Comment content processing method, apparatus, device, medium, and program product

Abstract

Embodiments of the present disclosure disclose a review content processing method, device, equipment, medium and program product. A specific implementation of the method comprises: obtaining review content input in real time by a review user on a review publishing page and real-time user information corresponding to the review user; generating first review risk information by using a first multi-modal large model for real-time risk prediction according to the review content and the real-time user information; in response to the first review risk information representing that the review content has risks, adjusting the page content of the review publishing page according to the first review risk information to add risk verification content; and in response to completing the content reply to the risk verification content, submitting the complete review content input on the review publishing page for processing. The implementation is related to page review, and can perform risk monitoring in the process of the review user inputting the review content, so as to perform corresponding page content adjustment and realize risk verification processing.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and specifically to comment content processing methods, apparatus, devices, media, and program products. Background Technology

[0002] Currently, with the continuous development of e-commerce and local services, user reviews have become an important basis for consumers' purchasing decisions. However, there is a proliferation of fake reviews (including fake positive reviews and malicious positive reviews), which seriously damages the platform's credit system. To verify the authenticity of review content, compliant automated or manual review methods are used to verify the authenticity of user comments after they are posted.

[0003] However, the inventors discovered that the following technical problems often arise when using the above method: Risk assessment of the review content should be conducted after it is published. Even if a risk is identified, the review content has already been uploaded to the platform and displayed, causing a negative impact and resulting in a delay in eliminating the impact. Summary of the Invention

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

[0005] Some embodiments of this disclosure provide methods, apparatus, devices, media, and program products for processing comment content to address the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide a method for processing review content, including: obtaining review content input in real time by a review user on a review posting page and real-time user information corresponding to the review user; generating first review risk information based on the review content and the real-time user information using a first multimodal large model for real-time risk prediction; in response to the first review risk information indicating that the review content has a risk, adjusting the page content of the review posting page to add risk verification content based on the first review risk information; and submitting the complete review content input on the review posting page in response to completing the content response to the risk verification content.

[0007] Optionally, the aforementioned risk verification content is review guidance information; and the aforementioned adjustment of the page content of the aforementioned review publishing page based on the aforementioned first review risk information includes: in response to the aforementioned first review risk information indicating that the risk level corresponding to the aforementioned review content is the first risk level, displaying the aforementioned review guidance information on the aforementioned review publishing page.

[0008] Optionally, the aforementioned risk verification content is an anomaly alert pop-up; and the aforementioned adjustment of the page content of the aforementioned review publishing page based on the aforementioned first review risk information includes: in response to the aforementioned first review risk information indicating that the aforementioned risk level is a second risk level, an anomaly alert pop-up window is displayed on the aforementioned review publishing page to provide anomaly alerts and anomaly verification operation information.

[0009] Optionally, before submitting the complete review content entered on the review posting page in response to completing the content response to the aforementioned risk verification content, the method further includes: in response to detecting that the review user has completed the verification operation corresponding to the aforementioned abnormal verification operation information, determining that the content response to the aforementioned risk verification content has been completed.

[0010] Optionally, the above method further includes: obtaining offline user information corresponding to the aforementioned review users; generating second review risk information using a second multimodal large model for risk prediction based on the aforementioned offline user information and the aforementioned complete review content; and performing content processing on the aforementioned complete review content based on the aforementioned second review risk information.

[0011] Optionally, the above-mentioned content processing of the complete review content based on the second review risk information includes: in response to the second review risk information indicating that the risk label corresponding to the complete review content is a first label, hiding the complete review content; and in response to the second review risk information indicating that the risk label is a second label, reviewing the complete review content again.

[0012] Optionally, the above method further includes: storing the complete review content, corresponding user information, and review risk results in a review data repository; in response to determining that the first review dataset in the review data repository meets the target conditions, training the first multimodal large model and the second multimodal large model based on the first review dataset to obtain a first trained multimodal large model and a second trained multimodal large model; replacing the deployed first multimodal large model with the first trained multimodal large model, and replacing the deployed second multimodal large model with the second trained multimodal large model.

[0013] Optionally, the first multimodal large model and / or the second multimodal large model are trained through the following steps: obtaining a second review dataset, wherein the second review data includes: review content in multimodal form, user behavior information, and review risk results; unifying the information format of the review content set in the second review dataset and performing time-series encoding on the user behavior information set to obtain a review processing dataset; training the first initial multimodal large model and / or the second initial multimodal large model based on the review processing dataset to learn multimodal content alignment knowledge and the correlation between review content and user behavior to obtain a first preliminary multimodal large model and / or a second preliminary multimodal large model; adding each review processing data in the review processing dataset to a thought chain prompt template for fake review detection to obtain a thought chain prompt information set; and retraining the first preliminary multimodal large model and / or the second preliminary multimodal large model based on the thought chain prompt information set to obtain the first multimodal large model and / or the second multimodal large model.

[0014] Secondly, some embodiments of this disclosure provide a review content processing apparatus, including: an acquisition unit configured to acquire review content input in real time by a review user on a review publishing page and real-time user information corresponding to the review user; a generation unit configured to generate first review risk information based on the review content and the real-time user information using a first multimodal large model for real-time risk prediction; an adjustment unit configured to, in response to the first review risk information indicating that the review content has a risk, adjust the page content of the review publishing page to add risk verification content; and a submission unit configured to, in response to completing a response to the risk verification content, submit the complete review content input on the review publishing page.

[0015] Optionally, the aforementioned risk verification content is review guidance information; and the adjustment unit can be configured to: in response to the aforementioned first review risk information indicating that the risk level corresponding to the aforementioned review content is a first risk level, display the aforementioned review guidance information on the aforementioned review publishing page.

[0016] Optionally, the aforementioned risk verification content is an anomaly alert pop-up; and the adjustment unit can be configured to: in response to the aforementioned first review risk information indicating that the aforementioned risk level is a second risk level, pop up an anomaly alert pop-up on the aforementioned review publishing page to provide anomaly alerts and anomaly verification operation information.

[0017] Optionally, the device further includes: in response to detecting that the user who made the review has completed the verification operation corresponding to the abnormal verification operation information, determining that a content response to the risk verification content has been completed.

[0018] Optionally, the device further includes: acquiring offline user information corresponding to the aforementioned reviewer; generating second review risk information using a second multimodal large model for risk prediction based on the aforementioned offline user information and the aforementioned complete review content; and performing content processing on the aforementioned complete review content based on the aforementioned second review risk information.

[0019] Optionally, the device further includes: in response to the second review risk information indicating that the risk label corresponding to the complete review content is a first label, hiding the complete review content; and in response to the second review risk information indicating that the risk label is a second label, re-reviewing the complete review content.

[0020] Optionally, the apparatus further includes: storing the complete review content, corresponding user information, and review risk results in a review data repository; in response to determining that the first review dataset in the review data repository meets the target conditions, training the first multimodal large model and the second multimodal large model based on the first review dataset to obtain a first trained multimodal large model and a second trained multimodal large model; replacing the deployed first multimodal large model with the first trained multimodal large model, and replacing the deployed second multimodal large model with the second trained multimodal large model.

[0021] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, such that when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0022] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method as described in any implementation of the first aspect.

[0023] Fifthly, some embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0024] The above-described embodiments of this disclosure have the following beneficial effects: Through the comment content processing methods of some embodiments of this disclosure, risk monitoring can be performed during the real-time input of comment content by comment users, allowing for corresponding page content adjustments and risk verification processing. Specifically, the reason for the delayed impact elimination due to insufficient timely risk verification is that even if risk identification is performed after the comment content is published, the comment content has already been uploaded to the platform for display, causing negative impact and resulting in a delay in impact elimination. Based on this, the comment content processing methods of some embodiments of this disclosure first obtain the comment content input in real-time by the comment user on the comment publishing page and the corresponding real-time user information. Here, by obtaining the input comment content and corresponding real-time user information in real-time on the comment publishing page, a data basis is provided for subsequent analysis of the risk information corresponding to the comment content, enabling accurate and timely determination of the first comment risk information. Then, based on the comment content and the real-time user information, the first comment risk information can be accurately and efficiently generated using a first multimodal large model for real-time risk prediction. Here, by generating initial review risk information, review risks can be identified in real-time as users input their reviews, allowing for subsequent risk mitigation and preventing delayed negative impacts. Next, in response to the initial review risk information indicating a risk in the review content, the review posting page is adjusted to add risk verification content. Here, by displaying the risk verification content on the review posting page, users are visually prompted to mitigate the risk, increasing the authenticity of the review and eliminating potential risks. Finally, in response to the completed risk verification, the complete review content entered on the review posting page is submitted, resulting in a review with significantly reduced risk. In summary, by generating initial review risk information during real-time review input, accurate and efficient real-time risk detection can be performed, eliminating potential risks in the review content. In addition, by adjusting the content on the review posting page and adding risk-related content, content risks can be effectively eliminated in a visual way. Attached Figure Description

[0025] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0026] Figure 1 This is a schematic diagram illustrating an application scenario of a comment content processing method according to some embodiments of this disclosure; Figure 2 This is a flowchart of some embodiments of the comment content processing method according to this disclosure; Figure 3 These are flowcharts of other embodiments of the comment content processing method according to this disclosure; Figure 4 These are schematic diagrams illustrating the structure of some embodiments of the comment content processing apparatus according to this disclosure; Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0027] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0028] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0029] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0030] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0031] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0032] Before performing any of the operations involving the collection, storage, or use of user personal information (such as review content) disclosed in this disclosure, the relevant organizations or individuals shall fulfill their obligations, including conducting personal information security impact assessments, informing personal information subjects, and obtaining prior authorization and consent from personal information subjects.

[0033] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0034] Figure 1 This is a schematic diagram illustrating an application scenario of a comment content processing method according to some embodiments of this disclosure.

[0035] exist Figure 1 In this application scenario, firstly, electronic device 101 can obtain the review content 103 entered in real time by the review user on the review posting page 102 and the corresponding real-time user information 104. In this application scenario, the review content 103 could be "Great value for money, very easy to use, large cup opening, and easy to clean". The real-time user information 104 could be: User ID: U100231, Login device: Mobile device 1, Product browsing path: Home carousel (Beauty Zone) → Product details page (Brand A cosmetics, ID: P10086) → Category page (Skincare → Serum) → Product details page → Shopping cart (Add A cosmetics) → Checkout page (Select "100 off for orders over 500" coupon) → Place order, Page dwell time: "Home: 1 minute 20 seconds, Product details page: 3 minutes 15 seconds, Shopping cart page: 45 seconds, Checkout page: 1 minute 10 seconds". Then, based on the aforementioned review content 103 and the aforementioned real-time user information 104, the electronic device 101 can generate first review risk information 106 using the first multimodal large model 105 for real-time risk prediction. In this application scenario, the first review risk information 106 can be "70 points". Furthermore, in response to the aforementioned first review risk information 106 indicating that the review content has a risk, the electronic device 101 can adjust the page content of the review publishing page 103 to add risk verification content 107. In this application scenario, the risk verification content 107 can be the page verification area corresponding to "Please upload a real picture of the cup". Finally, in response to completing the content response to the aforementioned risk verification content 107, the electronic device 101 can submit the complete review content entered on the aforementioned review publishing page 103. In this application scenario, the content response to the risk verification content 107 can be a real picture of the cup that the review user has uploaded. A complete review could be something like, "Great value for money, very easy to use, the cup opening is large, and it's easy to clean. It's a great deal for this price."

[0036] It should be noted that the aforementioned electronic device 101 can be either hardware or software. When the electronic device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the electronic device is software, it can be installed in the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0037] It should be understood that Figure 1 The number of electronic devices shown is merely illustrative. Any number of electronic devices can be used depending on the implementation requirements.

[0038] Continue to refer to Figure 2 The diagram illustrates a flow 200 of some embodiments of a comment content processing method according to the present disclosure. This comment content processing method includes the following steps: Step 201: Obtain the review content entered in real time by the review user on the review posting page and the real-time user information corresponding to the aforementioned review user.

[0039] In some embodiments, the entity executing the above-mentioned comment content processing method (e.g.) Figure 1 The electronic device 101 shown can acquire the real-time review content entered by the review user on the review posting page and the real-time user information corresponding to the review user via wired or wireless connection. The review user can be the user who posts the review content. In practice, the review user can be any reviewer. The review content can be the review result written by the reviewer on the reviewed object. For example, the review content could be "The product quality is very good, the price is very reasonable, and the delivery is fast." The reviewed object can be the object being reviewed. For example, the reviewed object could be the product purchased by the reviewer. The review posting page can be the page where the review content is posted. For example, the review posting page could be a product review page. The real-time entered review content can be the review content that the reviewer is currently entering on the review posting page and is yet to be posted. Here, the review content can be all the review content entered by the reviewer, or it can be content that has not yet been entered but is already displayed on the review posting page (i.e., the review content is not the complete content that the reviewer intends to enter). The real-time user information can be the user information corresponding to the reviewer at the current time. Here, the real-time user information can include: the behavioral chain information of the user's page operation at the current time. The behavioral link information may include: login device / location, product browsing path, page dwell time, whether a coupon was used, the time interval between placing an order and leaving a review, historical review order records, and user membership information.

[0040] It should be noted that the execution entity can use EventTimeSessionWindow to group the continuous activities of the reviewing user within 30 minutes into a session, and use this as a unit to construct the behavioral chain information of the user's page operations in the previous time.

[0041] Here, data acquisition (e.g., review content and real-time user information) can be achieved by connecting to a unified reporting component via a data collection terminal (e.g., client SDK). Events such as page jumps, clicks, and exposure dwell times are reported to the execution entity in a unified format. For data transmission, Kafka can be used to receive high-concurrency real-time data streams, ensuring no data loss. DataBus is used to periodically synchronize batch data such as transaction and logistics data from the business database. In addition, Kafka data streams can be consumed, and all events of the same reviewer can be aggregated to the same processing node using KeyBy(userId). Dynamic features such as "browse-order interval," "page bounce rate," and "current session review tendency" are calculated in real-time within a window to supplement behavioral link information. Furthermore, a time-series database can store raw event sequences, facilitating efficient querying of the original behavioral trajectories of reviewers within a time range. In addition, a graph database (Neo4j) is set up specifically for the e-commerce field, using users, products, and orders as nodes, and browsing, purchasing, and reviews as edges to construct a dynamic relationship graph for mining group associations (e.g., multiple users reviewing the same batch of products).

[0042] Step 202: Based on the above comments and the above real-time user information, generate the first comment risk information using the first multimodal large model for real-time risk prediction.

[0043] In some embodiments, the aforementioned executing entity can generate first review risk information based on the review content and the real-time user information, using a first multimodal large model for real-time risk prediction. The first multimodal large model can be a multimodal large model that performs real-time risk information detection on the review content entered on the review posting page. In practice, the first multimodal large model here is a large model whose real-time detection reaction time meets the target real-time response condition. That is, the first multimodal large model here can be a multimodal large model that achieves efficient real-time detection. The real-time detection reaction time can be the reaction time for the model to perform real-time detection and output the review risk result. The target real-time response condition here can be a large model whose real-time detection reaction time is less than the target reaction time. The target reaction time can be a time used to measure whether the large model can react quickly. The target reaction time can be a pre-set value. The first multimodal large model can be a large model whose input information and / or output can be multimodal content. The multimodal content here can include: text modality content, image modality content, and audio modality content. For example, the first multimodal large model can be a DeepSeek-VL model. First-review risk information can be information that characterizes the degree of risk associated with the review content. Here, the degree of risk associated with the review content can represent the probability that the review content is false, fraudulent, or maliciously reviewed (i.e., content that is not based on the user's own feelings). In practice, first-review risk information can be in numerical form or in the form of tags. When the first-review risk information is in numerical form, the higher the value, the greater the severity of the risk associated with the review content.

[0044] It should be noted that the timing for generating the first review risk information is when the first multimodal large model identifies... As an example, firstly, a prompt message is generated to indicate the first review risk information based on the review content and real-time user information. Then, the generated prompt message is input into the first multimodal large model to obtain the first review risk information.

[0045] Step 203: In response to the aforementioned first review risk information indicating that the review content is risky, adjust the page content of the review posting page according to the aforementioned first review risk information to add risk verification content.

[0046] In some embodiments, in response to the aforementioned first review risk information indicating that the review content poses a risk, the implementing entity may adjust the page content of the review posting page based on the aforementioned first review risk information to add risk verification content. This risk verification content may be a prompt reminding the review user to perform risk verification.

[0047] As an example, in response to the aforementioned risk information in the first review indicating that the review content poses a risk, the risk information and corresponding risk verification prompts are displayed in the risk display area of ​​the review posting page, allowing review users to perform the corresponding verification operations based on the risk verification prompts. For example, the verification operation here could be uploading a physical image of the reviewed object.

[0048] In some optional implementations of certain embodiments, the aforementioned risk verification content is review guidance information. This review guidance information can be information guiding review users to upload content related to the reviewed object in relation to their published review content. In practice, the content related to the reviewed object can be, but is not limited to, at least one of the following: a physical image of the reviewed object, a packaging image of the reviewed object, or a question related to the reviewed object. For example, a question related to the reviewed object could be, "What do you think is the most prominent feature of this product?" For example, the review guidance information could be guidance information guiding review users to upload physical images, or it could be guidance information guiding review users to answer related questions. Here, the answers to related questions can be in the form of multiple-choice questions.

[0049] Optionally, in response to the aforementioned first review risk information indicating that the risk level corresponding to the review content is the first risk level, the aforementioned implementing entity may display the aforementioned review guidance information on the aforementioned review publication page.

[0050] It should be noted that for review guidance information that directs users to upload product photos, videos, packaging images, etc., users can upload their content by clicking the upload control in the page area where the review guidance information is located. This content can be the uploaded product photos, videos, packaging images, etc. For review guidance information that directs users to answer related questions, users can select the answer to the question (i.e., select the answer to the relevant question) in the answer area of ​​the page area where the review guidance information is located, thus uploading their content. This content can be the answer to the relevant question.

[0051] In some optional implementations of certain embodiments, the aforementioned risk verification content is an anomaly prompt pop-up. This anomaly prompt pop-up can be a pop-up indicating that the review content is abnormal and providing a solution to the abnormality. For example, the anomaly prompt pop-up might display the message, "The currently entered review content may be risky; please carefully review it!" The solution displayed in the anomaly prompt pop-up might be, "Please perform the anomaly verification operation according to the anomaly verification operation information."

[0052] Optionally, in response to the aforementioned first review risk information indicating that the risk level is a second risk level, the aforementioned implementing entity may display an anomaly prompt pop-up window on the aforementioned review posting page to provide anomaly notification and anomaly verification operation information. Here, the anomaly verification operation information may be: Please perform SMS verification or answer a verification question about the review object (such as "Please describe the main logo on the packaging box").

[0053] Step 204: In response to completing the content response to the above-mentioned risk verification content, submit the complete comment content entered on the above-mentioned comment posting page.

[0054] In some embodiments, in response to completing a content response to the aforementioned risk verification content, the executing entity may submit the complete review content entered on the review publishing page. The content response may be the review user performing content verification on the risk verification content. The complete review content may be the complete content entered by the review user on the review publishing page. The submission process may involve the review user selecting the review content submission control on the review publishing page to publish the review content to the comment display page corresponding to the review object.

[0055] In some optional implementations of certain embodiments, before step 204, the steps further include: In response to the detection that the aforementioned user has completed the verification operation corresponding to the above-mentioned abnormal verification information, the aforementioned executing entity can determine to complete the content response to the above-mentioned risk verification content. For example, the verification operation here can be one of the following: completing the verification via SMS, or completing the response content to the relevant questions.

[0056] The above-described embodiments of this disclosure have the following beneficial effects: Through the comment content processing methods of some embodiments of this disclosure, risk monitoring can be performed during the real-time input of comment content by comment users, allowing for corresponding page content adjustments and risk verification processing. Specifically, the reason for the delayed impact elimination due to insufficient timely risk verification is that even if risk identification is performed after the comment content is published, the comment content has already been uploaded to the platform for display, causing negative impact and resulting in a delay in impact elimination. Based on this, the comment content processing methods of some embodiments of this disclosure first obtain the comment content input in real-time by the comment user on the comment publishing page and the corresponding real-time user information. Here, by obtaining the input comment content and corresponding real-time user information in real-time on the comment publishing page, a data basis is provided for subsequent analysis of the risk information corresponding to the comment content, enabling accurate and timely determination of the first comment risk information. Then, based on the comment content and the real-time user information, the first comment risk information can be accurately and efficiently generated using a first multimodal large model for real-time risk prediction. Here, by generating initial review risk information, review risks can be identified in real-time as users input their reviews, allowing for subsequent risk mitigation and preventing delayed negative impacts. Next, in response to the initial review risk information indicating a risk in the review content, the review posting page is adjusted to add risk verification content. Here, by displaying the risk verification content on the review posting page, users are visually prompted to mitigate the risk, increasing the authenticity of the review and eliminating potential risks. Finally, in response to the completed risk verification, the complete review content entered on the review posting page is submitted, resulting in a review with significantly reduced risk. In summary, by generating initial review risk information during real-time review input, accurate and efficient real-time risk detection can be performed, eliminating potential risks in the review content. In addition, by adjusting the content on the review posting page and adding risk-related content, content risks can be effectively eliminated in a visual way.

[0057] Further reference Figure 3 The diagram illustrates a flow 300 of another embodiment of the comment content processing method according to this disclosure. This comment content processing method includes the following steps: Step 301: Obtain the review content entered in real time by the review user on the review posting page and the real-time user information corresponding to the aforementioned review user.

[0058] Step 302: Based on the above comments and the above real-time user information, generate the first comment risk information using the first multimodal large model for real-time risk prediction.

[0059] Step 303: In response to the aforementioned first review risk information indicating that the review content is risky, adjust the page content of the review posting page according to the aforementioned first review risk information to add risk verification content.

[0060] Step 304: In response to completing the content response to the above-mentioned risk verification content, submit the complete comment content entered on the above-mentioned comment posting page.

[0061] In some embodiments, the specific implementation of steps 301-304 and the resulting technical effects can be found in [reference needed]. Figure 2 Steps 201-204 in the corresponding embodiments will not be repeated here.

[0062] Step 305: Obtain the offline user information corresponding to the above-mentioned review users.

[0063] In some embodiments, the executing entity (e.g. Figure 1 The electronic device 101 shown can acquire offline user information corresponding to the aforementioned reviewers. This offline user information can be user information stored offline. In practice, the user content corresponding to offline user information is more extensive than that corresponding to real-time user information. Offline user information includes not only real-time user information but also historical user information corresponding to the reviewers. That is, offline user information provides more comprehensive user content than real-time user information.

[0064] Step 306: Based on the aforementioned offline user information and the aforementioned complete review content, generate second review risk information using the second multimodal large model for risk prediction.

[0065] In some embodiments, the aforementioned executing entity can generate second review risk information based on the aforementioned offline user information and the aforementioned complete review content, using a second multimodal large-scale model for risk prediction. Here, the second multimodal large-scale model can be the same as the first multimodal large-scale model, or it can be a different large-scale model. When the two are different large-scale models, the real-time detection response time corresponding to the first multimodal large-scale model should be shorter than the real-time detection response time corresponding to the second multimodal large-scale model. That is, the second multimodal large-scale model can be a large-scale model that does not require a fast real-time response but requires sufficient accuracy for risk detection. For example, when the two are different, the second multimodal large-scale model can be a large-scale model based on the Transformer architecture. The second review risk information can also characterize whether the complete review content carries risk. In practice, the second review risk information can be in numerical form or in the form of tags.

[0066] As an example, firstly, a generation prompt message for review risk information is generated based on offline user information and complete review content. Then, the generated prompt message is input into the second multimodal large model to obtain the second review risk information.

[0067] Step 307: Based on the aforementioned second review risk information, process the complete review content.

[0068] In some embodiments, the aforementioned executing entity may process the complete review content based on the second review risk information. The currently complete review content may be the complete review comment after the review user has clicked the publish control.

[0069] As an example, in response to determining that the risk value corresponding to the second review risk information falls within the first value range, the complete review content is deleted. In response to determining that the risk value corresponding to the second review risk information falls within the second value range, the complete review content is returned for the review user to readjust. The value corresponding to the first value range is higher than the value corresponding to the second value range.

[0070] As an example, in response to the determination that the risk information in the second review indicates no risk, the content display weight corresponding to the aforementioned review is increased so that it is displayed on subsequent review content display pages. Content display weight can be the relative importance of the content on the display page. The higher the corresponding content display weight, the higher the page content will appear on the content display page.

[0071] In some optional implementations of certain embodiments, the aforementioned execution entity may perform content processing on the aforementioned complete review content based on the aforementioned second review risk information, including the following steps: The first step is to hide the complete review content in response to the aforementioned second review risk information indicating that the corresponding risk label is the first label. The first label can be a label indicating that the complete review content contains suspected risks. For example, the first label could be "suspected risk".

[0072] The second step involves re-reviewing the complete review content in response to the aforementioned risk information indicating that the risk label is a second label. The second label can indicate that the complete review content carries a significant risk. For example, the second label could be "significant risk." This re-review process may involve sending the review content to a risk-assessing terminal for further risk review.

[0073] In some optional implementations of certain embodiments, after step 307, the steps further include: The first step is to store the complete review content, corresponding user information, and review risk results in the review data repository. The user information here can include offline user information and real-time user information. The review risk result can be the final review result of the complete review content. The review risk result can be one of the following: no risk, first tag, or second tag. The review data repository can be a database that stores review data.

[0074] As an example, the aforementioned implementing entity can correlate complete review content, corresponding user information, and review risk results to obtain review data. This review data is then stored in a review data repository.

[0075] The second step involves, in response to the determination that the first review dataset in the aforementioned review data repository meets the target condition, training the first multimodal large model and the second multimodal large model based on the first review dataset, resulting in the first trained multimodal large model and the second trained multimodal large model. The first review dataset can be any review data stored in the review data repository. The target condition can be that the amount of data corresponding to the first review dataset reaches a target quantity or that the model training time is reached.

[0076] As an example, the aforementioned execution entity can use the first review dataset and conventional large model training methods to train the first multimodal large model and the second multimodal large model to obtain the first trained multimodal large model and the second trained multimodal large model.

[0077] The third step is to replace the deployed first multimodal large model with the aforementioned first trained multimodal large model, and replace the deployed second multimodal large model with the aforementioned second trained multimodal large model.

[0078] In some optional implementations of certain embodiments, the first multimodal large model and / or the second multimodal large model described above are trained through the following steps: The first step is to obtain the second review dataset. This second review data includes: multimodal review content, user behavior information, and review risk results. The multimodal review content can include text-based content and image-based content. User behavior information can be... The second step involves standardizing the information format of the review content set in the second review dataset and performing time-series encoding on the user behavior information set to obtain the review processing dataset.

[0079] Here, the user behavior information set is time-series encoded to preserve the time dimension information.

[0080] The third step involves training the first initial multimodal large model and / or the second initial multimodal large model based on the aforementioned review processing dataset. This training aims to learn multimodal content alignment knowledge and the correlation between review content and user behavior, resulting in the first preliminary multimodal large model and / or the second preliminary multimodal large model. In practice, multimodal content alignment knowledge may include image-text alignment knowledge. The correlation between review content and user behavior can be a mapping relationship. The first and second preliminary multimodal large models can be large models that have undergone preliminary training.

[0081] The fourth step involves adding each review processing data point from the aforementioned review processing dataset to the chain of thought prompt template used for fake review detection, resulting in a chain of thought prompt information set. The chain of thought prompt template in the large language model is a prompting engineering technique that guides the model to simulate the step-by-step reasoning process of humans, showcasing intermediate thinking steps to improve the ability to solve complex problems, output accuracy, and interpretability. The core logic is to transform "directly asking for the answer" into "requiring clear thinking step by step before providing the answer," turning the model's reasoning process from a "black box" to a "white box."

[0082] Fifth, based on the aforementioned set of thought chain prompts, the first preliminary multimodal large model and / or the second preliminary multimodal large model are retrained to obtain the first multimodal large model and / or the second multimodal large model. Details will not be elaborated further.

[0083] from Figure 3 It can be seen from this that, with Figure 2 Compared to the description of some corresponding embodiments, Figure 3In some corresponding embodiments, the process 300 of the comment content processing method involves using offline user information and complete comment information, along with a second multimodal big data model, to more accurately and comprehensively detect risks in the comment content after it has been published. This allows for more effective processing of the comment content. When risks are detected, content processing can be used to eliminate risks in a timely manner.

[0084] Further reference Figure 4 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a comment content processing apparatus, which are similar to... Figure 2 Corresponding to the method embodiments shown, this review content processing device can be specifically applied to various electronic devices.

[0085] like Figure 4 As shown, a review content processing device 400 includes: an acquisition unit 401, a generation unit 402, an adjustment unit 403, and a submission unit 404. The acquisition unit 401 is configured to acquire review content entered in real-time by a review user on a review posting page and corresponding real-time user information. The generation unit 402 is configured to generate first review risk information based on the review content and the real-time user information, using a first multimodal large model for real-time risk prediction. The adjustment unit 403 is configured to, in response to the first review risk information indicating that the review content poses a risk, adjust the page content of the review posting page to add risk verification content. The submission unit 404 is configured to, in response to completing a response to the risk verification content, submit the complete review content entered on the review posting page.

[0086] In some optional implementations of some embodiments, the risk verification content is review guidance information; and the adjustment unit 403 can be further configured to: in response to the first review risk information representing the risk level corresponding to the review content as a first risk level, display the review guidance information on the review publishing page.

[0087] In some optional implementations of some embodiments, the risk verification content is an abnormal prompt pop-up window; and the adjustment unit 403 can be further configured to: in response to the first review risk information indicating that the risk level is a second risk level, pop up an abnormal prompt pop-up window on the review publishing page to provide abnormal prompts and abnormal verification operation information.

[0088] In some optional implementations of certain embodiments, the apparatus 400 further includes a determining unit (not shown in the figure). This determining unit can be configured to: in response to detecting that the user has completed the verification operation corresponding to the abnormal verification operation information, determine that a response to the risk verification content has been completed.

[0089] In some optional implementations of certain embodiments, the apparatus 500 further includes: an information acquisition unit, an information generation unit, and a processing unit (not shown in the figure). The information acquisition unit can be configured to acquire offline user information corresponding to the reviewing user. The information generation unit can be configured to generate second review risk information based on the offline user information and the complete review content, using a second multimodal large model for risk prediction. The processing unit can be configured to perform content processing on the complete review content based on the second review risk information.

[0090] In some optional implementations of some embodiments, the processing unit may be configured to: hide the complete review content in response to the second review risk information indicating that the risk tag corresponding to the complete review content is the first tag; and review the complete review content again in response to the second review risk information indicating that the risk tag is the second tag.

[0091] In some optional implementations of certain embodiments, the apparatus 500 further includes a storage unit, a training unit, and a replacement unit (not shown in the figure). The storage unit can be configured to store the complete review content, corresponding user information, and review risk results in a review data repository. The training unit can be configured to, in response to determining that the first review dataset in the review data repository meets the target conditions, train the first multimodal large model and the second multimodal large model based on the first review dataset to obtain a first trained multimodal large model and a second trained multimodal large model. The replacement unit can be configured to replace the deployed first multimodal large model with the first trained multimodal large model, and replace the deployed second multimodal large model with the second trained multimodal large model.

[0092] It is understandable that the units described in the review content processing device 400 are related to the reference. Figure 2 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the comment content processing device 400 and the units contained therein, and will not be repeated here.

[0093] The following is for reference. Figure 5It illustrates electronic devices suitable for implementing some embodiments of this disclosure (e.g., Figure 1 A schematic diagram of the structure of electronic device 101)500. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0094] like Figure 5 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory 502 or a program loaded from a storage device 508 into a random access memory 503. The random access memory 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, the read-only memory 502, and the random access memory 503 are interconnected via a bus 504. An input / output interface 505 is also connected to the bus 504.

[0095] Typically, the following devices can be connected to the input / output interface 505: input devices 506 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 508 including, for example, magnetic tape, hard disk, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0096] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a read-only memory 502. When the computer program is executed by the processing device 501, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0097] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0098] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0099] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: acquire the review content entered in real-time by the review user on the review posting page and the corresponding real-time user information; generate first review risk information based on the review content and the real-time user information using a first multimodal large model for real-time risk prediction; in response to the first review risk information indicating that the review content is risky, adjust the page content of the review posting page to add risk verification content based on the first review risk information; and submit the complete review content entered on the review posting page in response to completing the content response to the risk verification content.

[0100] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0101] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0102] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an acquisition unit, a generation unit, an adjustment unit, and a submission unit. The names of these units do not necessarily limit the unit itself; for example, the acquisition unit may also be described as "a unit that acquires the real-time input of review content by review users on the review posting page and the corresponding real-time user information of the review users."

[0103] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0104] Some embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements any of the above-described comment content processing methods.

[0105] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for processing review content, comprising: Obtain the review content entered in real time by the review user on the review posting page and the real-time user information corresponding to the review user; Based on the review content and the real-time user information, a first review risk information is generated using a first multimodal large model for real-time risk prediction; In response to the first review risk information indicating that the review content is risky, the page content of the review publishing page is adjusted according to the first review risk information to add risk verification content; In response to the completion of the content response to the risk verification content, the complete review content entered on the review posting page is submitted for processing.

2. The method according to claim 1, wherein, The risk verification content is a commentary and guidance information; as well as The step of adjusting the page content of the review publishing page based on the first review risk information includes: In response to the first review risk information indicating that the risk level corresponding to the review content is the first risk level, the review guidance information is displayed on the review publishing page.

3. The method according to claim 1, wherein, The risk verification content is an anomaly alert pop-up window; as well as The step of adjusting the page content of the review publishing page based on the first review risk information includes: In response to the first review risk information indicating that the risk level is the second risk level, an abnormal prompt pop-up window appears on the review publishing page to provide abnormal prompts and abnormal verification operation information.

4. The method according to claim 3, wherein, Before submitting the complete review content entered on the review posting page in response to the completion of the content response to the risk verification, the method further includes: In response to detecting that the reviewing user has completed the verification operation corresponding to the abnormal verification operation information, it is determined that a content response to the risk verification content has been completed.

5. The method according to claim 1, wherein, The method further includes: Obtain the offline user information corresponding to the reviewed user; Based on the offline user information and the complete review content, a second review risk information is generated using a second multimodal large model for risk prediction; Based on the second review risk information, the complete review content is processed.

6. The method according to claim 5, wherein, The step of processing the complete review content based on the second review risk information includes: In response to the second review risk information indicating that the risk tag corresponding to the complete review content is the first tag, the complete review content is hidden; In response to the second review risk information indicating that the risk label is the second label, the complete review content is reviewed again.

7. The method according to claim 5, wherein, The method further includes: The complete review content, corresponding user information, and review risk results are stored in the review data repository. In response to determining that the first review dataset in the review data repository meets the target conditions, the first multimodal large model and the second multimodal large model are trained based on the first review dataset to obtain the first trained multimodal large model and the second trained multimodal large model. Replace the deployed first multimodal large model with the first trained multimodal large model, and replace the deployed second multimodal large model with the second trained multimodal large model.

8. The method according to claim 5, wherein, The first multimodal large model and / or the second multimodal large model are trained through the following steps: Obtain the second review dataset, which includes: review content in multimodal form, user behavior information, and review risk results; The review content set in the second review dataset is standardized in information format and the user behavior information set is time-series encoded to obtain the review processing dataset; Based on the review processing dataset, the first initial multimodal large model and / or the second initial multimodal large model are trained to learn multimodal content alignment knowledge and the correlation between review content and user behavior, thereby obtaining the first preliminary multimodal large model and / or the second preliminary multimodal large model. Each review processing data in the review processing dataset is added to the mind chain prompt template used for fake review detection to obtain a mind chain prompt information set; Based on the thought chain prompt information set, the first preliminary multimodal large model and / or the second preliminary multimodal large model are retrained to obtain the first multimodal large model and / or the second multimodal large model.

9. A device for processing review content, comprising: The acquisition unit is configured to acquire the review content that the review user enters in real time on the review posting page and the real-time user information corresponding to the review user; The generation unit is configured to generate first review risk information based on the review content and the real-time user information, using a first multimodal large model for real-time risk prediction. The adjustment unit is configured to respond to the first review risk information indicating that the review content has a risk, and adjust the page content of the review publishing page according to the first review risk information to add risk verification content; The submission unit is configured to submit the complete review content entered on the review publishing page in response to a completed response to the risk verification content.

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

11. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-8.

12. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-8.

Images