An ai-powered content filtering and recommendation system

Abstract

This invention relates to an artificial intelligence-powered content filtering and recommendation system that enables users to watch films, series, and similar content on online streaming platforms in a customized manner by advanced filtering of unwanted or inappropriate scenes before or during viewing, and / or recommending alternative content to watch.

Description

[0001] DESCRIPTION

[0002] AN AI-POWERED CONTENT FILTERING AND RECOMMENDATION SYSTEM

[0003] Technical Field

[0004] This invention relates to an artificial intelligence-powered content filtering and recommendation system that enables users to watch films, series, and similar content on online streaming platforms in a customized manner by advanced filtering of unwanted or inappropriate scenes that they do not wish to watch before or during the viewing, and / or recommending an alternative content to watch.

[0005] The system utilizes generative artificial intelligence (Al) models in the stages of scene tagging, user command reception, and scene representation generation. Designed primarily to facilitate personalized media experiences, this system offers a multi-layered structure comprising components such as a metadata extractor, a user content viewing history database, content recommendation tools, and a filtering module. All these components interact with an application client. Taking into account diverse cultural and geographic norms, this system learns the user's specific genre and sensitive content preferences for each scene and provides personalized recommendations accordingly. This innovative system significantly enhances the user experience, reduces labeling costs, and offers better content recommendations tailored to the user's individual settings and genre preferences.

[0006] Prior Art

[0007] Current content recommendation and filtering systems available today rely on surfacelevel information such as film summaries, view counts, and genre to calculate similarity between contents. These systems recommend content without taking into account users' preferences for sensitive content details, transcripts, or scene analysis.

[0008] Additionally, film regulatory authorities often establish a general age category for content. However, attitudes towards sensitive content can vary across cultures and geographical locations, making such superficial classification often insufficient. Furthermore, conducting detailed analysis of film scenes based on inappropriate scene and object types is a costly process, and there is a need to automate these processes to a large extent.

[0009] For user-friendly use, it is preferred that the interface of this content filtering and recommendation system is easy to navigate and highly customizable. There are various advancements related to the technical aspects mentioned.

[0010] One of the advancements mentioned in United States patent document US 10255503 describes a system that recommends films by calculating similarity, based on rule-based values, derived from scene pixels. There is no mention of using artificial intelligence or text-based scene descriptions. Instead, advanced similarity calculation can be achieved by utilizing scene descriptions, labels, or detection results obtained through generative Al methods, ultimately generating tailored recommendations for the user.

[0011] Currently available tools / products like Vidangel and Clearplay offer services that allow users to skip sensitive scenes in films based on predetermined categories. However, the process of tagging scenes by categories is done manually, resulting in significant labor costs and increased operational expenses.

[0012] Another document mentioned in the technical field is Turkish patent document TR 2021 / 021336, which describes a system that uses artificial intelligence to detect and block inappropriate content in real-time while watching. This method incurs significant server expenses due to real-time detection on the server side, and it is risky to deploy without post-processing, as it may miss certain instances. Additionally, no specific information is provided about the artificial intelligence methods to be used.

[0013] In the other Turkish patent documents mentioned, TR 2022 / 010163 and TR 2022 / 016108, usage of artificial intelligence for sensitive content filtering is discussed in a similar manner. However, they do not provide detailed information on how the artificial intelligence will be implemented. As mentioned earlier, none of the existing methods mentioned address an automated and intelligent system that learns users' attitudes towards unwanted sensitive scenes by utilizing unwanted sensitive scene information in the scenes and the users' filter usage. Additionally, a common limitation of previous methods is that users are restricted to selecting filters from a fixed set of categories in the interface. However, they do not provide an ergonomic and truly customizable solution as users cannot provide detailed filtering commands beyond predetermined categories using natural language, such as voice, gestures, or text.

[0014] Detailed Description of the Invention

[0015] The invention is an artificial intelligence-powered content filtering and recommendation system enabling users to watch films, series, and similar content on online streaming platforms in a customized manner by advanced filtering of unwanted or inappropriate scenes before or during the viewing and / or recommending alternative content to watch. The system comprises:

[0016] At least one multimedia device that presents the video and audio broadcast to the users by accepting the selection commands of the unwanted scenes and the selected online streaming platform and content given by the users over its interface,

[0017] - A scene detection module that determines the start and end time intervals of scenes in the content to be watched,

[0018] - A preprocessing module that applies audio, text, and image processing techniques to scenes detected by the scene detection module, transforming them into a format suitable for advanced artificial intelligence module,

[0019] At least one advanced artificial intelligence module equipped with one or more generative Al models based on convolutional neural networks or transformer architectures, which utilize scene information such as audio, text, and images to generate label, bounding box, description, classification output, attention, and activation maps, or representation vectors of the aforementioned outputs. It stores the Al outputs in a database and retrieves them based on vector proximity when needed,

[0020] An operator supervision module that allows the final scene data to be recorded in the database after operator verification through human inspection and potential addition or removal of scenes,

[0021] - An advisory module that provides users with filtered content through the application client, ensuring that the content is free from unwanted scenes or presented with filtered scenes based on the accumulated statistical information of the filtered content in the database, using support from the advanced artificial intelligence module,

[0022] - An application client operating within the multimedia device, which allows users to command the removal of unwanted scenes via voice, text, or gestures. It retrieves filter information from the database and enables the presentation of the new content by skipping or censoring unwanted scenes. It also transfers statistical information about the content watched by users to the database for recording.

[0023] To illustrate the operational system of the invention, when a family wants to watch a film, series, or any other content with their children on an online streaming platform, they can choose to watch the content with unwanted scenes, such as a scene containing blood, either in a blurred and censored format or completely removed from the content. Users can perform this action by giving commands to the application client, which operates on a multimedia device such as a television or a similar device. After the command is given, the scene detection module, preprocessing module, advanced artificial intelligence module, and operator control module work in the background to utilize the unwanted scene and object information stored in the database, enabling users to watch the content in a filtered manner through the application client. Additionally, based on the accumulated statistical information in the database before watching, the user's preferences for unwanted scenes are recorded, and the recommendation module provides alternative content through the application client.

[0024] Our invention offers several important and unique advantages compared to existing and previous technologies.

[0025] 1. Use of Generative Al: This system uses generative artificial intelligence in the stages of scene tagging, user command taking and scene representation. This allows the ability to understand scenes from the movie and what happens in the scene and incorporate this information into the recommendation system. This approach offers a higher precision in content similarity calculation compared to recommendation systems based only on superficial information.

[0026] 2. Personalized Sensitive Content Filtering: This system learns the user's scenespecific preferences for genre and sensitive content and uses this information to provide personalized recommendations. This means that the user's preferences towards sensitive content are taken into account and more granular filtering is provided instead of the film control authorities' general age category classification.

[0027] 3. Filtering Using Natural Language: This system provides an interface that allows it to issue detailed filtering commands outside of the predetermined categories. This allows users to customize their content using natural language via voice or text, providing a truly customizable experience unlike the static categories mentioned earlier.

[0028] 4. Cost Efficiency: Our invention provides cost efficiency, especially by using artificial intelligence for content tagging and filtering stages. This provides significant cost savings compared to previous systems such as traditional workforce-based labeling processes and real-time detection methods through the server. This invention significantly enhances the user experience, reduces labeling and filtering costs, and provides users with personalized content recommendations that better align with their preferences for sensitive content.

Claims

CLAIMS An artificial intelligence-powered content filtering and recommendation system system that enables users to watch films, series, and similar content on online streaming platforms in a customized manner by advanced filtering of unwanted inappropriate scenes before or during viewing, and / or recommending alternative content to watch, including:• At least one multimedia device that accepts user commands for selecting unwanted scenes through its interface and provides users with audiovisual broadcasts by accepting the chosen online streaming platform and content.• A scene detection module that determines the start and end time intervals of scenes in the content to be watched by users. and characterized by• A preprocessing module that preprocesses the scenes detected by the scene detection module, applying audio, text, and image processing techniques to transform them into a format suitable for the advanced artificial intelligence module to work on.• At least one advanced artificial intelligence module equipped with one or more generative Al models based on convolutional neural networks or transformer architectures, which utilize scene information such as audio, text, and images to generate labels, bounding boxes, descriptions, classification outputs, attention and activation maps, or representation vectors of the mentioned outputs. It stores the Al outputs in a database and retrieves them based on vector proximity when needed.• An operator supervision module that allows the final scene data to be recorded in the database after operator verification through human inspection and potential addition or removal of scenes.• An advisory module that provides users with filtered content through the application client, utilizing the accumulated statistical information in the database and with support from the advanced artificial intelligence module. It offers content without unwanted scenes or presents alternative content with filtered scenes to users before viewing.• An application client operating within the multimedia device, which allows users to watch content in a filtered manner by retrieving filter information from the database based on commands given through voice, text, or gestures. It enables the presentation of new content by skipping or censoring unwanted scenes, based on relevant filtering statistics. Additionally, it transfers statistical information about the content watched by users to the database for recording.