Description:
RELATED ART

[0001] Embodiments of the present disclosure relate generally to content delivery, and more particularly to personalized recommendation of media content.
[0002] With the advent of on-demand content delivery platforms, viewers often find themselves mindlessly scrolling through an ocean of content, leaving them confused, frustrated, or overwhelmed about what to watch. Present day content delivery platforms such as Netflix, YouTube, Hotstar, Apple TV, and the like, therefore, employ content recommendation algorithms that recommend tailored content to viewers based on their content viewing behavior. Conventionally, these recommendation algorithms consider a viewer’s watch history and viewership history of users from same demographics to provide recommendations for various content.
[0003] In particular, present day content delivery platforms use collaborative filtering to analyze viewing behavior of multiple users grouped by demographics, region, language, and other factors to identify and recommend content. While collaborative filtering aims to personalize content recommendations, it is primarily based on the assumption that users who have watched same movies will have similar preference for content. This approach leverages the preferences of similar users to make predictions for an individual, which can be relevant, but is not entirely personalized for each user’s unique preferences and interests.
[0004] Additionally, conventional platforms allow only keyword-based searches for genre, actors or movies, director, theme, and the like. Search engines associated with these conventional platforms are not designed for advanced queries that capture the essence or certain specific elements of the content which the user liked or enjoyed. As a result, conventional recommendations may not fully align with each user's distinct preferences, potentially leading to a less personalized viewing experience and reduced user satisfaction.
[0005] Accordingly, there is a need for an improved technique that enables personalized recommendation of content that caters to each user's distinct tastes and is able to support advanced search queries from users.

BRIEF DESCRIPTION

[0006] It is an objective of the present disclosure to provide a method for providing personalized content recommendations. The method includes identifying one or more content insight tags associated with one or more watched content consumed by a user by a large language model subsystem associated with a content viewing platform. The method includes generating a user viewing profile knowledge graph comprising one or more nodes and one or more edges. Each of the nodes correspond to one of the identified content insight tags and each of the edges connect two of the nodes representing a specific association between the content insight tags associated with the two connected nodes. Further, the method includes continually updating the user viewing profile knowledge graph by adding one or more new nodes, adding one or more new edges, or a combination thereof, upon identifying one or more of a new content insight tag and a new association between the content insight tags associated with newly watched content consumed by the user. The method includes continually identifying a subset of nodes from the nodes and the new nodes in the user viewing profile knowledge graph by the large language model subsystem such that a number of edges connecting each of the subset of nodes is higher than a number of edges connecting other remaining nodes selected from the nodes and the new nodes. The method includes continually selecting one or more content as one or more content recommendations hyper-personalized for the user from a content database communicatively coupled to the large language model subsystem such that one or more content insight tags associated with the selected content match one or more of the content insight tags corresponding to the subset of nodes.
[0007] Further, the method includes continually generating one or more customized headers based on the content insight tags corresponding to the subset of nodes. The method includes continually displaying the one or more customized headers along with the selected content as the content recommendation hyper-personalized for the user on a designated user interface screen associated with the content viewing platform on a user device associated with the user by replacing one or more default headers originally present on the designated user interface screen.
[0008] The method of identifying the content insight tags associated with the watched content includes identifying one or more of standard and contextual information corresponding to the watched content. The standard information includes one or more of a cast, crew, theme and genre, and the contextual information comprises one or more of a plot summary, dialogue, sub-plot, description of a specific scene, specific audio element, specific video element, review, social media comment, trailer and timestamp corresponding to when the content is watched by the user.
[0009] The method of identifying one or more of the standard and contextual information corresponding to the watched content includes identifying the standard tags that are associated with the watched content and are stored in the content database. The method of identifying one or more of the standard and contextual information corresponding to the watched content also includes identifying contextual information corresponding to the watched content captured from one or more of the public content platforms and one or more social media platforms. The method identifying one or more of the standard and contextual information corresponding to the watched content further includes storing the identified contextual information corresponding to the watched content in the content database.
[0010] Further, the method of updating the user viewing profile knowledge graph includes adding the new nodes representative of the new content insight tags associated with newly watched content consumed by the user. The method of updating the user viewing profile knowledge graph includes connecting each of the new nodes representing a new content insight tag to one or more nodes selected from the nodes and the new nodes via one or more new edges representing one or more specific associations between the new content insight tag and one or more of the corresponding content insight tags associated with the selected nodes.
[0011] The method of continually displaying the one or more customized headers along with the recommended content on the designated user interface screen includes continually updating the one or more customized headers along with the recommended content for display on the designated user interface screen upon one or more of expiry of a predefined time period and a predefined increase in number of nodes in the user viewing profile knowledge graph since a previous update to the customized headers.
[0012] The method of continually displaying the one or more customized headers along with the selected content as content recommendation hyper-personalized for the user on a designated user interface screen includes recommending content having associated content insight tags corresponding to one or more of a specific season, theme, festival, event, content expiry date, and paid promotion.
[0013] Further, the method of continually selecting one or more content as one or more content recommendations hyper-personalized for the user from the content database includes selecting subscribed content that is available to the user based on a valid user subscription to the content viewing platform that is associated with the user.
[0014] The method includes the large language model subsystem employing one or more agentic artificial intelligence applications for one or more of identifying the one or more content insight tags, generating the user viewing profile knowledge graph. The method includes the large language model subsystem employing one or more agentic artificial intelligence applications for one or more of continually updating the user viewing profile knowledge graph, continually identifying the subset of nodes, continually selecting the content from a content database. The method includes the large language model subsystem employing one or more agentic artificial intelligence applications for one or more of continually generating one or more customized headers and continually displaying the one or more customized headers along with the selected content as content recommendation hyper-personalized for the user on the designated user interface screen associated with the content viewing platform on the user device.
[0015] The content includes one or more of an audio-video content, audio content, video content, news, educational course, housing, financial information, retail product, food and beverage option and service offering made available to the user via the content viewing platform.
[0016] It is further an objective of the present disclosure to provide a content viewing platform. The content viewing platform includes a content database storing a plurality of content available on the content viewing platform that is accessible via a content viewing application stored on a user device associated with a user. The content viewing application provides access to one or more of the plurality of content available on the content viewing platform to the user based on a valid user subscription associated with the content viewing platform. The content viewing platform includes an intelligent content search and recommendation system comprising a large language model subsystem communicatively coupled to the content viewing platform and the content database.
[0017] Further, the intelligent content search and recommendation system is configured to identify one or more content insight tags associated with one or more watched content consumed by the user. The intelligent content search and recommendation system is configured to generate a user viewing profile knowledge graph comprising one or more nodes and one or more edges. Each of the nodes correspond to one of the identified content insight tags, and each of the edges connect two of the nodes representing a specific association between the content insight tags associated with the two connected nodes. The intelligent content search and recommendation system is further configured to continually update the user viewing profile knowledge graph by adding one or more new nodes, adding one or more new edges, or a combination thereof, upon identifying one or more of a new content insight tag and a new association between the content insight tags associated with newly watched content consumed by the user. The intelligent content search and recommendation system is configured to continually identify a subset of nodes from the nodes and the new nodes in the user viewing profile knowledge graph by the large language model subsystem such that a number of edges connecting each of the subset of nodes is higher than a number of edges connecting other remaining nodes selected from the nodes and the new nodes. The intelligent content search and recommendation system is configured to continually select one or more content as one or more content recommendations hyper-personalized for the user from the content database such that one or more content insight tags associated with the selected content match one or more of the content insight tags corresponding to the subset of nodes.
[0018] Further, the intelligent content search and recommendation system is configured to continually generate one or more customized headers based on the content insight tags corresponding to the subset of nodes. The intelligent content search and recommendation system is configured to continually display the one or more customized headers along with the selected content as the content recommendation hyper-personalized for the user on a designated user interface screen associated with the content viewing platform on the user device associated with the user by replacing one or more default headers originally present on the designated user interface screen.
[0019] The large language model subsystem includes one or more agentic artificial intelligence applications for one or more of identifying the one or more content insight tags, generating the user viewing profile knowledge graph, continually updating the user viewing profile knowledge graph. The large language model subsystem includes one or more agentic artificial intelligence applications for one or more of continually identifying the subset of nodes, continually selecting the content from a content database. The large language model subsystem includes one or more agentic artificial intelligence applications for one or more of continually generating one or more customized headers and continually displaying the one or more customized headers along with the selected content as content recommendation hyper-personalized for the user on the designated user interface screen associated with the content viewing platform on the user device.
[0020] It is a further objective of the present disclosure to provide a content viewing platform, wherein the content viewing platform comprises one or more of an online news platform, banking platform, stock trading platform, product listing platform, service listing platform, educational platform, and food delivery platform.

BRIEF DESCRIPTION OF DRAWINGS

[0021] These and other features, aspects, and advantages of the claimed subject matter will become better understood when the following detailed description is read with reference to the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:
[0022] FIG. 1 illustrates a block diagram depicting an exemplary system for providing personalized content recommendation to a user from a content viewing platform, in accordance with aspects of the present disclosure;
[0023] FIG. 2 illustrates a block diagram depicting an exemplary intelligent content search and recommendation system for providing personalized content recommendation to the user from the content viewing platform, in accordance with aspects of the present disclosure;
[0024] FIG. 3 illustrates an exemplary user interface screen of the content viewing platform viewed on a user device depicting default headers, in accordance with aspects of the present disclosure;
[0025] FIG. 4 illustrates an exemplary user interface screen of the content viewing platform viewed on a user device depicting customized headers generated by the intelligent content search and recommendation system of FIG. 2, in accordance with aspects of the present disclosure;
[0026] FIGS. 5A-5B illustrate a flowchart depicting an exemplary method for providing personalized content recommendation to the user on the content viewing platform, in accordance with aspects of the present disclosure; and
[0027] FIG. 6 illustrates a graphical representation of an exemplary user viewing profile knowledge graph associated with the user for providing personalized content recommendation to the user from the content viewing platform, in accordance with aspects of the present disclosure.

DETAILED DESCRIPTION

[0028] The following description presents an exemplary method and a system for providing personalized content recommendations to a user. Particularly, embodiments described herein disclose an intelligent content search and recommendation system that employs a large language model (LLM) to provide personalized content recommendations to the user from a content viewing platform. To that end, the LLM utilizes nuanced content insight tags associated with content consumed by the user to generate a unique user viewing profile knowledge graph to provide personalized content recommendations to the user that perfectly align with the user’s unique and changing preferences. Utilization of the nuanced content insight tags enables the LLM subsystem to provide recommendations not only based on the user’s watch history but also recommendations that accurately match content retrieved in response to user queries. As used herein, the term “content” may refer to all kinds of content that a user may consume at any given time using any device on any content viewing platform.
[0029] Conventional content recommendation systems employ collaborative filtering to enable content viewing platforms to group content by genre, actors, and certain standard or default set of keywords. Particularly, conventional content recommendation systems analyze the viewing behaviors of different users by grouping a plurality of users based on factors such as demography and similarities in content watch histories. Based on such groupings, the conventional recommendation systems may select and recommend one or more content watched and liked by a first set of users in the group to a second set of users in the group. However, what the recommendation system recommends may not always be appropriate, suitable, or preferred by one or more of the users for consumption.
[0030] The present disclosure describes an intelligent content search and recommendation system that utilizes an LLM subsystem that analyzes varied information associated with the watched content in a nuanced manner to overcome the aforementioned disadvantages of conventional content search and recommendation systems. In particular, the LLM subsystem automatically generates a user viewing profile knowledge graph that accurately captures and lists various content insight tags associated with one or more watched content consumed by the user. The content insight tags include certain standard information associated with the watched content including metadata tags, such as frame information, resolution, bitrates, frame length, and other associated standard information such as title of the content, cast, crew, genre and theme. The content insight tags further include contextual information such as, plot summary, user reviews, description of a scene, dialogue, sub-plot, specific audio element, specific video element, trailers and timestamp corresponding to the time the content was watched by the user. Notably, the LLM subsystem not only uses information intrinsically tagged with the media file, but also information from external sources such as news, social media posts, comments, and reviews from the user and others, and additional information repositories to generate the content insight tags and corresponding associations. The LLM subsystem subsequently navigates across content databases to identify similar content that includes one or more of the content insight tags available in the user viewing profile knowledge graph to recommend to the user. The intelligent content search and recommendation system, thus, captures the user’s unique interests and preferences of content from the user viewing profile knowledge graph. Further, the usage of various content insight tags enables the intelligent content search and recommendation system to provide personalized recommendations that perfectly match the user’s unique preferences and interests.
[0031] Further, the LLM subsystem adaptively generates and displays customized headers along with the recommended content on a designated user interface screen associated with the content viewing platform. The present intelligent content search and recommendation system, thus, not only recommends hyper personalized content but also generates customized headers, which simplifies the task of finding suitable content for the user, thereby reducing wasted doom scrolling time. Finding the right content to watch with ease improves user satisfaction and platform engagement, which benefits both users and the streaming platform. An embodiment of the present intelligent content search and recommendation system is described in greater detail with reference to FIGs. 1 and 2.
[0032] FIG. 1 shows a block diagram depicting an exemplary intelligent content search and recommendation system (102) for providing personalized recommendation of content from a content viewing platform (104) to a user that matches the user’s unique preferences. In one embodiment, the content viewing platform (104) corresponds to a server owned by a content distributor such as an over-the-top (OTT) media content provider or a video-on-demand (VOD) service provider. Some non-limiting examples of the content viewing platforms (104) include YouTube, Netflix, Prime, Hulu, Hotstar, Facebook, Instagram, Spotify, Reuters, Forbes, Fidelity, Amazon, Etsy, Uber Eats, and similar platforms. The content viewing platform (104) may include a content database (210) and a user viewing profile knowledge graph (212), both shown in FIG. 2. The content database (210) stores a plurality of content that is made available to subscribed users by the content viewing platform (104) for a subscription fee. The content may be stored in one or more designated formats, bitrates, and resolutions. Examples of the designated formats may include MPEG-4-part 14 format, MKV format, and AVI format.
[0033] Additionally, the content database (210) stores various information associated with the plurality of content, which may be referred to as content insight tags. The content insight tags may include standard information such as theme, genre, cast, crew and associated metadata corresponding to the plurality of content. Moreover, the content insight tags may also include contextual information such as plot summary, user reviews, user comments, description of specific scenes, trailers, and transcripts corresponding to the plurality of content. Further, the user viewing profile knowledge graph (212) includes information corresponding to watched content consumed by the subscribed user. Information corresponding to content consumed by the subscribed user may include the content insight tags associated with the watched content. In one or more embodiments, the content insight tags associated with the watched content in the user viewing profile knowledge graph (212) are arranged as nodes, which are interconnected via edges that represent one or more associations among the content insight tags.
[0034] In one embodiment, the intelligent content search and recommendation system (102) may be integrated into the content viewing platform (104). Alternatively, the intelligent content search and recommendation system (102) may be a standalone system communicatively coupled to the content viewing platform (104) via a communication link (108). The intelligent content search and recommendation system (102) may be able to access the user viewing profile knowledge graph (212) and the content database (210) via the communication link (108). Examples of the communication link (108) include a satellite-based communications system, an over-the-top (OTT) system, and the internet, among other generally available communication systems.
[0035] In certain embodiments, the content viewing platform (104) may be accessed by the user via a corresponding content viewing application (105) that is available over the internet and/or is installed on his or her user device (106), such as a smartphone, a laptop, a desktop, a gaming device, or a smart television. When a user opens the content viewing application (105) corresponding to the content viewing platform (104), such as Netflix, using the user device (106), a designated user interface (UI) screen is displayed on a user interface (110) of the user device (106). The designated UI screen displays one or more content as recommendations under one or more headers. The one or more displayed content may be subscribed content stored in the content database (210), which is available to users who have subscribed to the content viewing platform (104).
[0036] Conventionally, these default headers are static and are often same for all users globally. However, the position of these headers may vary on the designated UI screen of individual users depending upon one or more content the user watched or searched for in one or more previous instances of engagement with the content viewing platform (104). Additionally, or alternatively, the position of these headers may vary based on other factors, such as, content upload date, content expiry date, content release date, trending content, date and time or season of the year.
[0037] Additionally, one or more recommended content included under the default headers also vary for different users based on the viewing behaviors of the users captured by the conventional recommendation systems. Particularly, conventional recommendation systems utilize data corresponding to both user’s watch history and viewing behaviors of other viewers grouped based on factors such as demography and similarities in watch histories to recommend relevant content. While content is recommended based on the user’s past watch history, such groupings may not continue to be of relevance to the user’s current viewing behavior and preference.
[0038] In an example scenario, user A and user B have recently watched the movie "Arrival," a space-themed movie on a content viewing platform. Ideally, upon logging in to the content viewing platform, both users may be provided with multiple content recommendations that may be similar in genre or theme to the movie “Arrival”. The recommended content may include movies such as "Star Wars", “Prometheus” and “Alien” because all these movies correspond to the genre or theme “space”. However, user A might have primarily appreciated "Arrival" for its deep emotional storytelling and its exploration of human relationships. Meanwhile, user B might have been more captivated by the alien interaction aspect of "Arrival". Hence, user B might find “Alien” more to his or her liking and may watch the movie “Alien”. However, the recommendation of the movie “Alien” may not perfectly align with the preferences of user A as “Alien” may have horror elements attached to it, which may be inappropriate for user A. Hence, user A may not engage with the recommended content biased towards the theme “space”, resulting in unending scrolling.
[0039] Failure to find any preferred content that may match a user’s unique interest and preference even after continual scrolling may lead users to search content by the name of a preferred content. However, users may not always know or remember the titles of content they want to watch. Further, users may often want to specify specific elements or scenes of the content in their searches. However, conventional search engines utilized by the current content viewing platforms are not designed or equipped with the capability to return perfectly matching results for queries that may capture certain specific elements of the content instead of standard keywords such as cast and standard themes. Conventional search and recommendation systems, therefore often return content that may not be relevant or preferred by the user for consumption, thus leading to frustration in the user, eventually resulting in poor viewing experience and decreased user satisfaction.
[0040] To that end, the present intelligent content search and recommendation system (102) provides hyper-personalized recommendations by utilizing the user viewing profile knowledge graph (212). More particularly, the intelligent content search and recommendation system (102) includes an LLM subsystem (206), shown in FIG. 2, that uses an LLM and one or more agentic artificial intelligence (AI) applications to identify and recommend recently preferred content specific to the user by analyzing one or more content insight tags associated with watched content consumed by the user. In certain embodiments, the LLM subsystem (206) builds the user viewing profile knowledge graph (212) based on watched content and associated content insight tags. The LLM subsystem (206) identifies and processes the content insight tags and associations between the multiple content insight tags to generate the user viewing profile knowledge graph (212). In particular, the identified standard and contextual tags related to the watched content and corresponding associations are used to generate various nodes and corresponding connecting edges in the user viewing profile knowledge graph (212). Subsequently, the LLM subsystem (206) identifies the most appropriate content among multiple subscribed content from the content database (210) to be provided as recommendation to the user, for example, based on the nodes including the highest number of connecting edges.
[0041] In certain embodiments, the LLM subsystem (206) further generates and displays customized and unique headers and presents the recommended content under the customized headers as personalized recommendations to the specific user. Additionally, the LLM subsystem (206) adaptively updates the recommended content and customized headers periodically based on the user viewing profile knowledge graph that is continually updated based on the continuing watch history of the user. The intelligent content search and recommendation system (102), thus provides hyper-personalized recommendations for content to the user based on nuanced insights generated from user queries and history of content consumed by the user. An embodiment depicting certain exemplary components of the intelligent content search and recommendation system (102) that enables providing hyper-personalized content recommendations in response to search queries and based on continuing watch history of the user is described in greater detail with reference to FIG. 2.
[0042] FIG. 2 illustrates a block diagram depicting an embodiment of the exemplary intelligent content search and recommendation system (102) of FIG. 1 that provides a user with hyper-personalized content recommendations under customized headers. In one embodiment, the intelligent content search and recommendation system (102) and associated functions performed by the intelligent content search and recommendation system (102) may be implemented by suitable code on a processor-based system such as a general-purpose or a special-purpose computer. Accordingly, intelligent content search and recommendation system (102) includes one or more general-purpose processors, specialized processors, graphical processing units (GPU), microprocessors, programming logic arrays (PLA), field programming gate arrays (FPGA), application specific integrated circuits (ASIC), systems on chips (SOCs), and/or other suitable computing devices.
[0043] The intelligent content search and recommendation system (102) may address the shortcomings of the conventional content recommendation systems via use of one or more sub-components including, but not limiting to, an intelligent search engine (202), an analytics engine (204), and an LLM subsystem (206) In an embodiment, each of the intelligent search engine (202), analytics engine (204), and LLM subsystem (206) may be implemented as one or more GPUs, ASICs, AI on SOCs, microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or other suitable hardware devices capable of performing designated functionality based on commands from the intelligent content search and recommendation system (102).
[0044] It shall be noted that the terms user, subscriber, and subscribed user are used interchangeably throughout the description and refer to a person who has a valid subscription to the content viewing platform (104) to consume subscribed content. It shall further be noted that the plurality of content that is available to subscribed users for consumption on the content viewing platform (104) is referred to as subscribed content throughout the description. Further, subscribed content that the subscribed user prefers to watch, has watched, or has searched for may be referred to as preferred content throughout the description.
[0045] In one embodiment, accessing the content viewing application (105) associated with the content viewing platform (104) on the user device (106) results in display of the designated UI screen on the user interface (110) of the user device (106). The designated UI screen further displays a designated search field provided by the intelligent search engine (202), which may be communicatively coupled to the content viewing platform (104) via the communication link (108). The corresponding designated search field may be configured to receive various types of user queries from the user, for example, in textual form, audio form, image form and/or video form. However, for better understanding of the various embodiments of the invention, it will be assumed that the intelligent search engine (202) receives user queries in textual form and audio form via the designated search field. The queries may be in the form of a single keyword specifying the title, a theme or a genre of the content such as “Interstellar”, “Alice in Wonderland”, “Money Heist”, “Sports”, and “Horror”.
[0046] Alternatively, the queries may be in the form of long sentences including multiple keywords such as “I want to watch a murder mystery with mind bending climax”, “I want to watch a movie that shows love for pet animals with a happy ending”. Further, the queries may be in the form of a phrase or group of keywords not forming a complete sentence such as “Hostage drama with multiple plot twists”, “Space Odyssey with visually stunning graphics”, “Adaptation of children’s classic with magic and rich visuals”. In some instances, the group of keywords present in the user queries may also describe a specific scene in the content such as “A group of people, while travelling, meet a woman who tells them their future”.
[0047] Upon receiving the user queries, the analytics engine (204) processes the associated keywords related to the user queries to extract semantic information from the user queries. The analytics engine (204) further communicates with the content database (210) to identify subscribed content from a plurality of available content that may include semantic similarity with the keywords in the user queries. The analytics engine (204) identifies the semantic similarity of the subscribed content based on the keywords included in the queries by analyzing the content insight tags associated with the plurality of content. As previously noted, these content insight tags are stored in the content database (210) and may include standard information including, but not limited to, genre, cast, crew, theme, and metatags associated with the media file, and contextual information including a plot summary, description of a scene, a dialogue, a sub-plot, a description of a specific scene, an audio element, a video element, a review, a trailer and a timestamp corresponding to when the content is watched by the user. In certain embodiments, the analytics engine (204) accesses one or more public content platforms (214) such as the internet movie database (IMDB), Rotten Tomatoes, BookMyShow, Google and one or more social media platforms (216) such as Instagram, TikTok, and Snapchat to identify content insight tags corresponding to the subscribed content.
[0048] To that end, in one embodiment, the LLM subsystem (206) in the analytics engine (204) is trained and/or finetuned to extract semantic information to generate content insight tags associated with the content available in the content database (210) based on information that is provided as one or more of text, video, and audio using a large and diverse dataset. The diverse dataset may include millions of texts, documents, blogs, articles, dictionaries, and audios and videos, among others. Once trained, the LLM subsystem (206) aids the intelligent content search and recommendation system (102) to extract semantic information to generate the content insight tags to be stored in correlation with the associated content information in the content database (210).
[0049] In certain embodiments, the LLM subsystem (206) uses one or more LLMs such as BERT, Lambda, Llama, Falcon 180B, Vicuna 13B, Grok AI, Claude, Macaw-LLM, SALMONN, DeepSeek, and Meta ImageBind. Additionally, the LLM subsystem (206) may use one or more Agentic AI frameworks and intelligence tools such as Manus, CrewAI, Microsoft AutoGen, Meta AI Agent Framework, and Google Cloud Video Intelligence to compare semantic information associated with content retrieved or watched by the user in response to user queries with content insight tags associated with content available in the content database (210).
[0050] The trained LLM subsystem (206) uses the user viewing profile knowledge graph (212) to recommend content that are determined to have corresponding content insight tags that match the extracted semantic information corresponding to the retrieved and/or watched content. For example, in response to the user query, “hostage drama with multiple plot twists,” the LLM subsystem (206) parses the content database (210) to review the content insight tags associated with the subscribed content to identify content that may correspond to a hostage story as a plot, having serious tone in the story and including suspense. Subsequently, the LLM subsystem (206) recommends movies such as “Parasite” and “Oldboy” to the user as opposed to regular hostage movies such as “Taken” and “Die Hard”, typically retrieved and recommended by conventional content recommendation systems. Likewise, in response to the user query, “adaptation of children’s classic with magic and rich visuals”, the LLM subsystem (206) may recommend movies such as “A series of unfortunate events” and “Alice in wonderland” to the user. Thus, the LLM subsystem (206) uses semantic information and content insight tags available in content databases to provide nuanced search results as recommendations in response to user queries.
[0051] In an embodiment, the LLM subsystem (206) may generate the user viewing profile knowledge graph (212) after the user has consumed a content for the first time. The LLM subsystem (206) may continue displaying default headers and recommending content based on the continually consumed content by the user until a certain period or until the generated user viewing profile knowledge graph (212) expands to a certain level such that it includes sufficient information for providing hyper personalized content recommendations.
[0052] As previously noted, each node in the user viewing profile knowledge graph (212) corresponds to a content insight tag associated with a watched content, where the node is connected to one or more other nodes via one or more edges representing one or more associations between the corresponding content insight tags. Upon consumption of a new content by the user, the user viewing profile knowledge graph (212) adds new nodes representing content insight tags that may define the new watched content and/or new connecting edges between one or more existing and/or new nodes representing associations between the corresponding content insight tags.
[0053] Accordingly, the LLM subsystem (206) continuously monitors the user viewing profile knowledge graph (212) to identify a subset of nodes from the nodes and the new nodes that may have a higher number of connecting edges than a number of edges connecting other remaining nodes. The LLM subsystem (206) utilizes the information to provide recommendations to the user. Specifically, the LLM subsystem (206) identifies subscribed content from the content database (210) that include content insight tags that match one or more of the content insight tags corresponding to the subset of nodes having the highest number of edges. The LLM subsystem (206) then selects the identified subscribed content to be provided as personalized recommendations to the user. As previously noted, the LLM subsystem (206) continually updates the user viewing profile knowledge graph (212) to provide an accurate representation of the user’s viewing behavior and preferences to ensure that the recommended content accurately aligns with the user’s prevailing interests, leading to superior user experience.
[0054] In order to keep track of user’s prevailing interests and preferences, in certain embodiments, the LLM subsystem (206) adds a node representative of a recent timestamp connected to other nodes having content insight tags such as title, cast and theme associated with content, for example, watched by the user in the previous fifteen days. The LLM subsystem (206) identifies one or more subscribed content in the content database (210) that include content insight tags that match the content insight tags associated with the watched content consumed on recent dates. The LLM subsystem (206), thus, biases the content recommendations for the user based on the user’s recent watch history.
[0055] In certain other embodiments, the LLM subsystem (206) biases the content recommendations for the user to also include content that the content viewing platform (104) wants to promote. To that end, in one embodiment, the LLM subsystem (206) employs one or more agentic AI applications to tag and recommend various content with nodes having content insight tags, for example, representative of a selected season, theme, festival, event, content expiry date, and paid promotion, among others. In an embodiment, the LLM subsystem (206) may generate a list of all subscribed content that includes a specific content insight tag, for example Christmas or Olympics, to be promoted. Subsequently, the LLM subsystem (206) identifies a subset of nodes from the nodes associated with the content in the generated list that have, for example, the ten highest number of edges connected to them. The LLM subsystem (206) then recommends the content associated with the subset of nodes to the user.
[0056] In one or more embodiments, the identified content is included as personalized recommendations for the user along with customized headers in place of originally present default headers. To that end, the LLM subsystem (206) generates one or more customized headers using the content insight tags corresponding to the subset of nodes with the highest number of edges in the user viewing profile knowledge graph (212). It may be noted that nodes with the highest number of edges may correspond to semantic aspects of content that are most common in the plurality of content watched by the user and therefore are indicative of user’s viewing preference. The customized headers, thus, may include a group of keywords indicative of one or more of the content insight tags to allow the user to quickly identify preferred content for future consumption. Subsequently, the LLM subsystem (206) displays the customized headers on the designated UI screen of the content viewing platform (104) that is displayed on the user interface (110) of the user device (106). Additionally, the LLM subsystem (206) displays recommended content below the customized headers, where the recommended content includes content insight tags associated with the subscribed content that match the content insight tags corresponding to the nodes with the highest number of edges in the user viewing profile knowledge graph (212).
[0057] In an example, if the user watches Arrival, Alien and Alice in Wonderland, the nodes in the user viewing profile knowledge graph (212) having the top three highest number of edges may correspond to the content insight tags “Thought Provoking”, “Space movies”, and “Adaptation of children’s classic.” Accordingly, the LLM subsystem (206) generates customized headers such as “Thought Provoking space movies” and “Adaptation of children’s classic” for display on the user interface (110) in place of certain originally present default headers such as “Trending now”, Because you watched”, “Thriller”, and “Feel-good movies” to recommend content that most accurately matches user’s current viewing preferences. The LLM subsystem (206), thus, enables the intelligent content search and recommendation system (102) to capture and analyze each individual user’s unique preference or interest and provides customized headers specific to the user’s interest to reduce doom scrolling, thus significantly enhancing each user’s content viewing experience.
[0058] The LLM subsystem (206) may update recommended content, and the customized headers based on one or more nodes that may have the highest or the greatest number of edges connected to them in the user viewing profile knowledge graph (212) at the time of accessing the designated UI screen by user. Alternatively, the LLM subsystem (206) may further update recommended content, and the one or more customized headers based on the user viewing profile knowledge graph (212) based on a predefined duration, the expiry of which triggers an updated display of the recommended content on the designated UI screen. Examples of the predefined duration may include one day, one week and one fortnight, among others. The LLM subsystem (206) may also periodically change the positions of one or more customized headers on the designated UI screen based on requirements specified by a moderator or an administrator of the content viewing platform (104).
[0059] FIG. 3 illustrates an exemplary designated UI screen (300) of the content viewing platform (104) viewed on the user device (106) depicting one or more default headers (302A-302D) and corresponding content (304A-304C). The designated UI screen (300) as seen in FIG. 3 corresponds to a default screen displayed to the user when the user viewing profile knowledge graph (212) is at an initial stage, for example, when the user has newly subscribed to the content viewing platform (104) and thus has limited watch history.
[0060] As the user continues to consume content on the content viewing platform (104), the user viewing profile knowledge graph (212) also expands, for example, with addition of new content insight tags associated with the subsequently watched content. Subsequently, the analytics engine (204) replaces one or more of the default headers (302C-302D) displayed on the designated UI screen (300) with customized headers (402A-402B), for example, corresponding to “Thought provoking space movies,” and “Adaptation of children’s classic,” as shown in FIG.4.
[0061] Further, one or more personalized content (404A-404B) is presented along with the customized headers (402A-402B) as recommendations that are specifically based on the user’s unique preferences and interests captured in the user viewing profile knowledge graph (212). An embodiment of a method of providing personalized content recommendations and generating and displaying the customized headers along with hyper-personalized recommendations by replacing original default headers and associated content recommendations is described in greater detail with reference to FIGS. 5A and 5B.
[0062] FIGS. 5A-5B illustrate a flowchart (500) depicting an exemplary method for providing personalized content recommendations. The order in which the exemplary method is described is not intended to be construed as a limitation, and any number of the described blocks may be combined in any order to implement the exemplary method disclosed herein, or an equivalent alternative method. Additionally, certain blocks may be deleted from the exemplary method or augmented by additional blocks with added functionality without departing from the claimed scope of the subject matter described herein.
[0063] At step (502), the LLM subsystem (206) in the intelligent content search and recommendation system (102) identifies content insight tags associated with the watched content consumed by the user. As previously noted, the content insight tags include various standard as well as contextual information associated with the watched content in addition to metadata associated with the watched content. The content insight tags may be made available to the LLM subsystem (206) from the content database (210), the public content platforms (214) and the social media platforms (216). In an embodiment, the LLM subsystem (206) may obtain content insight tags such as plot summary, trailers, cast, crew and genre from the content database (210) and/or the public content platforms (214), while identifying other content insight tags such as user reviews, social media comments, dialogues, sub-plots, themes, specific audio or video elements, and specific scene descriptions from one or more of the public content platforms (214) and the social media platforms (216).
[0064] In a scenario where the user is accessing the content viewing platform (104) for the first time, the LLM subsystem (206) generates the user viewing profile knowledge graph (212) by identifying the content insight tags associated with the watched content, as noted at step (504). When the user continues to consume content on the content viewing platform (104), the LLM subsystem (206) adds additional content insight tags associated with the subsequently watched content to the user viewing profile knowledge graph (212), thereby continually expanding or updating the generated user viewing profile knowledge graph (212).
[0065] The content insight tags in the user viewing profile knowledge graph (212) are arranged in a relational representation including the interconnected nodes and connecting edges. Each edge represents a specific association between the content insight tags corresponding to two interconnecting nodes. More specifically, each content insight tag associated with a watched content may be represented by a node and each node is connected to another node via an edge representing an association or relation therebetween. As the user continues to consume more content, new nodes representing content insight tags associated with the newly watched content are added and new edges are created from both previous and new nodes based on the association between a previous node and a new node or between two new nodes, at step (506).
[0066] The LLM subsystem (206) continually monitors the user viewing profile knowledge graph (212) to identify a subset of nodes from the nodes and the new nodes in the user viewing profile knowledge graph (212) that may have a higher number of connecting edges than a number of edges connecting other remaining nodes in the user viewing profile knowledge graph (212), at step (508). The LLM subsystem (206) identifies the subset of nodes having the higher number of connected edges as corresponding to content insight tags that most accurately define the user’s interest and preferences. At step (510), the LLM subsystem (206) selects content from the content database (210) having matching content insight tags as the content insight tags corresponding to the identified subset of nodes having the higher number of edges in the user viewing profile knowledge graph (212) for recommendation to the user.
[0067] At step (512), the LLM subsystem (206) continually generates customized headers (402A-402B) based on the content insight tags corresponding to the subset of nodes. In some embodiments, the LLM subsystem (206) is further configured to generate customized headers (402A-402B) comprising a group of keywords selected from the one or more matching content insight tags associated with the recommended content. The customized headers (402A-402B) and the selected content are displayed as recommended content (404A-404B) hyper-personalized to match the preference and interest of the user on a designated user interface screen (300) on the user device (106) associated with the user, at step (514). It may be noted that the LLM subsystem (206) selects the recommended content from the subscribed content available to the user for viewing on the content viewing platform (104). The content viewing platform (104) displays the recommended content (404A-404B) and the customized headers (402A-402B) by replacing the default headers (302C-302D) originally present on the designated user interface screen (300). In one embodiment, the recommended content (404A-404B) is presented under or next to the generated customized headers (402A-402B) on the designated UI screen (300) of the content viewing platform (104).
[0068] In certain embodiments, the one or more customized headers (402A-402B) along with the recommended content (404A-404B) may be displayed to the user for a predefined time period and/or until there is a predefined increase in number of nodes in the user viewing profile knowledge graph (212) since a previous update made to the default headers (302C-302D) or the customized headers (402A-402B). Additionally, the LLM subsystem (206) continually analyzes the user viewing profile knowledge graph (212) to determine other relevant content to be provided as recommended content (404A-404B) . The present intelligent content search and recommendation system (102), thus attempts to recommend content that may more accurately align with each user's distinct preferences by utilizing the user viewing profile knowledge graph (212). An embodiment of the user viewing profile knowledge graph (212) is shown and described in greater detail with reference to FIG. 6.
[0069] FIG. 6 illustrates a graphical representation of an exemplary user viewing profile knowledge graph associated with a specific user of the content viewing platform (104) for providing personalized content recommendations. In an embodiment, when a content is first watched or consumed by the user, the LLM subsystem (206) generates the user viewing profile knowledge graph (212), which is stored in the content database (210). Upon subsequent consumption of content by the user over time, the LLM subsystem (206) continually updates the user viewing profile knowledge graph (212) with additional nodes and edges upon identifying new content insight tags and new associations between the content insight tags, respectively.
[0070] As seen in FIG. 6, when the user watches the movie “Arrival” on the content viewing platform (104), the LLM subsystem (206) generates a node defining the title “Arrival”. Similarly, multiple other nodes including context insight tags such as “space”, “drama”, “alien interaction”, “English”, “duration 120 minutes”, “Amy Adams”, among others will be generated. The LLM subsystem (206) generates these content insight tags for defining semantics and contextual information associated with the movie “Arrival”. Accordingly, the LLM subsystem (206) connects the node “Arrival” to “Amy Adams”, “space”, “drama” and other nodes via corresponding edges. The LLM subsystem (206) defines an association or a relationship between two interconnecting nodes that are connected via an edge. For example, the LLM subsystem (206) defines the relationship between “Arrival” and “space” as “theme”. Likewise, the LLM subsystem (206) defines the relationship between “Arrival” and “Amy Adams” as “stars in”.
[0071] When the user watches new content, the LLM subsystem (206) updates the user viewing profile knowledge graph (212) with additional nodes associated with the newly watched content. If the newly watched content includes one or more content insight tags that are similar to content insight tags of previously watched content, the LLM subsystem (206) may not add new nodes corresponding to the similar content insight tags. Instead, the LLM subsystem (206) extends new edges from the older or previous nodes to the new nodes that are associated with the newly watched content. In an exemplary scenario, the user may watch the movie “Alien” after watching “Arrival”. In this scenario, the LLM subsystem (206) does not generate an additional node corresponding to the content insight tag “space”. Instead, the LLM subsystem (206) connects the existing node corresponding to “space” associated with “Arrival” with a newly generated node corresponding to the newly watched movie “Alien” via a new edge representing the relationship between “Alien” and “space” as “theme”. Additional new edges may be extended from node “space” to the newly generated content insight tags associated with “Alien”, such as nodes “horror” and “Ridley Scott”. Likewise, the LLM subsystem (206) defines the relationship between “Alien” and “Ridley Scott” as “is directed by”. Reusing already generated nodes representing content insight tags ensures lesser resource utilization by the user viewing profile knowledge graph (212).
[0072] Embodiments of the present intelligent content search and recommendation system (102), thus, mitigate limitations of conventional collaborative filtering content recommendation systems by providing hyper-personalized and continually evolving content recommendations for each user. The present intelligent content search and recommendation system (102) is powered by the LLM subsystem (206) to continually generate the user viewing profile knowledge graph (212), including content insight tags associated with watched content consumed by the user based on nuanced semantics and contextual information corresponding to the watched content. The LLM subsystem (206) further defines one or more associations between the content insight tags, a number of associations of each node subsequently being used to identify the most preferred content among multiple subscribed content from the content database (210) to be provided as recommendation to the user. The intelligent content search and recommendation system (102), thus provides hyper-personalized and continually evolving recommendations for content to the user based on more nuanced information related to the watched content.
[0073] Furthermore, unlike conventional recommendation systems, the intelligent content search and recommendation system (102) adaptively generates customized headers and identifies associated content for recommendation to users. The intelligent content search and recommendation system (102) displays the customized headers (402A-402B) along with the identified content (404A-404B) as recommendations on a designated user interface screen (300) of the content viewing platform (104) viewed on the user device (106) by replacing one or more originally present default headers (302C-302D). The intelligent content search and recommendation system (102) disclosed in the present disclosure uses the LLM subsystem (206) to adaptively update the customized headers from time to time based on user queries and watch history to present content recommendations that more accurately represent user preferences at the time. The present intelligent content search and recommendation system (102) thus, improves user satisfaction and platform engagement, thereby driving increased viewership, benefiting both platform and users.
[0074] For clarity, embodiments of the intelligent content search and recommendation system (102) are described in the present disclosure with reference to an audio and video content viewing platform (104). However, it may be noted that the present intelligent content search and recommendation system (102) may similarly be used for recommending other types of content. These other types of content may include, but are not limited to, content such as news, educational courses, housing, financial information, retail products, food and beverage options and various service offerings made available to users through the corresponding content viewing platform (104). The content viewing platform (104), for example, may include online news platforms, banking platforms, stock trading platforms, product and service listing platforms, educational platforms, and food delivery platforms.
[0075] Although specific features of various embodiments of the present systems and methods may be shown in and/or described with respect to some drawings and not in others, this is for convenience only. It is to be understood that the described features, structures, and/or characteristics may be combined and/or used interchangeably in any suitable manner in the various embodiments shown in the different figures.
[0076] While only certain features of the present systems and methods have been illustrated and described herein, many modifications and changes will occur to those skilled in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes.

LIST OF NUMERAL REFERENCES:

102 Intelligent content search and recommendation system
104 Content viewing platform
106 User device
108 Communication link
110 User interface
202 Intelligent search engine
204 Analytics engine
206 LLM subsystem
210 Content database
212 User viewing profile knowledge graph
214 Public content platforms
216 Social media platforms
300 Designated user interface
302A-302D Default headers
304A-304D Recommended content corresponding to default headers
402A-402B Customized headers
404A-404B Recommended content corresponding to customized headers
500 Method for providing personalized content recommendations
502-514 Steps of method for providing personalized content recommendations

, Claims:
We claim:

1. A method for providing personalized content recommendations, the method comprising:
identifying one or more content insight tags associated with one or more watched content consumed by a user by a large language model subsystem (206) associated with a content viewing platform (104);
generating a user viewing profile knowledge graph (212) comprising one or more nodes and one or more edges, each of the nodes corresponding to one of the identified content insight tags, and each of the edges connecting two of the nodes representing a specific association between the content insight tags associated with the two connected nodes;
continually updating the user viewing profile knowledge graph (212) by adding one or more new nodes, adding one or more new edges, or a combination thereof, upon identifying one or more of a new content insight tag and a new association between the content insight tags associated with newly watched content consumed by the user;
continually identifying a subset of nodes from the nodes and the new nodes in the user viewing profile knowledge graph (212) by the large language model subsystem (206), wherein a number of edges connecting each of the subset of nodes is higher than a number of edges connecting other remaining nodes selected from the nodes and the new nodes;
continually selecting one or more content as one or more content recommendations hyper-personalized for the user from a content database (210) communicatively coupled to the large language model subsystem (206), wherein one or more content insight tags associated with the selected content match one or more of the content insight tags corresponding to the subset of nodes;
continually generating one or more customized headers (402A-402B) based on the content insight tags corresponding to the subset of nodes; and
continually displaying the one or more customized headers (402A-402B) along with the selected content (404A-404B) as the content recommendation hyper-personalized for the user on a designated user interface screen (300) associated with the content viewing platform (104) on a user device (106) associated with the user by replacing one or more default headers (302C-302D) originally present on the designated user interface screen (300).

2. The method as claimed in claim 1, wherein identifying the content insight tags associated with the watched content comprises identifying one or more of standard and contextual information corresponding to the watched content, wherein the standard information comprises one or more of a cast, crew, theme and genre, and wherein the contextual information comprises a plot summary, description of a scene, dialogue, sub-plot, description of a specific scene, specific audio element, specific video element, review, social media comment, trailer, and timestamp corresponding to when the content is watched by the user.

3. The method as claimed in claim 2, wherein identifying one or more of the standard and contextual information corresponding to the watched content comprises:
identifying the standard tags that are associated with the watched content and are stored in the content database (210);
identifying contextual information corresponding to the watched content captured from one or more of the public content platforms (214) and one or more social media platforms (216); and
storing the identified contextual information corresponding to the watched content in the content database (210).

4. The method as claimed in claim 1, wherein updating the user viewing profile knowledge graph (212) comprises:
adding the new nodes representative of the new content insight tags associated with newly watched content consumed by the user; and
connecting each of the new nodes representing a new content insight tag to one or more nodes selected from the nodes and the new nodes via one or more new edges representing one or more specific associations between the new content insight tag and one or more of the corresponding content insight tags associated with the selected nodes.

5. The method as claimed in claim 1, wherein continually displaying the one or more customized headers (402A-402B) along with the recommended content (404A-404B) on the designated user interface screen (300) comprises continually updating the one or more customized headers (402A-402B) along with the recommended content (404A-404B) for display on the designated user interface screen (300) upon one or more of expiry of a predefined time period and a predefined increase in number of nodes in the user viewing profile knowledge graph (212) since a previous update made to the customized headers (402A-402B).

6. The method as claimed in claim 1, wherein continually displaying the one or more customized headers (402A-402B) along with the selected content (404A-404B) as content recommendation hyper-personalized for the user on a designated user interface screen (300) comprises recommending content having associated content insight tags corresponding to one or more of a specific season, theme, festival, event, content expiry date, and paid promotion.

7. The method as claimed in claim 1, wherein continually selecting one or more content as one or more content recommendations hyper-personalized for the user from the content database (210) comprises selecting subscribed content that is available to the user based on a valid user subscription to the content viewing platform (104) that is associated with the user.

8. The method as claimed in claim 1, wherein the large language model subsystem (206) employs one or more agentic artificial intelligence applications for one or more of identifying the one or more content insight tags, generating the user viewing profile knowledge graph (212), continually updating the user viewing profile knowledge graph (212), continually identifying the subset of nodes, continually selecting the content from a content database (210), continually generating one or more customized headers (302C-302D), and continually displaying the one or more customized headers (302C-302D) along with the selected content (304C-304D) as content recommendation hyper-personalized for the user on the designated user interface screen (300) associated with the content viewing platform (104) on the user device (106).

9. The method as claimed in claim 1, wherein the content comprises one or more of an audio-video content, audio content, video content, news, educational course, housing, financial information, retail product, food and beverage option and service offering made available to the user via the content viewing platform (104).

10. A content viewing platform (104), comprising:
a content database (210) storing a plurality of content available on the content viewing platform (104) that is accessible via a content viewing application (105) stored on a user device (106) associated with a user, wherein the content viewing application (105) provides access to one or more of the plurality of content available on the content viewing platform (104) to the user based on a valid user subscription associated with the content viewing platform (104); and
an intelligent content search and recommendation system (102) comprising a large language model subsystem (206) communicatively coupled to the content viewing platform (104) and the content database (210), wherein the intelligent content search and recommendation system (102) is configured to:
identify one or more content insight tags associated with one or more watched content consumed by the user;
generate a user viewing profile knowledge graph (212) comprising one or more nodes and one or more edges, each of the nodes corresponding to one of the identified content insight tags, and each of the edges connecting two of the nodes representing a specific association between the content insight tags associated with the two connected nodes;
continually update the user viewing profile knowledge graph (212) by adding one or more new nodes, adding one or more new edges, or a combination thereof, upon identifying one or more of a new content insight tag and a new association between the content insight tags associated with newly watched content consumed by the user;
continually identify a subset of nodes from the nodes and the new nodes in the user viewing profile knowledge graph (212) by the large language model subsystem (206), wherein a number of edges connecting each of the subset of nodes is higher than a number of edges connecting other remaining nodes selected from the nodes and the new nodes;
continually select one or more content as one or more content recommendations hyper-personalized for the user from the content database (210), wherein one or more content insight tags associated with the selected content match one or more of the content insight tags corresponding to the subset of nodes;
continually generate one or more customized headers (402A-402B) based on the content insight tags corresponding to the subset of nodes; and
continually display the one or more customized headers (402A-402B) along with the selected content (404A-404B) as the content recommendation hyper-personalized for the user on a designated user interface screen (300) associated with the content viewing platform (104) on the user device (106) associated with the user by replacing one or more default headers originally present on the designated user interface screen (300).

11. The content viewing platform (104), as claimed in claim 10, wherein the content viewing platform (104) comprises one or more of an online news platform, banking platform, stock trading platform, product listing platform, service listing platform, educational platform, and food delivery platform.