144 results about "Content filtering" patented technology

The invention provides a large model dynamic protection method and system based on a zero-trust architecture. The method comprises the steps that a security proxy gateway receives an access request; authenticating an initiating main body of the access request, collecting context information and transmitting the context information to a strategy decision point; the strategy decision point calculates a trust score in real time based on a dynamic trust evaluation model and performs real-time evaluation in combination with an access control strategy to generate a dynamic authorization judgment result; if the access is allowed, forwarding the access request to a large language model server, and performing input security filtering; the large language model server generates response content and performs output security filtering; and returning the final response subjected to the output security filtering to the initiating main body through the security proxy gateway. According to the method, a multi-layer protection framework is constructed, a dynamic trust evaluation model is introduced, and a content filtering layer is deployed, so that continuous permission verification, risk adaptive control and full-link content security protection are realized, and the service security of a large model is effectively guaranteed.

The invention discloses a large-model sensitive large-content filtering method in a learning scene, and the method comprises the steps: extracting an image-side visual coding feature and a text-side semantic coding feature of a learning problem from the learning problem inputted by a user through employing a neural network model based on deep learning; and multi-modal feature fusion is further carried out on the characters and the pictures to form more comprehensive and accurate question joint representation, and the joint representation can fully mine the deep relation between the characters and the pictures, so that sensitive or improper contents in complex input are effectively identified, and automatic and intelligent filtering of the compound learning questions is realized. The problems that a traditional single-mode filtering scheme is narrow in coverage, low in detection accuracy, low in manual auditing efficiency and the like are solved, the recognition capacity of a platform for sensitive content is improved, and then a firm and reliable safety guarantee is provided for an intelligent education platform.

A genetic algorithm is implemented to generate prompts that evade content filters of generative artificial intelligence (AI) systems. The genetic algorithm applies grammar operations to mutate candidate prompts, communicates the candidate prompts to generative AI systems, and selects candidate prompts that successfully evade content filters according to corresponding responses. A disambiguation model that corrects grammar in prompts is tested on the selected prompts to determine if grammar is properly corrected. Once tested, the disambiguation model is deployed in an ensemble with a classifier that outputs verdicts for prompts with grammar corrected by the disambiguation model.

This disclosure provides methods, components, devices, and systems for network security and for providing threat notifications in association with threat detections. Some aspects relate to notifications for network security threats detected by a gateway of a network. The gateway may monitor network traffic on the gateway and detect threats using packet inspection, traffic logging, content filtering, and other tools. Upon identification, determination, or detection of a potential threat, a security response mechanism may be implemented that may include providing a notification to a device on the network, to a user on the network, or both. The notification may include one or more of a visual indication, an audible indication, a tactile indication, and an electronic message. The gateway may output the notification to a group of devices based on the predefined preference of each device. The notification may include mitigation procedures usable to mitigate damage associated with the security threat.

The invention discloses an intelligent customer opinion processing system based on multiple agents. A receiving module receives complaint information of a user and standardizes the complaint information. And the task distribution module realizes splitting of customer complaint tasks and intelligent agent adaptation for standardized customer complaint data through a multi-intelligent agent iterative learning framework. And the subtask execution module executes the received customer complaint task. And the feedback optimization module collects full-link indexes in real time, detects abnormal events by adopting a sliding time window, and optimizes the construction of a decision tree in the sub-task execution module. And the content filtering module is used for carrying out post-processing on output texts of the consultation agent and the risk identification agent by using a text filtering and optimizing framework based on an off-line large model. And the result storage module adopts a multi-level storage strategy to record a full-link operation log. According to the method, the problems of low efficiency, insufficient accuracy, weak dynamic adaptive capacity and the like of a traditional customer complaint processing method can be effectively solved.

The present disclosure distributes processing capabilities throughout different nodes in a wireless network. Methods and apparatus consistent with the present disclosure increase the efficiency of communications in a wireless network because they help minimize the need to forward communications to other nodes in the network. Apparatus and methods consistent with the present disclosure perform a function of elastic content filtering because rating information may be stored in different memories of different mesh nodes according to rules or profiles associated with a wireless mesh network as responses to requests are sent back along a route in a wireless mesh network in a manner that may not increase an amount of network traffic. When, however, network traffic dips below a threshold level, additional messages may be sent to certain mesh nodes that update rating information stored at those certain mesh nodes.

Methods and apparatus for filtering panoramic video signals using visual fixation are disclosed. A panoramic video signal organized into indexed frames is displayed and durations of visual fixations of a viewer of the display are measured by means of detecting the viewer's gaze positions and identifying clusters of adjacent gaze positions. For each visual fixation attaining a prescribed duration threshold, a segment of the panoramic video signal corresponding to a respective view region of prescribed shape and dimensions is extracted to form a filtered video signal. Contents of successive frames of the panoramic signal are cyclically supplied to a bank of content-filtering units operating concurrently to produce individual content-filtered frames which are concatenated for transmission to respective destinations.

Context-aware content filtering adapted for a knowledge domain is provided. In certain examples, a classification confidence score by a classifier indicates a level of confidence that a prompt from a user is associated with the knowledge domain. The classification confidence score is compared with a threshold. When the score is below a threshold, a violation notice is provided to the user without submitting the prompt to a generative artificial intelligence (GAI) model. When the classification confidence score is above the threshold, the prompt is further processed to determine, according to rules, whether the prompt should be submitted to the GAI model. In various examples, the rules are applied to contextual information, safety score information, and intent information derived from the prompt.

The present disclosure is directed to distributing processing capabilities throughout different nodes in a wireless mesh network. Methods and apparatus consistent with the present disclosure increase the efficiency of communications in a wireless mesh network because they help minimize the need to forward communications to other nodes in the wireless mesh network such that an evaluation can be performed. Apparatus and methods consistent with the present disclosure may distribute ratings or verdicts associated with previous requests to access data to different nodes in a wireless mesh network without generating additional wireless communications through the wireless mesh network. Apparatus and methods consistent with the present disclosure distribute content ratings to different nodes in a wireless network such that different wireless nodes may block redundant requests to undesired content without increasing messaging traffic.

The invention belongs to the technical field of computer information security, particularly discloses a content security protection method and device based on semantic consistency, and aims to solve the problem that high-level cue word injection attacks are difficult to effectively recognize in the prior art. Carrying out real-time interception on dominant illegal texts through a content filtering module; quantizing generation rationality differences of prompt words between attack and normal language models by using a word vector confusion degree calculation module; evaluating the logic coherence of the sentence structure of the prompt word through a statement semantic consistency judgment module; integrating the multi-dimensional feature data and carrying out risk classification by adopting a machine learning algorithm; and routing the suspected attack request to a value fine tuning model for processing according to a judgment result. According to the technical scheme, high-precision recognition and response to complex cue word injection attacks are achieved, and the content safety protection capacity and compliance guarantee level of a large model in an interaction scene are remarkably improved.

The invention discloses a multi-mode private illegal content retrieval method and system based on semantic granularity perception, and belongs to the technical field of image-text multi-mode content retrieval. In order to solve the technical problem that private and illegal contents are difficult to effectively retrieve through explicit text description, the method comprises the following steps of: constructing an image identifier consisting of clustering, matrix decomposition, quantization and unique coding, and realizing generative learning of an image and the image identifier by utilizing a multi-modal large model; and constructing a soft label based on the similarity between the image and the text representation, and outputting a corresponding image identifier through text-driven generative retrieval learning. According to the method, the illegal image content with semantic matching can be retrieved from the private text description, and the method is suitable for scenes such as illegal content examination, risk identification and content filtering.

A source content processor receives content from a crawler and calls a text mining engine. The text mining engine mines the content and provides metadata about the content. The source content processor applies a source content filtering rule to the content utilizing the metadata from the text mining engine. The source content filtering rule is previously built based on at least one of a named entity, a category, or a sentiment. The source content processor determines whether to persist the content according to a result from applying the source content filtering rule to the content and either stores the content in a data store or deletes the contents from the data ingestion pipeline such that the content is not persisted anywhere. Embodiments disclosed herein can significantly reduce the amount of irrelevant content through the data ingestion pipeline, prior to data persistence.

The invention discloses a text embedding optimization-based sensitive content filtering and defending method for a text generation graph model, and relates to the technical field of large model security. The method comprises the steps of preparing a positive and negative sample data set and a common data set; parameters of a text encoder are adjusted, and a teacher-student model is used for training; stabilizing a text encoder by using a common data set; and designing a comprehensive loss function, performing training optimization in combination with the plurality of data sets to generate an adjusted student model, and generating a final security image through text embedding. According to the method, when text input containing sensitive vocabularies is processed through fine tuning of a text encoder, text embedding of the sensitive vocabularies is converted into text embedding of corresponding benign vocabularies, and the benign text embedding is used for guiding generation of a security image; on the premise of not changing the model generation capability, the generation of images containing sensitive pornography or other images protected by copyright by the text graph model is avoided, and the effective filtering, conversion and defense of sensitive contents are realized.

A dedicated interactive device designed for children comprises an embedded electronic module uniquely paired to a single child via a secure identifier. The device enables conversational and educational interactions, applying a multi-layer artificial intelligence filtering system to block sensitive content. Interaction data are encrypted and stored locally when offline, with secure synchronization to the cloud upon reconnection. A cloud-based parental dashboard allows management of one or multiple child profiles, language and voice preferences, review of time-stamped interaction logs, and configuration of usage controls. The system features privacy-by-design principles, including data minimization, strong encryption, authentication, and regulatory compliance. Optional components, such as cameras or displays, are integrated under the same safety and privacy framework. The platform unifies secure device-child pairing, robust AI filtering, comprehensive parental control, and enhanced protection of children's data in a single system.

Techniques for filtering out undesirable generative machine learned model (or LLM) output are discussed herein. A system may receive a subset of an LLM output. That is, the system may stream the LLM output to a user device by receiving one or more tokens from the LLM and outputting such token(s) to a user device. However, prior to outputting the token(s) to the user device, the system may determine whether the token(s) include undesirable content that is to be blocked. The system may use synchronous blocking components (e.g., blocks the undesirable token(s) before such token(s) get output to the user device) and / or asynchronous blocking components (e.g., blocks the undesirable token(s) after the token(s) have been output to the user device) to filter out undesirable content. The synchronous and / or asynchronous blocking components may be designed to block one or more undesirable topics such as hateful speech, profanity, bias, toxicity, factualness, etc.

The invention discloses a method and device for pushing recruitment platform content based on user fatigue, and the method comprises the steps: building a historical behavior baseline for the pushing content of different business types of a recruitment platform for a target user, and enabling the historical behavior baseline to be obtained through the rolling calculation of the historical behavior data of the target user based on a time decay algorithm; for any service type, calculating a comprehensive fatigue score of the target user to the push content of the service type based on the historical behavior baseline and the interactive behavior data of the target user to the push content of the service type; the comprehensive fatigue degree score is compared with a threshold value interval, a pushing strategy corresponding to the interval to which the comprehensive fatigue degree score belongs is executed according to the comparison result, and the pushing strategy comprises normal pushing, frequency reduction pushing, content filtering, pushing channel switching, content transformation and / or silence period entering. According to the application, accurate pushing of the content published by the recruitment platform can be realized, the waste of pushing resources is reduced, and the user satisfaction is improved.

This application describes methods, systems, and apparatus for allowing a user to easily determine whether the user and / or other user(s) have viewed content. For example, a program guide may display one or more graphics associated with the content (e.g., a movie poster), and in response to determining whether the user has viewed the content, the program guide may include an indication of whether the user has viewed the content. The program guide may include indications associated with multiple users (e.g., different colors, icons, graphics, or text). The program guide may filter the content based on whether the user has viewed the content (e.g., only display content the user has not viewed). A computing device storing the content may manage recording of the content based on whether the user has viewed the content (e.g., delete the recording if the user has viewed the content). A viewing history of the user may be maintained.

The invention provides a SaaS-based multi-party system data pushing method and system and a storage medium, and the method comprises the steps: obtaining a differentiated pushing rule pre-configured by a target tenant, the differentiated pushing rule comprising a field selection rule for a target interface, a data filtering rule and a data format conversion rule; obtaining original data to be sent to a target tenant, and performing field screening on the original data based on a field selection rule to obtain first intermediate data; performing content filtering on the first intermediate data based on a data filtering rule to obtain second intermediate data; performing format conversion on the second intermediate data based on a data format conversion rule to obtain to-be-transmitted data meeting the customized format requirement of the target tenant; and sending the to-be-transmitted data to a receiving system corresponding to the target tenant. According to the invention, multi-party system data pushing can be carried out safely and efficiently at low cost.

The application relates to the technical field of artificial intelligence, in particular to a multi-modal human-computer interaction content filtering method, device and medium. The method comprises the following steps: collecting multi-modal interaction data; obtaining a current application scene, calculating single invalidity scores of various data, obtaining joint invalidity scores in combination with multi-modal weights corresponding to the scene, and judging whether the joint invalidity scores are lower than a preset threshold. If yes, scene-specific violation judgment weights are called, multi-modal violation features are extracted from the data, and current fusion features are obtained by combining the weights; the latest N (N>1) historical fusion features are obtained, the current fusion features are combined to generate modified fusion features, the modified fusion features are input into a pre-trained deep learning model, and a risk confidence is obtained. When the risk confidence exceeds a preset threshold, an interaction instruction transmission link is immediately cut off, and multi-modal data is encrypted and stored as evidence, so that accurate filtering and high-risk prevention and control of multi-modal interaction content are realized. The application can effectively improve the judgment accuracy.

The invention discloses an optimization method and system for achieving software function extension based on a DLL and computing device.A DLL module is integrated in target software of a client in a hot plug mode, and the method comprises the steps that in response to a request for inputting external data through a data input interface of the target software, the external data are captured through a first interceptor; verifying the external data to obtain a verification result, and returning the verification result to the target software, so that the target software processes the external data according to the verification result; in response to a request for outputting to-be-output content through a data output interface of the target software, capturing the to-be-output content through a second interceptor; and performing content filtering and security detection on the to-be-output content to obtain target output content, and returning the target output content to the target software so as to be displayed at the client. According to the method, the plug-in can be flexibly and rapidly developed and deployed under the condition that software source codes are not modified, and then flexible expansion and rapid iteration of software functions are achieved through the plug-in.

To improve accuracy of content filtering.SOLUTION: An information processing system 1 includes a first extraction unit that extracts first BCT component information related to a component of a first behavior change technique (BCT) contained in a first content determined to be a filtering target, a second extraction unit that extracts second BCT component information related to a component of a second behavior change technique (BCT) contained in a second content not determined to be a filtering target, and a generation unit that estimates reference information that serves as a criterion for determining whether a content is a filtering target on the basis of the first BCT component information and the second BCT component information, and generates a model based on the reference information.SELECTED DRAWING: Figure 2B

The invention discloses a video analysis task unloading method and device based on mixed granularity content filtering under edge computing and a storage medium, and the method comprises the steps: 1, dividing each video frame image into a plurality of sub-blocks, extracting a candidate target region of each sub-block, dividing each sub-block of each frame of image into a target sparse block and a target dense block; step 2, for the target sparse blocks obtained in the step 1, compactly arranging all candidate target areas of each target sparse block on a plurality of virtual canvases with fixed sizes to form canvas blocks, and forming a block to be unloaded by each canvas block and a target dense block; step 3, generating an unloading scheduling strategy by using a deep reinforcement learning strategy network trained based on a near-end strategy optimization algorithm, and distributing the blocks to be unloaded to each edge node in parallel according to the unloading scheduling strategy; the equipment and the storage medium are used for implementing the method. According to the method, the problems of calculation redundancy and resource bottleneck of existing edge video analysis are solved.

According to the embodiments of the present disclosure, provided are a method and apparatus for content recommendation, and a device, a medium and a program product. The method comprises: using a plurality of content filtering strategies to determine a plurality of groups of candidate recommended content from a content library, wherein the plurality of content filtering strategies are respectively based on different content ranking criteria; using a trained machine learning model to determine a recommendation score of each piece of candidate recommended content in the plurality of groups of candidate recommended content with respect to a target user on the basis of user reference information of the target user and content reference information of the plurality of groups of candidate recommended content; and on the basis of the recommendation score corresponding to each piece of candidate recommended content in the plurality of groups of candidate recommended content, determining a group of target recommended content from among the plurality of groups of candidate recommended content, so as to provide the group of target recommended content to the target user. Thus, the accuracy and efficiency of content recommendation can be improved.

The invention relates to a vehicle sensitive word instruction rejection method and device.The method comprises the steps that text instruction data of a target user for a current vehicle are collected, keyword matching is conducted on the text instruction data through a sensitive word rule engine, and whether sensitive words exist in the text instruction data or not is judged; if the target sensitive word exists, generating a sensitive word rejection instruction to perform sensitive word instruction reminding on the target user; otherwise, inputting the text instruction data into the sensitive word rejection model, outputting a semantic analysis result to judge whether the text instruction data is a target invalid word or not, and outputting an invalid word rejection instruction to perform invalid instruction reminding on the target user under the condition that the text instruction data is the target invalid word. According to the method, the user input is subjected to content filtering and validity judgment so as to ensure the safety and processing efficiency of the subsequent service, and through keyword filtering and semantic analysis, the recognition service can quickly and accurately recognize the input which needs to be rejected, and resource waste is avoided.

Methods and a system for personalizing self-checkout terminals based on customer loyalty information includes receiving customer preferences during registration, generating themed personas with specific color schemes, graphics, and animations relevant to the customer's registered interests, and dynamically applying personalized interface themes when customers are identified at self-checkout terminals. The system maintains cached personas tagged with attributes like interests, religious preferences, and dietary restrictions, enabling sub-second personalization by matching customer preferences to pre-generated themes. The system avoids showing inappropriate content by filtering themes based on offensive content, collects user feedback on displayed themes through a rating mechanism, and employs time-delay factors to prevent theme repetition. New personas and corresponding themes are periodically generated to maintain customer engagement while keeping processing costs manageable.

To easily introduce a generation AI in a company and to perform optimum prompt engineering and filtering of illegal contents.SOLUTION: The system includes a user interface, a content filter, a prompt engineering infrastructure, an LLM engine, and a ConfigCenter. The user interface receives a user operation. The content filter filters content generated based on an operation received by the user interface. The prompt engineering infrastructure optimizes the prompts based on the content filtered by the content filter. The LLM engine generates an output based on the optimized prompt. ConfigCenter manages the configuration of the user interface, content filters, prompt engineering infrastructure, and LLM engine elements.SELECTED DRAWING: Figure 1

The invention belongs to the field of artificial intelligence recognition, and particularly relates to a method and application capable of recognizing credit investigation content, which can accurately and quickly recognize the credit investigation content. Through a series of technical means including the steps of training and application of a deep learning model, preprocessing and optimization of credit investigation pictures, content filtering, feature enhancement and the like, a paper-edition credit investigation material can be converted into an electronic edition, the content of the electronic edition is accurately recognized and audited, and the credit investigation efficiency is improved. Therefore, the accuracy and efficiency of credit investigation content identification are greatly improved, and a more convenient and accurate method is provided for personal credit evaluation.

The application discloses a private protocol transmission security management system and method based on a one-way communication device, and relates to the technical field of network security. The method comprises the following steps: extracting the communication characteristics of a private protocol and performing normalization processing, and constructing a private protocol feature library; deploying a protocol analysis engine on the low-security side of a one-way optical shutter, performing deep detection on the out-domain data stream, identifying the private protocol, and extracting the current session protocol characteristics; matching the extracted characteristics with the feature library, and performing compliance verification and content filtering according to a security policy after successful matching; the verified data is encapsulated into a transmission format allowed to pass through the one-way optical shutter, and is sent to the high-security side through the one-way optical shutter; the received data is reorganized, and integrity and consistency verification is performed on the high-security side, and after the verification is passed, an audit log is recorded and forwarded to a target system; the protocol behavior and the session state are monitored in real time, and a security alarm is triggered and the transmission is blocked for unknown protocols or abnormal behaviors.

The invention discloses an anti-serial connection method and system for network bidding. The method comprises the following steps: implementing information isolation on bidding related parties to ensure that any party cannot know identities and accurate quantity information of other participants, and displaying details, a total demand scale and a current price of an object to a buyer, a seller and a bystander; in the bidding process, a new competitor is dynamically introduced; anonymization control and content filtering are carried out on communication between the buyer, the seller, the bystander and the seller; monitoring a collaborative quotation behavior and triggering disposal; and the key data block chain is subjected to evidence storage, and a visual report is generated for inspection after bidding is finished. The method can effectively prevent series connection and activate competition.

Computer-implemented methods for a topical and contextual content filter for large language model chatbots. Aspects include receiving a user query of a user from a chatbot application. Aspects also include receiving a previous chatbot application conversation associated with the user. Aspects further include determining that the user query of the user from the chatbot application is not a continuation of the previous chatbot application conversation associated with the user. Aspects also include identifying a short list of possible topics using the user query and a topic database. Aspects include determining that the user query of the user from the chatbot application is not associated with the short list of possible topics. Aspects further include transmitting a notification to the chatbot application comprising a rejection of the user query.