Atmosphere inspection method and system

Abstract

The present invention relates to an atmosphere inspection method and device, the method comprising: monitoring a large number of posts and their comment data displayed online, and determining the recall data with violation risks through context combination strategy, comment violation rate strategy and violation comment similarity strategy; placing the recall data on an audit platform to audit the recall data and determine whether the recall data has any violation phenomenon; performing corresponding processing operations on the violation recall data with any violation phenomenon. Through this technical solution, the bad atmosphere problem can be quickly controlled to avoid affecting the atmosphere of the entire community due to a small amount of bad content or users.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an atmosphere inspection method and system. Background Art

[0002] For community products, it is very difficult to deal with a bad community atmosphere in a timely manner. Usually, the atmosphere problem can only be perceived when it has already formed and is particularly serious. When dealing with it, it can only deal with typical data of user feedback point by point. Not only is it difficult to pre-empt or timely discover bad atmosphere problems, but it is also impossible to cover most atmosphere problems and deal with them in a coordinated manner.

[0003] Existing technologies deal with atmosphere problems (large-scale deviations from the topic, and violations of the rules by combining comments and posts) mainly through high-exposure recalls (content with high click-through rates on the site), spontaneous reporting and feedback from users, proactive user behavior in reviewing the recall, and auditors recalling fixed materials. They are highly dependent on user feedback data and typical illegal materials to discover illegal data. Once users do not provide feedback or the feedback is not timely, or the materials are uncommon, they cannot be recalled for processing. Summary of the invention

[0004] In order to overcome the problems existing in the related technology, the present invention provides an atmosphere inspection method and system to recall atmosphere problems before they have a large-scale impact, shorten the time to discover problems, increase the coverage of atmosphere problems, and thus quickly control bad atmosphere problems to avoid affecting the atmosphere of the entire community due to a small amount of bad content or users.

[0005] According to a first aspect of an embodiment of the present invention, there is provided an atmosphere inspection method, the method comprising:

[0006] Monitor the massive amount of posts and their comments displayed online, and determine the recall data with violation risks through context combination strategy, comment violation rate strategy and violation comment similarity strategy;

[0007] Placing the recall data on an audit platform to audit the recall data and determine whether there are any violations in the recall data;

[0008] For the illegal recall data with violations, corresponding processing operations shall be carried out.

[0009] In one embodiment, preferably, the context combination strategy includes:

[0010] Obtain massive amounts of contextual data;

[0011] According to the preset context combination standard and the massive context data, a context-combined BERT model is trained;

[0012] The massive posts and their comment data are detected by using the context-combined BERT model to obtain target data with violation risks;

[0013] Determine a parent feature corresponding to the target data, and concatenate the parent feature with the feature of the target data to obtain a target feature;

[0014] The target features are input into a violation detection model to determine recall data with a risk of violation.

[0015] In one embodiment, preferably, the comment violation rate strategy includes:

[0016] Count the number of all illegal comments under each post, and calculate the ratio of the number of illegal comments to the total number of comments under the post to get the comment violation rate;

[0017] When the exposure value of a post is greater than or equal to a preset exposure value and the total number of comments under the post is greater than or equal to a preset level, and the comment violation rate is greater than or equal to a preset violation rate, a preset number of comment data arranged in front that can be displayed under the post are determined as the recalled data.

[0018] In one embodiment, preferably, the similarity strategy of the illegal comments includes:

[0019] Get the human-reviewed illegal comments under each post;

[0020] Use a text similarity algorithm to score the similarity between human-reviewed illegal comments under each post and the comments that have been displayed online;

[0021] When the similarity score is greater than a preset score value, the target comment corresponding to the similarity score is determined as the recalled data.

[0022] In one embodiment, preferably, the method further comprises:

[0023] On the review platform, all recall data under each post is displayed on a one-post-multiple-comments review page;

[0024] The illegal recall data is sampled in real time to the quality monitoring module so as to be reviewed again by the quality monitoring module to determine the consistency of judgment.

[0025] According to a second aspect of an embodiment of the present invention, there is provided an atmosphere inspection system, the system comprising:

[0026] The risk data recall module is used to monitor the massive amount of posts and their comment data displayed online, and determine the recall data with risks through context combination strategy, comment violation rate strategy and violation comment similarity strategy;

[0027] An atmosphere inspection and audit module is used to place the recall data on an audit platform to audit the recall data and determine whether there are any violations in the recall data;

[0028] The processing module is used to perform corresponding processing operations on the illegal recall data with illegal phenomena.

[0029] In one embodiment, preferably, the context combination strategy includes:

[0030] Obtain massive amounts of contextual data;

[0031] According to the preset context combination standard and the massive context data, a context-combined BERT model is trained;

[0032] The massive posts and their comment data are detected by using the context-combined BERT model to obtain target data with violation risks;

[0033] Determine a parent feature corresponding to the target data, and concatenate the parent feature with the feature of the target data to obtain a target feature;

[0034] The target features are input into a violation detection model to determine recall data with a risk of violation.

[0035] In one embodiment, preferably, the comment violation rate strategy includes:

[0036] Count the number of all illegal comments under each post, and calculate the ratio of the number of illegal comments to the total number of comments under the post to get the comment violation rate;

[0037] When the exposure value of a post is greater than or equal to a preset exposure value and the total number of comments under the post is greater than or equal to a preset level, and the comment violation rate is greater than or equal to a preset violation rate, a preset number of comment data arranged in front that can be displayed under the post are determined as the recalled data.

[0038] In one embodiment, preferably, the similarity strategy of the illegal comments includes:

[0039] Get the human-reviewed illegal comments under each post;

[0040] Use a text similarity algorithm to score the similarity between human-reviewed illegal comments under each post and the comments that have been displayed online;

[0041] When the similarity score is greater than a preset score value, the target comment corresponding to the similarity score is determined as the recalled data.

[0042] In one embodiment, preferably, the system further comprises:

[0043] A display module, used for displaying all recall data under each post on the review platform in a one-post-multiple-comments review page;

[0044] The quality monitoring module is used to sample the illegal recall data in real time to the quality monitoring module so as to re-examine it through the quality monitoring module to determine the consistency of judgment.

[0045] According to a third aspect of an embodiment of the present invention, there is provided a computer-readable storage medium having computer instructions stored thereon, wherein the instructions, when executed by a processor, implement the steps of the method as described in any one of the embodiments of the second aspect.

[0046] The technical solution provided by the embodiments of the present invention may have the following beneficial effects:

[0047] In the embodiment of the present invention, the recall data with violation risks are determined through context combination strategies, comment violation rate strategies and violation comment similarity strategies, so as to conduct further manual review and confirmation, so as to recall the atmosphere problems before they have a large-scale impact, shorten the time to discover problems, increase the coverage of atmosphere problems, and thus quickly control bad atmosphere problems, and avoid affecting the atmosphere of the entire community due to a small amount of bad content or users.

[0048] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0050] Figure 1 The figure is a flow chart of an atmosphere inspection method according to an exemplary embodiment.

[0051] Figure 2 The figure is a flow chart of a context combination strategy according to an exemplary embodiment.

[0052] Figure 3 The following is a flow chart of a comment violation rate strategy according to an exemplary embodiment.

[0053] Figure 4 The figure is a flow chart of a similarity strategy for illegal comments according to an exemplary embodiment.

[0054] Figure 5 is a block diagram of an atmosphere patrol system according to an exemplary embodiment. DETAILED DESCRIPTION

[0055] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0056] Figure 1 The figure is a flow chart of an atmosphere inspection method according to an exemplary embodiment.

[0057] like Figure 1 As shown, according to a first aspect of an embodiment of the present invention, an atmosphere inspection method is provided, comprising:

[0058] Step S101, monitoring a large amount of posts and their comment data displayed online, and determining recall data with violation risks through context combination strategy, comment violation rate strategy and violation comment similarity strategy;

[0059] Step S102, placing the recall data on an audit platform to audit the recall data and determine whether there is any violation in the recall data;

[0060] In one embodiment, preferably, the method further comprises:

[0061] On the review platform, all recall data under each post is displayed in a one-post-multiple-comments review page; through this mode, the overall atmosphere of the comments under the post can be quickly understood and the review processing speed can be improved.

[0062] The illegal recall data is sampled in real time to the quality monitoring module so as to be reviewed again by the quality monitoring module to determine the consistency of judgment.

[0063] The atmosphere inspection manually reviewed data, and real-time sampling flows into the quality monitoring module, which is reviewed by higher-level professionals to determine whether it is consistent. When the consistency rate is low, re-review is performed, and data is discarded without callback.

[0064] Step S103: For the recalled data with violations, corresponding processing operations are performed. For the content, there are self-visible, invisible and other treatments; for the account, if the number of violations reaches a certain level within a unit time, a ban penalty will be imposed.

[0065] like Figure 2 As shown, in one embodiment, preferably, the context combination strategy includes:

[0066] Step S201, obtaining a large amount of context data;

[0067] Step S202, training a context-integrated BERT model according to the preset context-integrated standard and the massive context data;

[0068] Step S203, detecting the massive posts and their comment data through the context-combined BERT model to obtain target data with violation risks;

[0069] Step S204, determining the parent feature corresponding to the target data, and concatenating the parent feature with the feature of the target data to obtain the target feature;

[0070] Step S205: input the target feature into a violation detection model to determine the recalled data with a violation risk.

[0071] The "context combination" strategy is aimed at posts and comments that have no obvious violations, but the content violates the rules in some specific post-comment combination scenarios. By defining the context combination standard and massive context data, a context combination BERT model is trained, and the online data is passed through the model. After the parent feature of the content that hits the model is taken, the parent and child features are combined and passed through the online risk control model. The data recalled by the risk control model is entered into the atmosphere inspection and review system.

[0072] like Figure 3 As shown, in one embodiment, preferably, the comment violation rate strategy includes:

[0073] Step S301, counting the number of all illegal comments under each post, and calculating the ratio of the number of illegal comments to the total number of comments under the post, to obtain the comment violation rate;

[0074] Step S302, when the exposure value of the post is greater than or equal to the preset exposure value and the total number of comments under the post is greater than or equal to the preset level, and the comment violation rate is greater than or equal to the preset violation rate, the preset number of comment data that can be displayed in the front under the post is determined as the recalled data.

[0075] Comment violation rate strategy: Calculate the ratio of the total number of comments that have been reviewed by humans and machines to the total number of comments under the post. When the exposure of a post reaches a certain value and the number of comments under the post reaches a certain level (to prevent a single illegal comment from having a large impact on the violation rate due to too few comments) and the comment violation rate under the post reaches a certain value, the top n data that can be displayed under the post will be recalled to the atmosphere inspection and review system.

[0076] like Figure 4 As shown, in one embodiment, preferably, the similarity strategy of the illegal comments includes:

[0077] Step S401, obtaining the human-reviewed illegal comments under each post;

[0078] Step S402, using a text similarity algorithm to score the similarity between the human-reviewed illegal comments under each post and the comments that have been displayed online;

[0079] Step S403: when the similarity score is greater than a preset score value, the target comment corresponding to the similarity score is determined as the recalled data.

[0080] like Figure 5 As shown, according to a second aspect of an embodiment of the present invention, an atmosphere inspection system is provided, the system comprising:

[0081] The risk data recall module 51 is used to monitor the massive amount of posts and their comment data displayed online, and determine the recall data with risks through context combination strategy, comment violation rate strategy and violation comment similarity strategy;

[0082] The atmosphere inspection and audit module 52 is used to place the recall data on the audit platform to audit the recall data and determine whether there is any violation in the recall data;

[0083] The processing module 53 is used to perform corresponding processing operations on the illegal recall data with illegal phenomena.

[0084] In one embodiment, preferably, the context combination strategy includes:

[0085] Obtain massive amounts of contextual data;

[0086] According to the preset context combination standard and the massive context data, a context-combined BERT model is trained;

[0087] The massive posts and their comment data are detected by using the context-combined BERT model to obtain target data with violation risks;

[0088] Determine a parent feature corresponding to the target data, and concatenate the parent feature with the feature of the target data to obtain a target feature;

[0089] The target features are input into a violation detection model to determine recall data with a risk of violation.

[0090] In one embodiment, preferably, the comment violation rate strategy includes:

[0091] Count the number of all illegal comments under each post, and calculate the ratio of the number of illegal comments to the total number of comments under the post to get the comment violation rate;

[0092] When the exposure value of a post is greater than or equal to a preset exposure value and the total number of comments under the post is greater than or equal to a preset level, and the comment violation rate is greater than or equal to a preset violation rate, a preset number of comment data arranged in front that can be displayed under the post are determined as the recalled data.

[0093] In one embodiment, preferably, the similarity strategy of the illegal comments includes:

[0094] Get the human-reviewed illegal comments under each post;

[0095] Use a text similarity algorithm to score the similarity between human-reviewed illegal comments under each post and the comments that have been displayed online;

[0096] When the similarity score is greater than a preset score value, the target comment corresponding to the similarity score is determined as the recalled data.

[0097] In one embodiment, preferably, the system further comprises:

[0098] A display module, used for displaying all recall data under each post on the review platform in a one-post-multiple-comments review page;

[0099] The quality monitoring module is used to sample the illegal recall data in real time to the quality monitoring module so as to re-examine it through the quality monitoring module to determine the consistency of judgment.

[0100] According to a third aspect of an embodiment of the present invention, there is provided a computer-readable storage medium having computer instructions stored thereon, wherein the instructions, when executed by a processor, implement the steps of the method as described in any one of the embodiments of the second aspect.

[0101] It is further understood that in the present invention, "plurality" refers to two or more than two, and other quantifiers are similar. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. The singular forms "a", "the" and "the" are also intended to include plural forms, unless the context clearly indicates other meanings.

[0102] It is further understood that the terms "first", "second", etc. are used to describe various information, but such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other, and do not indicate a specific order or degree of importance. In fact, the expressions "first", "second", etc. can be used interchangeably. For example, without departing from the scope of the present invention, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information.

[0103] It is further understood that, although the operations are described in a specific order in the accompanying drawings in the embodiments of the present invention, it should not be understood as requiring the operations to be performed in the specific order or serial order shown, or requiring the execution of all the operations shown to obtain the desired results. In certain environments, multitasking and parallel processing may be advantageous.

[0104] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

[0105] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An atmosphere inspection method, characterized in that: The method comprises: Monitor the massive amount of posts and their comments displayed online, and determine the recall data with violation risks through context combination strategy, comment violation rate strategy and violation comment similarity strategy; Placing the recall data on an audit platform to audit the recall data and determine whether there are any violations in the recall data; For the illegal recall data with violations, corresponding processing operations shall be carried out; Wherein, the context combination strategy includes: Obtain massive amounts of contextual data; According to the preset context combination standard and the massive context data, a context-combined BERT model is trained; The massive posts and their comment data are detected by the context-combined BERT model to obtain target data with violation risks; Determine a parent feature corresponding to the target data, and concatenate the parent feature with the feature of the target data to obtain a target feature; Inputting the target features into a violation detection model to determine recalled data with a risk of violation; The review violation rate strategy includes: Count the number of all illegal comments under each post, and calculate the ratio of the number of illegal comments to the total number of comments under the post to get the comment violation rate; When the exposure value of a post is greater than or equal to a preset exposure value and the total number of comments under the post is greater than or equal to a preset level, and the comment violation rate is greater than or equal to a preset violation rate, a preset number of comment data arranged in the front that can be displayed under the post is determined as the recalled data; Similar strategies for illegal comments include: Get the human-reviewed illegal comments under each post; Use a text similarity algorithm to score the similarity between human-reviewed illegal comments under each post and comments that have been displayed online; When the similarity score is greater than a preset score value, the target comment corresponding to the similarity score is determined as the recalled data.

2. The method according to claim 1, characterized in that The method further comprises: On the review platform, all recall data under each post is displayed on a one-post-multiple-comments review page; The illegal recall data is sampled in real time to the quality monitoring module so as to be reviewed again by the quality monitoring module to determine the consistency of judgment.

3. An atmosphere inspection system, characterized in that: The system comprises: The risk data recall module is used to monitor the massive amount of posts and their comment data displayed online, and determine the recall data with risks through context combination strategy, comment violation rate strategy and violation comment similarity strategy; An atmosphere inspection and audit module is used to place the recall data on an audit platform to audit the recall data and determine whether there are any violations in the recall data; A processing module is used to perform corresponding processing operations on the illegal recall data with illegal phenomena; Wherein, the context combination strategy includes: Obtain massive amounts of contextual data; According to the preset context combination standard and the massive context data, a context-combined BERT model is trained; The massive posts and their comment data are detected by the context-combined BERT model to obtain target data with violation risks; Determine a parent feature corresponding to the target data, and concatenate the parent feature with the feature of the target data to obtain a target feature; Inputting the target features into a violation detection model to determine recalled data with a risk of violation; The review violation rate strategy includes: Count the number of all illegal comments under each post, and calculate the ratio of the number of illegal comments to the total number of comments under the post to get the comment violation rate; When the exposure value of a post is greater than or equal to a preset exposure value and the total number of comments under the post is greater than or equal to a preset level, and the comment violation rate is greater than or equal to a preset violation rate, a preset number of comment data arranged in the front that can be displayed under the post is determined as the recalled data; Similar strategies for illegal comments include: Get the human-reviewed illegal comments under each post; Use a text similarity algorithm to score the similarity between human-reviewed illegal comments under each post and comments that have been displayed online; When the similarity score is greater than a preset score value, the target comment corresponding to the similarity score is determined as the recalled data.

4. The system according to claim 3, characterized in that The system further comprises: A display module, used for displaying all recall data under each post on the review platform in a one-post-multiple-comments review page; The quality monitoring module is used to sample the illegal recall data in real time to the quality monitoring module so as to re-examine it through the quality monitoring module to determine the consistency of judgment.

Images