 7 yes Figure 6 The screenshot acquisition method shown illustrates the interactive process of the rule judgment method. Figure 1 .

[0042] Figure 8 yes Figure 6 The screenshot acquisition method shown illustrates the interactive process of the rule judgment method. Figure 2 .

[0043] Figure 9 This is an illustrative interactive flow of a content moderation method according to one embodiment of the present disclosure. Figure 2 .

[0044] Figure 10 This is an illustrative interactive flow of a content moderation method according to one embodiment of the present disclosure. Figure 3 .

[0045] Figure 11 yes Figure 2 The illustrated interactive flow of semantic analysis methods in the content moderation approach is shown. Figure 1.

[0046] Figure 12 yes Figure 2 The illustrated interactive flow of semantic analysis methods in the content moderation approach is shown. Figure 2 .

[0047] Figure 13 This is an illustrative interactive flow of a content moderation method according to one embodiment of the present disclosure. Figure 4 .

[0048] Figure 14 This is an illustrative interactive flowchart of a content moderation method according to one embodiment of the present disclosure.

[0049] Figure 15 This is a schematic structural block diagram of a content review device according to one embodiment of the present disclosure.

[0050] Figure 16 This is a schematic structural block diagram of an electronic device according to one embodiment of the present disclosure. Detailed Implementation

[0051] The present disclosure will now be described in further detail with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0052] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0053] This disclosure proposes a content moderation method.

[0054] Figure 1 A schematic diagram illustrating an application scenario of this disclosure is shown. In this application scenario, a user terminal 100 and a server 200 may be included. The user terminal 100 is connected to the server 200 via a network.

[0055] Figure 2 A schematic diagram illustrating the overall flow of a content moderation method according to one embodiment of this disclosure is shown. Figure 2 The content moderation method M200 shown includes steps S210 to S260. This content moderation method can be executed by an electronic device such as a server.

[0056] In step S210, obtain screenshots of the review communication related to the content to be reviewed.

[0057] In some embodiments of this disclosure, the content to be reviewed in step S210 is usually business data or operational information that needs to be judged by the system or by humans in terms of compliance, authenticity, validity, etc.

[0058] The information to be reviewed in this disclosure includes, but is not limited to, the following types of information: business operation request information (such as room check-out application, refund application, order modification application, contract change application, etc.), attachment or voucher information (such as scanned copies of contracts, invoices, vouchers, identity verification materials, etc.), and operation behavior information (such as user-submitted feedback, appeal content, abnormal operation reports, etc.).

[0059] When submitting a review application for content to be reviewed, relevant users can proactively upload screenshots of the review communication, so that step S210 can obtain the review communication screenshots through the upload interface.

[0060] Step S210 can also retrieve screenshots of audit communications through database association tables or file storage indexes. Step S210 can also automatically capture audit communication screenshots through the communication platform's open interface.

[0061] The review communication screenshots obtained through step S210 are image files that record the content of communication between users, generated through instant messaging tools, collaboration platforms, or online communication systems.

[0062] Screenshots of communication reviews can include communication information, the identities of the participants, and the time of the chat. These screenshots typically use a bubble-style message layout, where each message is presented as an independent bubble. The bubbles for different participants differ in color; for example, one party's bubble might be light blue while the other's is white, or one party's bubble might be dark blue while the other's is gray.

[0063] In step S220, the text content, background color, text location information, and identity information of the communication participants are extracted from the screenshot of the communication review. The communication participants include the content submitter and the content reviewer.

[0064] In some embodiments of this disclosure, step S220 may employ methods such as structured extraction, deep learning models, or template matching to extract the aforementioned information.

[0065] Specifically, to improve the accuracy of information extraction, the screenshots of the review communication can be preprocessed before extraction (e.g., size normalization, noise removal, and color space conversion, or one or more of these). The information extracted in step S220 can be presented in the form of structured data records, thereby significantly improving the efficiency and accuracy of content review. Before extraction, the completeness, authenticity, and consistency of the review communication screenshots can also be verified, thereby enhancing the security and compliance of the review process.

[0066] The background color extracted in step S220 is typically the color of the message bubble carrying the communication information. The location information extracted in step S220 is typically the boundary coordinates of the message bubble, such as the coordinates of the top-left and bottom-right corners. The communication party identity information extracted in step S220 may include the current user's identity information and the other party's user identity information. The communication party identity information can include username, avatar, account ID, etc.

[0067] In step S230, after determining the time sequence of text communication information based on location information and determining the sender information of text communication information based on background color and identity information of communication participants, structured information is generated. The structured information contains text communication information that identifies the sender information, and the text communication information is arranged in the order of sending time.

[0068] In step S240, the structured information is semantically parsed to obtain the content reviewer's response intent information in the text communication information.

[0069] In some embodiments of this disclosure, step S240 may employ rule-based methods (such as keyword matching, regular expressions, template matching), traditional machine learning methods, deep learning methods, and other methods for semantic parsing.

[0070] In step S250, if the response intent information indicates that the content reviewer has approved the content to be reviewed, the review result of the content to be reviewed is determined as approved.

[0071] In some embodiments of this disclosure, step S250 may employ methods such as automatic system marking, workflow control, or log recording to determine the review result of the content to be reviewed as approved. Once step S250 determines the review result of the content to be reviewed as approved, it indicates that the content complies with platform rules or superior instructions, and subsequent operational steps (such as resource allocation, transaction execution, room release operations, etc.) can continue to be executed.

[0072] In step S260, if the response intent information indicates that the content reviewer does not approve the review of the content to be reviewed, the review result of the content to be reviewed is determined as unapproved.

[0073] The content moderation method disclosed herein decomposes the content moderation process into automated steps such as screenshot acquisition, information extraction, message sequence reconstruction, and semantic parsing, thereby transforming the manual moderation process into intelligent decision-making, thus shortening the moderation time, improving consistency, and ensuring the compliance and efficiency of digital business.

[0074] The content moderation method disclosed herein solves the problem of relying on manual methods to identify and review communication records in the prior art.

[0075] Regarding step S230, in some embodiments of this disclosure, it may include, for example... Figure 3 Steps S231 to S234 are shown.

[0076] In step S231, the text communication information is sorted based on the text location information to obtain the sorted content.

[0077] In some embodiments of this disclosure, for single-column layout screenshots of review communications, step S231 can directly sort the text position information in descending order. For multi-column or complex layout screenshots of review communications, step S231 can combine text position information for hierarchical sorting.

[0078] Step S231 can also use methods such as geometric clustering sorting or deep learning sorting models to sort text communication information (e.g., dialogue messages).

[0079] The sorted content obtained through step S231 can accurately reflect the actual exchange order of messages between the two parties, enabling the system to accurately reconstruct the complete dialogue process and provide a time-consistent data foundation for subsequent processing.

[0080] In step S232, the background color is matched with a preset color role mapping rule to determine the role of the sender of the text communication information in the communication process.

[0081] In some embodiments of this disclosure, step S232 may employ color feature matching methods, color clustering algorithms, etc., for matching. In step S232, the communicating parties of the message can distinguish whether a message was sent by the current user or by the other user. The roles determined through step S232 may include the requester (i.e., the content submitter), the content reviewer, etc.

[0082] In step S233, the sender information corresponding to the role is determined based on the identity information of the communication participants.

[0083] In step S234, the sorted content and the corresponding sender information are combined in chronological order to obtain structured information.

[0084] The structured information obtained through step S234 is structured dialogue data, which typically includes a message sequence arranged in chronological order, the identities of the sender and receiver of each message, etc.

[0085] Steps S231 to S234 enable the sequentialization and role-based representation of scattered, unstructured chat information in the review communication screenshots, thereby improving the machine understandability of communication records and the degree of review automation, and significantly improving the system's processing efficiency and compliance traceability.

[0086] Regarding step S230, in some embodiments of this disclosure, it may also include, as follows: Figure 4 Steps S235 to S236 are shown.

[0087] In step S235, the text communication information and the identity information of the communication participants are verified based on the content to be reviewed, and the verification result is obtained.

[0088] In some embodiments of this disclosure, step S235 can perform comprehensive verification of the semantic consistency, identity authenticity, and rule compliance of the content to be reviewed, ensuring the accuracy and credibility of the review decision and providing a reliable foundation for subsequent automated review.

[0089] The verification result obtained through step S235 can include verification passed and verification failed.

[0090] If the verification result obtained through step S235 is that the verification passed, then step S236 is executed.

[0091] When the verification result obtained through step S235 is that the verification fails, it can be downgraded to manual review. That is, the content to be reviewed, the screenshot of the review communication and the verification result are pushed to the reviewer, who will review and judge to determine whether to approve it. Alternatively, the message or field that fails the verification can be marked to facilitate subsequent traceability or review analysis.

[0092] In step S236, after determining the temporal order of the text communication information based on the location information and determining the sender information of the text communication information based on the background color and the identity information of the communication participants, the structured information described above is generated.

[0093] In particular, during the verification process, anomaly detection and automatic early warning mechanisms can be further introduced.

[0094] When the system detects semantic inconsistencies, field conflicts, or risky behavior patterns (such as repeatedly initiating high-amount refunds within a short period or frequently modifying the same contract field) in multiple submissions of content awaiting review, the system can mark high-risk content using clustering detection or anomaly distribution analysis methods and automatically trigger the review process by auditors. This mechanism can detect potential violations or malicious behavior in advance, enhancing the system's risk control capabilities in complex business scenarios.

[0095] Steps S235 to S236 ensure that the above information is accurate and reliable, improve the efficiency and accuracy of automatic review, and enhance the traceability and compliance of the data.

[0096] Regarding step S235, in some embodiments of this disclosure, it can be replaced with, for example... Figure 5 Step S237 is shown.

[0097] In step S237, based on the content to be reviewed, one or more of the following checks are performed on the text communication information and the identity information of the communication participants: semantic consistency check, business authenticity check, and identity consistency check, to obtain the check result.

[0098] In some embodiments of this disclosure, the semantic consistency verification in step S237 refers to semantically comparing the content to be reviewed with the communication information to determine whether the business intent or key information expressed by the two is consistent, so as to prevent misjudgment or misoperation caused by information misunderstanding.

[0099] Step S237, business authenticity verification, refers to verifying whether the business information involved in the communication and the content to be reviewed is true, reasonable and in line with the actual situation, to prevent false applications or data tampering, to ensure that the review decision is based on real business information, and to enhance the reliability of the process and the ability to prevent fraud.

[0100] In step S237, the identity consistency verification refers to verifying whether the identity information of the sender or receiver is consistent with the system registration information or the identity approved by the superior, ensuring that the source of the message is legal and reliable, preventing impersonation, tampering or unauthorized operation, and providing security for business review.

[0101] Step S237 allows for the elimination of abnormal or non-compliant information before reconstruction, improving the accuracy, efficiency, and business compliance of audits. It also creates a traceable and reliable audit data chain, further ensuring business compliance and security.

[0102] Before performing step S210 and obtaining screenshots of review communications related to the content to be reviewed, in some embodiments of this disclosure, the content review method of this disclosure may include, for example: Figure 6 Steps S211 to S213 are shown.

[0103] In step S211, it is determined whether the content to be reviewed conforms to the preset review rules.

[0104] In some embodiments of this disclosure, step S211 may use methods such as rule engine comparison, conditional expression or logical judgment, model or algorithm prediction, knowledge graph or rule graph to determine whether it conforms to the preset review rules.

[0105] Pre-defined review rules are a set of standards or conditions that are predefined to determine whether the content to be reviewed meets business requirements.

[0106] Preset audit rules can be reflected in the following content or form: business constraints, compliance and policy rules, format and field rules, and multi-condition combination rules, etc.

[0107] If step S211 determines that the content to be reviewed meets the preset review rules, proceed to step S212; if step S211 determines that the content to be reviewed does not meet the preset review rules, proceed to step S213.

[0108] In step S212, the content to be reviewed is set to "approved".

[0109] In step S213, obtain screenshots of the review communication related to the content to be reviewed.

[0110] Steps S211 to S213 can reduce unnecessary acquisition of review communication screenshots, save system resources, and make the review process more efficient and intelligent.

[0111] Regarding step S211, in some embodiments of this disclosure, it may include, for example... Figure 7 Steps S2111 to S2112 are shown.

[0112] In step S2111, the content to be reviewed is formatted and cleaned up to obtain the processed content.

[0113] In some embodiments of this disclosure, step S2111 can employ methods such as character standardization, number and date normalization, field standardization, and space and line break normalization for format standardization. Step S2111 can employ methods such as noise character removal, duplicate content cleanup, pinyin and grammar correction, and non-business information filtering for text cleaning. The processed content obtained through step S2111 is typically clean, standardized, and structured.

[0114] In step S2112, it is determined whether the processed content conforms to the preset review rules.

[0115] Steps S2111 to S2112 can ensure the accuracy and efficiency of rule judgment, while reducing manual intervention and improving the automation of the review process.

[0116] Regarding step S211, in some embodiments of this disclosure, it may also include, as follows: Figure 8 Steps S2113 to S2114 are shown.

[0117] In step S2113, the review rules corresponding to the content to be reviewed are obtained from the preset rule base.

[0118] In some embodiments of this disclosure, step S2113 may employ rule matching methods, rule query and retrieval methods, conditional logic mapping methods, model-assisted matching methods, etc., to obtain the audit rules.

[0119] In step S2113, the preset rule base is a set of rules that are predefined or configured in the system and used to determine whether the content to be reviewed meets business and compliance requirements.

[0120] The rules in the preset rule base can be divided according to dimensions such as region, department, risk level, priority, time, effective period, rule type, and triggering conditions.

[0121] In step S2114, it is determined whether the content to be reviewed conforms to the review rules.

[0122] Furthermore, in the rule matching stage from step S2113 to step S2114, knowledge graph and semantic reasoning techniques can be combined to achieve dynamic expansion and adaptive updating of rules.

[0123] Specifically, the system can construct semantic relationships between rule nodes based on domain knowledge graphs, and dynamically identify rule subsets related to the current content to be reviewed through logical reasoning algorithms (such as rule reasoning based on first-order predicate logic and semantic matching based on graph embedding).

[0124] For example, when the content to be reviewed involves "contract modification" and the related fields "rent adjustment" and "renewal upon expiration" are detected, the system can automatically activate the composite rule set under the "contract modification" scenario to ensure the completeness of the review coverage. Through this knowledge graph-based intelligent rule mapping mechanism, the system can achieve scalable management and self-evolutionary updates of the rule base, thereby reducing manual maintenance costs and improving the flexibility and accuracy of the review strategy.

[0125] Steps S2113 to S2114 can quickly, accurately, and automatically determine the compliance of the content to be reviewed, improve review efficiency, reduce labor costs, and ensure the standardization and traceability of business operations.

[0126] Regarding step S260, in some embodiments of this disclosure, it may include, for example... Figure 9 Steps S261 to S262 are shown.

[0127] In step S261, if the response intent information indicates that the content reviewer does not approve the content to be reviewed, the content to be reviewed and the screenshot of the review communication will be pushed to the reviewer.

[0128] In some embodiments of this disclosure, step S261 may push the content to be reviewed and screenshots of review communications to the reviewers through methods such as system message push, task or work order creation, and permission and role control. After receiving the content to be reviewed and the communication records, the reviewers may perform content verification, review, and other operations, and submit the manual review results to the server.

[0129] In step S262, in response to receiving a rejection from the reviewer that the content to be reviewed has been rejected, the review result of the content to be reviewed is determined to be rejected.

[0130] Steps S261 to S262 can handle human intervention in abnormal situations, while improving the compliance, traceability and overall audit efficiency of the process.

[0131] The content moderation method provided in this disclosure, which responds to semantic parsing using a large language model, may further include, after step S260, the following: Figure 10 Steps S270 to S280 are shown.

[0132] In step S270, the manual review results entered by the reviewers for the content to be reviewed are obtained.

[0133] In step S280, in response to the manual review result indicating approval, the large language model is optimized based on the manual review result.

[0134] In some embodiments of this disclosure, step S280 may employ a model optimization method based on feedback learning, reinforcement learning, or rule distillation. When the manual review result indicates approval, the manual review result is used to perform incremental training and parameter tuning on the large language model.

[0135] In steps S270 to S280, a "human-machine collaborative feedback link" can be constructed based on the results of manual review to continuously optimize the review and judgment capabilities of the large language model. This link can structurally store the reasons for manual review, correction opinions, and final decision results, and incrementally train the model through knowledge distillation or reinforcement learning algorithms, enabling the model to gradually learn the logic and implicit rules of manual review. For example, when reviewers repeatedly correct the "partially approved" type content that the model misjudged, the system can automatically adjust the model's confidence weight under similar semantics, thereby reducing the misjudgment rate in subsequent automatic reviews.

[0136] Steps S270 to S280 enable the large language model to learn and continuously improve the logic of human judgment, thereby improving the accuracy and consistency of automatic review results, reducing the burden of human review, and building an iteratively optimized intelligent review closed-loop mechanism.

[0137] Regarding step S240 (semantically parsing the structured information to obtain the content reviewer's response intent information in the text communication information), in some embodiments of this disclosure, it may include, for example... Figure 11 Steps S241 to S242 are shown.

[0138] In step S241, structured information is input into multiple artificial intelligence models to obtain multiple intent information.

[0139] In some embodiments of this disclosure, the multiple artificial intelligence models in step S241 may include a variety of natural language understanding models, intent recognition models, sentiment analysis models, dialogue understanding models, contextual reasoning models, and multimodal fusion models.

[0140] In step S242, the results are fused and weighted based on multiple intent information to obtain the response intent information.

[0141] In some embodiments of this disclosure, step S242 may employ one or more of the following methods to perform result fusion and weighted decision-making based on multiple intent information: weighted voting, confidence fusion, Bayesian inference, ensemble learning, feature-level fusion, or dynamic weight adjustment.

[0142] Steps S241 to S242 can fully utilize the differences and complementary advantages of semantic understanding among multiple artificial intelligence models to achieve high-precision semantic recognition and robust judgment, thereby improving the system's automatic review and response decision-making capabilities in complex communication scenarios.

[0143] Regarding step S240, in some embodiments of this disclosure, it may also include, as follows: Figure 12 Steps S243 to S244 are shown.

[0144] In step S243, the business context information of the content to be reviewed is obtained.

[0145] In some embodiments of this disclosure, step S243 may employ methods such as structured data extraction, semantic analysis, user behavior mining, knowledge graph association reasoning, or log tracking to obtain business context information.

[0146] In step S243, the business context information is auxiliary information related to the content to be reviewed in terms of business process, semantic scenario, user identity, operation behavior, or system status.

[0147] Business context information can include business process information, communication context information, user attribute information, time and location information, system status information, etc.

[0148] In step S244, semantic parsing is performed based on structured information and business context information to obtain the content reviewer's response intent information in the text communication information.

[0149] Steps S243 to S244 can significantly improve the accuracy of semantic parsing and business matching, ensuring that the review judgment is more intelligent and consistent.

[0150] Furthermore, the content moderation method provided in this disclosure may further include, after step S250 or step S260, the following steps: Figure 13 Step S290 is shown.

[0151] In step S290, audit log information for the content to be audited is generated.

[0152] In some embodiments of this disclosure, step S290 may use event tracking methods, process auditing methods, data persistence methods, etc., to generate audit log information.

[0153] The review log information is structured data recorded during the content review process, which can reflect the various operations and system status in the review process.

[0154] Audit log information may include the audit initiation time, the auditor, the audit result, a detailed description of human intervention or model decision-making, and communication records involved in the audit process.

[0155] Step S290 enables full-process traceability and management of content review, ensuring transparency in the review decision-making process and verifiability of results. It also provides data support for anomaly review, compliance audit, and model optimization, thereby enhancing the credibility of the system and the compliance of business management.

[0156] Figure 14 An exemplary flowchart based on the content moderation method of this disclosure is shown.

[0157] Figure 14 In the flowchart shown, taking the rental contract as an example, the content review process can include steps S310 to S370.

[0158] In step S310, in response to receiving the rental contract, it is determined whether the rental contract meets the preset review rules.

[0159] In some embodiments of this disclosure, if step S310 determines that the rental contract meets the preset review rules, it means that the rental contract can be directly approved, and step S320 is executed. If step S310 determines that the rental contract does not meet the preset review rules, it means that the rental contract does not meet the platform's room rental conditions, and step S330 is executed.

[0160] In step S320, the rental contract is set to be approved.

[0161] In step S330, obtain the screenshot of the review communication uploaded by the user when submitting the rental contract for review.

[0162] In step S340, the text content, background color, text location information, and identity information of the communication participants are extracted from the screenshot of the communication review. The communication participants include the content submitter and the content reviewer.

[0163] In step S350, after determining the time sequence of text communication information based on text location information and determining the sender information of text communication information based on background color and identity information of communication participants, the structured information described above is generated.

[0164] In step S360, the structured information is semantically parsed to obtain the content reviewer's response intent information in the text communication information.

[0165] In some embodiments of this disclosure, if the response intent information obtained in step S360 indicates that the content moderator has approved the rental contract, step S320 is executed. If the response intent information obtained in step S360 indicates that the content moderator has not approved the rental contract, step S370 is executed.

[0166] In step S370, the review result of the rental contract is determined to be "approved".

[0167] This disclosure also provides a content moderation device.

[0168] Figure 15 This is a schematic structural block diagram of a content review device according to one embodiment of the present disclosure.

[0169] like Figure 15As shown, the content review device includes a screenshot acquisition module 1010, an information extraction module 1020, a message reconstruction module 1030, a semantic parsing module 1040, a review approval module 1050, and a review rejection module 1060.

[0170] The screenshot acquisition module 1010 is used to acquire screenshots of review communications related to the content to be reviewed.

[0171] The information extraction module 1020 is used to extract text communication information, background color of text communication information, location information of text communication information, and identity information of communication participants from the review communication screenshot. Communication participants include the content submitter and the content reviewer.

[0172] The message reconstruction module 1030 is used to determine the time sequence of text communication information based on location information, and after determining the sender information of the text communication information based on background color and the identity information of the communication participants, it generates structured information of text communication information that identifies the sender information and arranges them in chronological order.

[0173] The semantic parsing module 1040 is used to perform semantic parsing on structured information to obtain the content reviewer's response intent information in text communication information.

[0174] The approval module 1050 is used to determine the approval result of the content to be reviewed as approved if the response intent information indicates that the content reviewer has approved the content to be reviewed.

[0175] The review rejection module 1060 is used to determine the review result of the content to be reviewed as "rejected" if the response intent information indicates that the content reviewer does not approve the review of the content to be reviewed.

[0176] The content review device disclosed herein can be implemented by computer software programs. The specific implementation of each module in the device can refer to the implementation process of the corresponding steps in the above-described method implementation method of this disclosure, and will not be repeated here.

[0177] Figure 16 This is a schematic structural block diagram of an electronic device according to one embodiment of the present disclosure.

[0178] like Figure 16As shown, the hardware architecture of an electronic device can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application and overall design constraints of the hardware. Bus 1100 connects various circuits including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400 such as peripherals, voltage regulators, power management circuits, external antennas, etc. Bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, only one connection line is used in this figure, but this does not indicate that there is only one bus or one type of bus.

[0179] For ease of explanation, certain steps of the above method are described in relation to modules. It should be understood that the corresponding module performing one or more steps of the above method may be one or more hardware modules specifically configured to perform the corresponding step, or implemented by a processor configured to perform the corresponding step, or stored in a computer-readable medium for implementation by a processor, or implemented by some combination thereof.

[0180] This disclosure also provides a readable storage medium storing a computer program that, when executed by a processor, is used to implement the methods described above. A "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples of a readable storage medium include: an electrical connection with one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM), etc.

[0181] This disclosure also provides a computer program product, the methods of which can be implemented wholly or partially through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented wholly or partially as a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed, all or part of the processes or functions of this disclosure are performed.

[0182] Computer programs or instructions can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any available medium capable of access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video optical disc; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or it can include both volatile and non-volatile types of storage media.

[0183] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0184] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0185] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0186] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0187] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., refer to specific features, structures, or characteristics described in connection with that embodiment / mode or example, which are included in at least one embodiment / mode or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0188] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. A content moderation method, characterized in that, include: Obtain screenshots of communication related to the content to be reviewed; Extract the text content, background color, text location information, and identity information of the communication participants from the screenshot of the review communication. The communication participants include the content submitter and the content reviewer. After determining the time sequence of the text communication information based on the location information, and determining the sender information of the text communication information based on the background color and the identity information of the communication participants, structured information is generated. The structured information contains text communication information that identifies the sender information, and the text communication information is arranged in order of sending time. Semantic parsing is performed on the structured information to obtain the response intent information of the content reviewer in the text communication information; as well as If the response intent information indicates that the content reviewer has approved the content to be reviewed, the review result of the content to be reviewed will be determined as approved; If the response intent information indicates that the content reviewer has failed to approve the content to be reviewed, the review result of the content to be reviewed will be determined as a failure to approve.

2. The content moderation method as described in claim 1, characterized in that, After determining the temporal order of the text communication information based on the location information, and determining the sender information of the text communication information based on the background color and the identity information of the communication participants, structured information is generated, including: The text communication information is sorted based on the location information to obtain the sorted content; The background color is matched with a preset color role mapping rule to determine the role of the sender of the text communication information in the communication process; Based on the identity information of the communication participants, determine the sender information corresponding to the role; and The sorted content and the corresponding sender information are combined in chronological order to obtain the structured information.

3. The content moderation method as described in claim 1, characterized in that, After determining the temporal order of the text communication information based on the location information, and determining the sender information of the text communication information based on the background color and the identity information of the communication participants, structured information is generated, including: Based on the content to be reviewed, the text communication information and the identity information of the communication participants are verified to obtain the verification result; and In response to the verification result being successful, the time sequence of the text communication information is determined based on the location information, and the sender information of the text communication information is determined based on the background color and the identity information of the communication participants, and then the structured information is generated.

4. The content review method as described in claim 3, characterized in that, The verification of the text communication information and the identity information of the communication participants based on the content to be reviewed, and the resulting verification results, include: Based on the content to be reviewed, one or more of the following checks are performed on the text communication information and the identity information of the communication participants: semantic consistency check, business authenticity check, and identity consistency check, to obtain the check result.

5. The content moderation method as described in any one of claims 1 to 4, characterized in that, Before obtaining screenshots of review communications related to the content to be reviewed, the method further includes: Determine whether the content to be reviewed conforms to the preset review rules; and In response to the fact that the content to be reviewed does not conform to the preset review rules, obtain the review communication screenshots related to the content to be reviewed.

6. The content moderation method as described in claim 5, characterized in that, The step of determining whether the content to be reviewed meets the preset review rules includes: The content to be reviewed is formatted and cleaned to obtain the processed content; and Determine whether the processed content conforms to the preset review rules.

7. The content moderation method as described in claim 5, characterized in that, The step of determining whether the content to be reviewed meets the preset review rules includes: Retrieve the review rules corresponding to the content to be reviewed from the preset rule base; and Determine whether the content to be reviewed conforms to the review rules.

8. The content moderation method as described in any one of claims 1 to 4, characterized in that, If the response intent information indicates that the content reviewer does not approve the review of the content to be reviewed, determining the review result of the content to be reviewed as unapproved includes: If the response intent indicates that the content reviewer does not approve the content to be reviewed, the content to be reviewed and the screenshot of the review communication will be pushed to the reviewer; and In response to receiving a notification from the reviewer that the content to be reviewed has failed the review, the review result of the content to be reviewed is determined as a failure to pass the review.

9. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes execution instructions stored in the memory, causing the processor to perform the content moderation method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the content moderation method as described in any one of claims 1 to 8.

Images