A content recommendation method and device based on an IPTV interactive scene and a medium

Abstract

The application discloses a content recommendation method and device based on an IPTV interactive scene and a medium. Through a multi-dimensional three-dimensional perception network on a data collection layer, user description data corresponding to a target user is obtained; in a recall stage in a calculation and analysis layer, each dimension data in the user description data is processed through a double-tower model to obtain low-dimensional dense user and low-dimensional dense content feature vectors, and an initial screening related content item is obtained through a vector space approximate search algorithm; in a sorting stage in the calculation and analysis layer, a target content to be recommended is obtained through calculation and sorting by a reinforcement learning sorting algorithm module; an intelligent service matrix module in a recommendation engine service layer is instructed to optimize and process the target content to be recommended and feed back to the target user. The application solves the problems of new content recommendation difficulty, single and inflexible content recommendation and high calculation complexity, and improves the flexibility and diversity of content recommendation.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a content recommendation method, apparatus and medium based on IPTV interactive scenarios. Background Technology

[0002] In interactive network television (IPTV), in order to better improve the user experience, it is necessary to recommend relevant content to different users.

[0003] In the process of developing this invention, the following shortcomings of existing technologies were discovered: Currently, recommendation methods generally employ collaborative filtering, content-based recommendation, or a hybrid approach combining both. Collaborative filtering relies on large amounts of user behavior data, leading to a cold start problem where new users or content are difficult to recommend. Content-based recommendation methods may result in overly simplistic and undiversified content, heavily relying on feature extraction; otherwise, the recommendation effectiveness is limited. Hybrid recommendation methods suffer from high computational complexity, especially in large-scale data scenarios, lacking dynamic adaptability and failing to respond in real-time to changes in user behavior. Summary of the Invention

[0004] This invention provides a content recommendation method, device, and medium based on IPTV interactive scenarios, which effectively improves the flexibility and diversity of content recommendation.

[0005] According to one aspect of the present invention, a content recommendation method based on IPTV interactive scenarios is provided, wherein the method is executed by a content recommendation system based on IPTV interactive scenarios, the system comprising: a data acquisition layer, a calculation and analysis layer, and a recommendation engine service layer, including:

[0006] By constructing a multi-dimensional, three-dimensional perception network on the data acquisition layer, user description data corresponding to the target user can be obtained;

[0007] In the recall phase of the computational analysis layer, the user description data of each dimension is processed by a pre-constructed dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors. Then, the initial relevant content items are obtained by using a pre-set vector space approximation search algorithm.

[0008] In the ranking stage of the computational analysis layer, the initial filtered content items are calculated and ranked by a pre-optimized reinforcement learning ranking algorithm module to obtain the target content to be recommended.

[0009] The system instructs a pre-built intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and then provide feedback to the target user.

[0010] According to another aspect of the present invention, a content recommendation device based on an IPTV interactive scenario is provided, wherein the device is executed by a content recommendation system based on an IPTV interactive scenario, the system comprising: a data acquisition layer, a calculation and analysis layer, and a recommendation engine service layer, including:

[0011] The user description data acquisition module is used to acquire user description data corresponding to the target user by building a multi-dimensional three-dimensional perception network on the data acquisition layer.

[0012] The initial screening related content item determination module is used in the recall stage of the calculation and analysis layer to process the data of each dimension in the user description data through a pre-built dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors, and to obtain the initial screening related content items through a pre-set vector space approximation search algorithm.

[0013] The target content to be recommended module is used to calculate and sort the initially screened relevant content items in the ranking stage of the calculation and analysis layer through a pre-optimized reinforcement learning ranking algorithm module to obtain the target content to be recommended.

[0014] The target content to be recommended feedback module is used to instruct the pre-built intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and provide feedback to the target user.

[0015] According to another aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the content recommendation method based on IPTV interactive scenarios as described in any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the content recommendation method based on IPTV interactive scenarios as described in any embodiment of the present invention.

[0017] The technical solution of this invention acquires user description data corresponding to a target user through a multi-dimensional, three-dimensional perception network constructed on the data acquisition layer. In the recall phase of the computational analysis layer, a pre-constructed dual-tower model processes the data of each dimension in the user description data to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors. An initial selection of relevant content items is obtained through a pre-set vector space approximation search algorithm. In the ranking phase of the computational analysis layer, a pre-optimized reinforcement learning ranking algorithm module calculates and ranks the initial selection of relevant content items to obtain target content to be recommended. Finally, a pre-constructed intelligent service matrix module in the recommendation engine service layer is instructed to optimize the target content to be recommended and provide feedback to the target user. This solves the problems of difficulty in recommending new content, limited and inflexible content recommendation, and high computational complexity, improving the flexibility and diversity of content recommendation.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a content recommendation method based on an IPTV interactive scenario provided in Embodiment 1 of the present invention;

[0021] Figure 2 This is a detailed flowchart of a content recommendation method based on an IPTV interactive scenario provided by Embodiment 2 of the present invention;

[0022] Figure 3 This is a schematic diagram of the structure of a content recommendation device based on an IPTV interactive scenario according to Embodiment 3 of the present invention;

[0023] Figure 4 This is a schematic diagram of the structure of an electronic device provided according to Embodiment 4 of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "target," "current," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] Example 1

[0027] Figure 1 The flowchart of a content recommendation method based on IPTV interactive scenarios is provided in Embodiment 1 of the present invention. This embodiment is applicable to the situation of recommending content in IPTV interactive scenarios. The method can be executed by a content recommendation device based on IPTV interactive scenarios, specifically by a content recommendation system based on IPTV interactive scenarios. The content recommendation device based on IPTV interactive scenarios can be implemented in hardware and / or software.

[0028] Correspondingly, this is performed by a content recommendation system based on IPTV interactive scenarios, which includes a data acquisition layer, a calculation and analysis layer, and a recommendation engine service layer.

[0029] like Figure 1 As shown, the method includes:

[0030] S110. By constructing a multi-dimensional, three-dimensional perception network on the data acquisition layer, user description data corresponding to the target user is obtained.

[0031] User description data can be multi-dimensional descriptions of users. The data acquisition layer is a processing layer that collects user description data.

[0032] Specifically, the user description data comprises multi-dimensional data, including basic user attribute data, interaction behavior sequence data, environmental context data, and content feature data. The basic user attribute data includes static age profile data and static geographic profile data. The interaction behavior sequence data includes dynamic click behavior, hovering behavior, swiping behavior, sharing behavior, and skipping behavior. The environmental context data includes current device network status scenario information and current device model scenario information. The content feature data includes video bitrate metadata and video category metadata.

[0033] In this embodiment, the data acquisition layer constructs a multi-dimensional perception network and uses a terminal tracking software development kit to collect user description data in real time. Specifically, it collects relevant data from user basic attribute data, interaction behavior sequence data, environmental context data, and content feature data.

[0034] S120. In the recall stage of the computational analysis layer, the data of each dimension in the user description data is processed by a pre-constructed dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors, and the initial relevant content items are obtained by a pre-set vector space approximation search algorithm.

[0035] In this embodiment, the computational analysis layer includes a recall phase and a ranking phase. A dual-tower model, comprising a user tower and a content tower, is deployed in the recall phase. A reinforcement learning ranking algorithm module is introduced in the ranking phase.

[0036] Optionally, the dual-tower model includes a user tower and a content tower; in the recall phase of the computational analysis layer, the user description data is processed using the pre-constructed dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors, including: in the recall phase of the computational analysis layer, the user basic attribute data, interaction behavior sequence data, and environmental context data are processed in a low-dimensional dense manner using the user tower of the dual-tower model to obtain low-dimensional dense user feature vectors; and the content feature data is processed in a low-dimensional dense manner using the content tower of the dual-tower model to obtain low-dimensional dense content feature vectors.

[0037] Specifically, the user tower transforms sparse, high-dimensional user behavior into a 256-dimensional dense vector using deep neural networks. The content tower processes text, image, and video features uniformly through multimodal fusion technology.

[0038] For example, the information for target user A includes data such as "gender: male", "age: 25", "city: Beijing", and "recent viewing history: movie B and movie C".

[0039] Traditional high-dimensional sparse vectors are described as:

[0040] [0, 1, 0, # Gender (3D);

[0041] 0,...,1,...,0, # City (1000 dimensions, Beijing corresponds to position 1);

[0042] 0,1,0,0,...,0, # Device (50 dimensions);

[0043] 1,0,0,1,0,...,0 # Viewing history (10,000 dimensions, with movie B / C corresponding to position 1)]

[0044] At this point, the description of the information for target user A has approximately 10,000 dimensions.

[0045] However, through the user pyramid, the information of target user A can be transformed into a low-dimensional dense vector, which might be described as:

[0046] [0.23, -0.56, 0.89, 0.12, -0.34, ..., 0.08] # All 256 dimensions are non-zero data, including compressed data such as gender (16 dimensions), age (2 dimensions), city (32 dimensions), and viewing (2 dimensions).

[0047] Additionally, Content Tower will use a multimodal fusion architecture to reduce the dimensionality of content features. Specifically, content can include textual data (title / synopsis), visual data (poster / stills), and structural data (genre / actors).

[0048] For example, the description of the TV series "AA" includes data such as "Genre: [Criminal Investigation, Suspense]"; "Director: D"; "Cast: E"; "Keywords: Campus, Exam"; and "Rating: 8.5".

[0049] Traditional high-dimensional sparse vectors are described as:

[0050] [0,1,0,...,0, # Type - Criminal Investigation (2nd dimension = 1);

[0051] 0,...,1,...,0,# Director-D (assuming the 1342nd dimension = 1);

[0052] 1,0,...,1,0, # Actor-E (1st dimension = 1);

[0053] 0,0,1,1,...,0,# Keywords - campus, exam (3rd dimension = 1, 4th dimension = 1);

[0054] 8.5 # Rating

[0055] At this point, the content description dimensions are approximately 50,000.

[0056] However, through the content pyramid, content can be transformed into a low-dimensional dense vector, which might be described as:

[0057] [0.35, -0.12, 0.28, ..., 0.18] # All 256 dimensions of non-zero data, including compressed 128-dimensional semantic vectors, 64-dimensional visual features, and structured data.

[0058] Optionally, obtaining the initial relevant content items through a pre-set vector space approximation search algorithm includes: performing an approximate feature vector search in the vector space based on the low-dimensional dense user feature vector and the low-dimensional dense content feature vector, to obtain the initial relevant content items corresponding to the approximate feature vector.

[0059] Among them, the vector space approximation search algorithm can be an algorithm that searches for approximate feature vectors in a vector space.

[0060] For example, assuming a vector search is performed in a candidate pool of tens of millions in the vector space, an approximate feature vector search is performed based on the low-dimensional dense user feature vector and the low-dimensional dense content feature vector to obtain the initial screening related content items corresponding to the approximate feature vector. Assuming that the obtained initial screening related content items are 500, this process can greatly reduce the cold start time for new users and improve the efficiency and flexibility of content recommendation.

[0061] S130. In the ranking stage of the calculation and analysis layer, the initial screening related content items are calculated and ranked by a pre-optimized reinforcement learning ranking algorithm module to obtain the target content to be recommended.

[0062] The reinforcement learning ranking algorithm module includes a state awareness unit, an action decision unit, and a reward feedback unit; wherein, the reward feedback unit adds a long-tail content incentive calculation item and a skip penalty calculation item when calculating the reward value.

[0063] Among them, the state perception unit can dynamically capture the needs of the target user through data such as real-time user behavior, environmental information and historical interaction information, and then optimize and calculate the reward value through the action decision unit and reward feedback unit, thereby obtaining one or more target content to be recommended.

[0064] S140, Instruct the pre-built intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and provide feedback to the target user.

[0065] Optionally, the intelligent service matrix module includes: a QOE optimization algorithm bitrate adaptation unit, a containerized deployment unit, and a tiered degradation strategy unit; the pre-built intelligent service matrix module in the instruction recommendation engine service layer optimizes the target content to be recommended and feeds it back to the target user, including: obtaining content description information corresponding to the target content to be recommended, and performing bitrate conversion on the target content to be recommended through the QOE optimization algorithm bitrate adaptation unit to obtain target content to be recommended with adapted bitrate; wherein, the content description information includes video quality score, video stuttering penalty, and bitrate switching penalty; the QOE optimization algorithm bitrate adaptation unit contains a QOE optimization algorithm bitrate adaptation method formula, which is... ,in, For the current adapted bitrate; Rate the video quality; This represents the stuttering penalty coefficient. This is the penalty coefficient for bitrate switching; For video quality; For the duration of buffering; The number of switching times; through the parameter server of the containerized deployment unit, the target bitrate-adapted content to be recommended is hot-updated in seconds, and combined with the non-core feature shutdown method of the hierarchical degradation strategy unit, the target bitrate-adapted content to be recommended is fed back to the target user.

[0066] Among them, the QOE optimization algorithm bitrate adaptation method can maximize the user's viewing experience and balance video quality, stuttering time and switching frequency for the target user's terminal device under limited network bandwidth.

[0067] For the formula of the QOE optimization algorithm's bitrate adaptation method, ; The subjective quality score for the video is the current video bitrate, which is generally between 1.0 and 2.0. For example, a 4K video (bitrate 8000kbps) has a quality score of 2.0, and a 720P video (bitrate 2000kbps) has a quality score of 1.2. For video quality, a logarithmic function based on bitrate is used. For example, when the bitrate is 2000kbps, the quality score is 1.5. This is the stuttering penalty coefficient, representing the negative impact of each buffering stutter on the user experience, typically ranging from 3.0 to 5.0. For example, if a user experiences a 2-second stutter while watching, the corresponding calculation... Deduct 3 x 2 = 6 points. The duration of the lag is the cumulative buffer time. This is the bitrate switching penalty coefficient, representing the loss of experience caused by frequent bitrate increases and decreases. The initial value is set to 0.5~1.5. For example, if the bitrate changes from 4K to 720P to 1080P within one minute, the cumulative switching penalty can be calculated. . To determine the number of switches, for example, 5 switches within 10 minutes, we can know... .

[0068] In this embodiment, the parameter server of the containerized deployment unit performs second-level hot updates on the target bitrate-adapted content to be recommended, and combines the non-core feature disabling method of the hierarchical degradation strategy unit to provide feedback on the target bitrate-adapted content to the target user.

[0069] Containerized deployment units can support A / B testing and canary releases, and achieve second-level hot updates through parameter servers, assuming that inference latency can be stabilized within 50 milliseconds.

[0070] For the non-core feature shutdown method of the hierarchical degradation strategy unit, when network congestion of the terminal device is detected, the system will sequentially shut down non-core features (such as personalized copywriting), reduce the complexity of the ranking model, and finally fall back to the fallback recommendation based on popularity, so as to maintain the pre-set millisecond-level response speed even in extreme cases.

[0071] In addition, a dynamic resource allocation algorithm was designed for sudden traffic surges. When concurrent requests exceed the threshold, computing resources are automatically allocated to the sorting stage to ensure the stability of the core link.

[0072] Specifically, each user interaction is fed back to the data acquisition layer in real time. After feature engineering processing by the computation and analysis layer, it triggers incremental model updates. The updated strategy immediately takes effect in the engine service layer, forming a positive cycle of continuous self-evolution. A specific scenario could be: a target user suddenly watches multiple suspense videos while resting at night. The short-term window captures this change in interest within 5 minutes, and the reinforcement learning model increases the weight of relevant content in the next recommendation cycle. At the same time, the QOE algorithm can automatically select the 1080P bitrate based on the home network conditions. The entire process requires no manual intervention.

[0073] The technical solution of this invention acquires user description data corresponding to a target user through a multi-dimensional three-dimensional perception network constructed on the data acquisition layer. In the recall phase of the computational analysis layer, a pre-constructed dual-tower model processes the data of each dimension in the user description data to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors. An initial selection of relevant content items is obtained through a pre-set vector space approximation search algorithm. In the ranking phase of the computational analysis layer, a pre-optimized reinforcement learning ranking algorithm module calculates and ranks the initial selection of relevant content items to obtain target content to be recommended. Finally, a pre-constructed intelligent service matrix module in the recommendation engine service layer is instructed to optimize the target content to be recommended and provide feedback to the target user. This solves the problems of difficulty in recommending new content, the lack of diversity and flexibility in content recommendation, and high computational complexity, improving the flexibility and diversity of content recommendation, reducing the difficulty of recommending new content, and lowering computational complexity.

[0074] Example 2

[0075] Figure 2 This is a detailed flowchart of a content recommendation method based on an IPTV interactive scenario according to Embodiment 2 of the present invention. This embodiment is a refinement based on the above embodiments. In this embodiment, during the ranking stage in the calculation and analysis layer, a pre-optimized reinforcement learning ranking algorithm module is used to calculate and rank the initially screened relevant content items to obtain the target content to be recommended for further refinement.

[0076] S210. By constructing a multi-dimensional, three-dimensional perception network on the data acquisition layer, user description data corresponding to the target user is obtained.

[0077] S220. In the recall phase of the computational analysis layer, the data of each dimension in the user description data is processed by a pre-constructed dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors. The initial relevant content items are obtained by using a pre-set vector space approximation search algorithm.

[0078] S230. Obtain the real-time state perception description parameters corresponding to the target user by instructing the state perception unit.

[0079] Among them, the state-aware unit can dynamically capture the needs of the target user through data such as real-time user behavior, environmental information, and historical interaction information.

[0080] S240. By instructing the action decision unit, the initial screening related content items are sorted according to the acquired real-time state perception description parameters, and the current interaction behavior sequence data corresponding to the target user is acquired in real time. The current reward value is calculated by the reward value calculation formula in the reward feedback unit.

[0081] Optionally, the step of acquiring the current interaction behavior sequence data corresponding to the target user in real time and calculating the current reward value using the reward value calculation formula in the reward feedback unit includes: calculating the current reward value based on the acquired current interaction behavior sequence data corresponding to the target user in real time using the reward value calculation formula in the reward feedback unit. The formula for calculating the reward value is as follows: ,in, , , , and These are the weight parameters; The click reward value corresponding to the aforementioned dynamic click behavior; The hovering reward value corresponding to the hovering dynamic behavior; The sharing reward value corresponding to the aforementioned sharing dynamic behavior; This is a calculation item for incentives for long-tail content; To skip the penalty calculation item; Rewards for long-tail content; To skip the penalty value.

[0082] Among these, click dynamic behavior refers to users explicitly clicking on content, which is a strong positive signal. Hover dynamic behavior refers to users lingering for more than 3 seconds without clicking, which is a weak positive signal. Sharing dynamic behavior refers to users sharing content to social media platforms, which is an extremely strong positive signal. The long-tail content incentive calculation item can be set to a reward value of 1 when the recommended content is niche or has low exposure, and 0 otherwise. The skip penalty calculation item is for users quickly swiping over content, specifically, it can be set to allow users to linger for less than 1 second.

[0083] In this embodiment, it can be based on the formula To calculate the current reward value.

[0084] S250, instruct the reward feedback unit to perform an update calculation of the adjustment strategy based on the current reward value using a dynamic strategy optimization calculation method, and obtain the current updated adjustment strategy.

[0085] Optionally, the step of updating and calculating the adjustment strategy based on the current reward value using a dynamic strategy optimization calculation method to obtain the current updated adjustment strategy includes: updating and calculating the adjustment strategy based on the current reward value using a dynamic strategy optimization calculation method to obtain the current updated adjustment strategy. Among them, the dynamic strategy optimization calculation method is as follows: ;in, The policy learning rate; Let s be the gradient of the logarithmic probability of choosing action a, where a is the action in the user interaction and s is the current real-time state of the user. The expected reward value corresponding to the current user's real-time status; It is the entropy coefficient; Let be the policy entropy.

[0086] Specifically, The policy learning rate controls the step size of policy updates, with a preset value of 0.0001~0.001. For Let be the gradient of the logarithmic probability of choosing action 'a' under the current policy. This gradient is used to measure the importance of the current action; the higher the probability of an action, the larger its gradient. For For example, let R be the advantage function, which is the current reward value R minus the expected reward value. This indicates the degree of surprise or delight associated with an action. For example, the expectation... However, if actual user clicks and shares (assuming R is 3.0) are achieved, the advantage value is +2.0, which can significantly improve the recommendation strategy.

[0087] in, Entropy coefficient, a parameter controlling the intensity of exploration. The larger the size, the more diverse the recommendations. The smaller the value, the more conservative the recommendation; let's assume a preset value of 0.01~0.1. The entropy is the probability distribution disorder of a strategy. The higher the entropy, the stronger the exploratory nature. For example, if the strategy is always [90% suspense, 10% criminal investigation], the entropy is very low (≈0.325), and the system will encourage the exploration of other categories of content.

[0088] In this embodiment, content is filtered from a massive pool of candidates based on real-time state-aware description parameters, and the display strategy (which can be sorting, style, or copywriting) is optimized. The system's evolution is driven by reward values. By quantifying user feedback (positive behaviors such as clicks, hovering, and sharing, combined with long-tail content incentives and skip penalties), the model is guided to adjust. This improves the real-time recommendation effect while ensuring the diversity of the content ecosystem, achieving a balance between accurate recommendation and interest exploration.

[0089] S260. Based on the current update adjustment strategy, obtain the target content to be recommended.

[0090] By processing the initial screening items according to the current updated adjustment strategy, the target content to be recommended can be further obtained.

[0091] S270, Instruct the pre-built intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and provide feedback to the target user.

[0092] The technical solution of this invention acquires user description data corresponding to a target user through a multi-dimensional three-dimensional perception network constructed on the data acquisition layer. In the recall phase of the computational analysis layer, a pre-constructed dual-tower model processes the data of each dimension in the user description data to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors. An initial selection of relevant content items is obtained through a pre-set vector space approximation search algorithm. In the ranking phase of the computational analysis layer, a pre-optimized reinforcement learning ranking algorithm module calculates and ranks the initial selection of relevant content items to obtain target content to be recommended. Finally, a pre-constructed intelligent service matrix module in the recommendation engine service layer is instructed to optimize the target content to be recommended and provide feedback to the target user. This solves the problems of difficulty in recommending new content, the lack of diversity and flexibility in content recommendation, and high computational complexity, improving the flexibility and diversity of content recommendation, reducing the difficulty of recommending new content, and lowering computational complexity.

[0093] Example 3

[0094] Figure 3 This is a schematic diagram of a content recommendation device based on an IPTV interactive scenario provided in Embodiment 3 of the present invention. The content recommendation device based on an IPTV interactive scenario provided in this embodiment can be implemented by software and / or hardware, and can be configured in a terminal device or server to implement a content recommendation method based on an IPTV interactive scenario according to an embodiment of the present invention. Figure 3 As shown, the device includes: a user description data acquisition module 310, an initial screening related content item determination module 320, a target recommended content determination module 330, and a target recommended content feedback module 340.

[0095] The content recommendation system based on IPTV interactive scenarios is executed by a data acquisition layer, a calculation and analysis layer, and a recommendation engine service layer.

[0096] The user description data acquisition module 310 is used to acquire user description data corresponding to the target user by constructing a multi-dimensional three-dimensional perception network on the data acquisition layer.

[0097] The initial screening related content item determination module 320 is used to process the data of each dimension in the user description data through a pre-constructed dual-tower model in the recall stage of the calculation and analysis layer, to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors, and to obtain the initial screening related content items through a pre-set vector space approximation search algorithm.

[0098] The target content to be recommended module 330 is used to calculate and sort the initial screening related content items through a pre-optimized reinforcement learning ranking algorithm module in the ranking stage of the calculation and analysis layer to obtain the target content to be recommended.

[0099] The target content to be recommended feedback module 340 is used to instruct the pre-built intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and provide feedback to the target user.

[0100] The technical solution of this invention acquires user description data corresponding to a target user through a multi-dimensional three-dimensional perception network constructed on the data acquisition layer. In the recall phase of the computational analysis layer, a pre-constructed dual-tower model processes the data of each dimension in the user description data to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors. An initial selection of relevant content items is obtained through a pre-set vector space approximation search algorithm. In the ranking phase of the computational analysis layer, a pre-optimized reinforcement learning ranking algorithm module calculates and ranks the initial selection of relevant content items to obtain target content to be recommended. Finally, a pre-constructed intelligent service matrix module in the recommendation engine service layer is instructed to optimize the target content to be recommended and provide feedback to the target user. This solves the problems of difficulty in recommending new content, the lack of diversity and flexibility in content recommendation, and high computational complexity, improving the flexibility and diversity of content recommendation, reducing the difficulty of recommending new content, and lowering computational complexity.

[0101] Based on the above embodiments, the user description data is multi-dimensional data, including user basic attribute data, interaction behavior sequence data, environmental context data, and content feature data; wherein, the user basic attribute data includes age static profile data and region static profile data; the interaction behavior sequence data includes click dynamic behavior, hover dynamic behavior, swipe dynamic behavior, share dynamic behavior, and skip dynamic behavior; the environmental context data includes current device network status scenario information and current device model scenario information; and the content feature data includes video bitrate metadata and video category metadata.

[0102] Based on the above embodiments, the dual-tower model includes a user tower and a content tower.

[0103] Based on the above embodiments, the initial screening related content item determination module 320 can be specifically used to: in the recall stage of the calculation and analysis layer, perform low-dimensional dense processing on user basic attribute data, interaction behavior sequence data and environmental context data through the user tower in the dual-tower model to obtain a low-dimensional dense user feature vector; and perform low-dimensional dense processing on content feature data through the content tower in the dual-tower model to obtain a low-dimensional dense content feature vector.

[0104] Based on the above embodiments, the initial screening related content item determination module 320 can also be specifically used to: perform an approximate feature vector search in the vector space according to the low-dimensional dense user feature vector and the low-dimensional dense content feature vector, and obtain the initial screening related content item corresponding to the approximate feature vector.

[0105] Based on the above embodiments, the reinforcement learning ranking algorithm module includes a state awareness unit, an action decision unit, and a reward feedback unit; wherein, the reward feedback unit adds a long-tail content incentive calculation item and a skip penalty calculation item when calculating the reward value.

[0106] Based on the above embodiments, the target content to be recommended module 330 may specifically include: a real-time state perception description parameter acquisition submodule, used to acquire the real-time state perception description parameters corresponding to the target user by instructing the state perception unit; a current reward value determination submodule, used to instruct the action decision unit to sort the initial screening related content items according to the acquired real-time state perception description parameters, and to acquire the current interaction behavior sequence data corresponding to the target user in real time, and to calculate the current reward value through the reward value calculation formula in the reward feedback unit; a current update adjustment strategy determination submodule, used to instruct the reward feedback unit to update the adjustment strategy according to the current reward value through a dynamic strategy optimization calculation method to obtain the current update adjustment strategy; and a target content to be recommended submodule, used to obtain the target content to be recommended according to the current update adjustment strategy.

[0107] Based on the above embodiments, the current reward value determination submodule can be specifically used to: calculate the current reward value according to the reward value calculation formula in the reward feedback unit based on the real-time acquired current interaction behavior sequence data corresponding to the target user. The formula for calculating the reward value is as follows: ,in, , , , and These are the weight parameters; The click reward value corresponding to the aforementioned dynamic click behavior; The hovering reward value corresponding to the hovering dynamic behavior; The sharing reward value corresponding to the aforementioned sharing dynamic behavior; This is a calculation item for incentives for long-tail content; To skip the penalty calculation item; Rewards for long-tail content; To skip the penalty value.

[0108] Based on the above embodiments, the current update adjustment strategy determination submodule can be specifically used to: perform update calculations on the current reward value using a dynamic strategy optimization calculation method to obtain the current update adjustment strategy. Among them, the dynamic strategy optimization calculation method is as follows: ;in, The policy learning rate; Let s be the gradient of the logarithmic probability of choosing action a, where a is the action in the user interaction and s is the current real-time state of the user. The expected reward value corresponding to the current user's real-time status; It is the entropy coefficient; Let be the policy entropy.

[0109] Based on the above embodiments, the intelligent service matrix module includes: a QOE optimization algorithm bitrate adaptation unit, a containerized deployment unit, and a tiered degradation strategy unit.

[0110] Based on the above embodiments, the target content to be recommended feedback module 340 can be specifically used to: obtain content description information corresponding to the target content to be recommended, and perform bitrate conversion on the target content to be recommended through the QOE optimization algorithm bitrate adaptation unit to obtain target content to be recommended with adapted bitrate; wherein, the content description information includes video quality score, video stuttering penalty, and bitrate switching penalty; the QOE optimization algorithm bitrate adaptation unit includes a QOE optimization algorithm bitrate adaptation method formula, which is... ,in, For the current adapted bitrate; Rate the video quality; This represents the stuttering penalty coefficient. This is the penalty coefficient for bitrate switching; For video quality; For the duration of buffering; The number of switching times; through the parameter server of the containerized deployment unit, the target bitrate-adapted content to be recommended is hot-updated in seconds, and combined with the non-core feature shutdown method of the hierarchical degradation strategy unit, the target bitrate-adapted content to be recommended is fed back to the target user.

[0111] The content recommendation device based on IPTV interactive scenarios provided in this embodiment of the invention can execute the content recommendation method based on IPTV interactive scenarios provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0112] Example 4

[0113] Figure 4 A schematic diagram of an electronic device 10, which can be used to implement Embodiment 4 of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0114] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0115] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0116] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as content recommendation methods based on IPTV interactive scenarios.

[0117] In some embodiments, the content recommendation method based on IPTV interactive scenarios can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the content recommendation method based on IPTV interactive scenarios described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the content recommendation method based on IPTV interactive scenarios by any other suitable means (e.g., by means of firmware).

[0118] This method, executed by a content recommendation system based on IPTV interactive scenarios, comprises a data acquisition layer, a computational analysis layer, and a recommendation engine service layer. The method includes: acquiring user description data corresponding to the target user through a multi-dimensional, three-dimensional perception network constructed on the data acquisition layer; in the recall phase of the computational analysis layer, processing the data of each dimension in the user description data using a pre-constructed dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors, and obtaining initial filtered relevant content items through a pre-set vector space approximation search algorithm; in the ranking phase of the computational analysis layer, calculating and ranking the initial filtered relevant content items through a pre-optimized reinforcement learning ranking algorithm module to obtain target content to be recommended; and instructing a pre-constructed intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and provide feedback to the target user.

[0119] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0120] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0121] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0122] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0123] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0124] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0125] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0126] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0127] Example 5

[0128] Embodiment 5 of the present invention also provides a computer-readable storage medium, wherein the computer-readable instructions, when executed by a computer processor, are used to execute a content recommendation method based on an IPTV interactive scenario, executed by a content recommendation system based on an IPTV interactive scenario. The system includes: a data acquisition layer, a computational analysis layer, and a recommendation engine service layer. The method includes: acquiring user description data corresponding to a target user through a multi-dimensional stereoscopic perception network constructed on the data acquisition layer; in the recall phase of the computational analysis layer, processing the data of each dimension in the user description data through a pre-constructed dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors, and obtaining initial filtered relevant content items through a pre-set vector space approximation search algorithm; in the ranking phase of the computational analysis layer, calculating and ranking the initial filtered relevant content items through a pre-optimized reinforcement learning ranking algorithm module to obtain target content to be recommended; and instructing a pre-constructed intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and provide feedback to the target user.

[0129] Of course, the computer-executable instructions provided in the embodiments of the present invention, which include a computer-readable storage medium, are not limited to the method operations described above, but can also perform related operations in content recommendation based on IPTV interactive scenarios provided in any embodiment of the present invention.

[0130] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0131] It is worth noting that in the above embodiments of content recommendation based on IPTV interactive scenarios, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0132] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A content recommendation method based on IPTV interactive scenarios, characterized in that, The content recommendation system based on IPTV interactive scenarios is executed by a data acquisition layer, a calculation and analysis layer, and a recommendation engine service layer. By constructing a multi-dimensional, three-dimensional perception network on the data acquisition layer, user description data corresponding to the target user can be obtained; In the recall phase of the computational analysis layer, the user description data of each dimension is processed by a pre-constructed dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors. Then, the initial relevant content items are obtained by using a pre-set vector space approximation search algorithm. In the ranking stage of the computational analysis layer, the initial filtered content items are calculated and ranked by a pre-optimized reinforcement learning ranking algorithm module to obtain the target content to be recommended. The system instructs the pre-built intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and then provide feedback to the target user. The user description data is multi-dimensional, including user basic attribute data, interaction behavior sequence data, environmental context data, and content feature data. The user basic attribute data includes age static profile data and region static profile data; The interactive behavior sequence data includes click dynamic behavior, hover dynamic behavior, swipe dynamic behavior, share dynamic behavior, and skip dynamic behavior; The environmental context data includes current device network status scenario information and current device model scenario information; The content feature data includes video bitrate metadata and video category metadata; The reinforcement learning ranking algorithm module includes a state awareness unit, an action decision unit, and a reward feedback unit; wherein, the reward feedback unit adds a long-tail content incentive calculation item and a skip penalty calculation item when calculating the reward value; In the ranking stage of the computational analysis layer, a pre-optimized reinforcement learning ranking algorithm module is used to calculate and rank the initially filtered relevant content items to obtain the target content to be recommended, including: By instructing the state perception unit, the real-time state perception description parameters corresponding to the target user are obtained; By instructing the action decision unit, the initial filtering related content items will be sorted according to the acquired real-time state perception description parameters, and the current interaction behavior sequence data corresponding to the target user will be acquired in real time. The current reward value will be calculated by the reward value calculation formula in the reward feedback unit. The reward feedback unit is instructed to perform an update calculation of the adjustment strategy based on the current reward value using a dynamic strategy optimization calculation method to obtain the current updated adjustment strategy. Based on the current update and adjustment strategy, the target content to be recommended is obtained; The step of acquiring the current interaction behavior sequence data of the target user in real time and calculating the current reward value using the reward value calculation formula in the reward feedback unit includes: Based on the real-time acquired sequence of current interaction behaviors of the target user, the current reward value is calculated using the reward value calculation formula in the reward feedback unit. ; The formula for calculating the reward value is as follows: ,in, , , , and These are the weight parameters; The click reward value corresponding to the aforementioned dynamic click behavior; The hovering reward value corresponding to the hovering dynamic behavior; The sharing reward value corresponding to the aforementioned sharing dynamic behavior; This is a calculation item for incentives for long-tail content; To skip the penalty calculation item; Rewards for long-tail content; To skip the penalty value; The step of updating and calculating the adjustment strategy based on the current reward value using a dynamic strategy optimization calculation method to obtain the current updated adjustment strategy includes: The current reward value is updated and adjusted using a dynamic strategy optimization calculation method to obtain the current update and adjustment strategy. ; The dynamic strategy optimization calculation method is as follows: ;in, The learning rate is the policy rate. Let s be the gradient of the logarithmic probability of choosing action a, where a is the action in the user interaction and s is the current real-time state of the user. The expected reward value corresponding to the current user's real-time status; It is the entropy coefficient; For policy entropy; Among them, click dynamic behavior is when a user explicitly clicks on content, and click dynamic behavior is a strong positive signal; hover dynamic behavior is when a user stays for more than 3 seconds but does not click, and hover dynamic behavior is a weak positive signal; share dynamic behavior is when a user shares content to a social media platform, which is an extremely strong positive signal; for the long-tail content incentive calculation item, when the recommended content is niche or low-exposure content, the long-tail content reward value is set to 1, otherwise the long-tail content reward value is 0; the skip penalty calculation item is for users to quickly swipe past.

2. The method according to claim 1, characterized in that, The dual-tower model includes a user tower and a content tower; In the recall phase of the computational analysis layer, a pre-constructed dual-tower model is used to process the data of each dimension in the user description data to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors, including: In the recall phase of the computational analysis layer, the user basic attribute data, interaction behavior sequence data and environmental context data are processed in a low-dimensional dense manner through the user tower in the dual-tower model to obtain a low-dimensional dense user feature vector. By using the content tower in the dual-tower model, the content feature data is processed in a low-dimensional and dense manner to obtain a low-dimensional and dense content feature vector.

3. The method according to claim 2, characterized in that, The process of obtaining initial filtered relevant content items through a pre-set vector space approximation search algorithm includes: Based on the low-dimensional dense user feature vector and the low-dimensional dense content feature vector, an approximate feature vector search is performed in the vector space using a pre-set vector space approximate search algorithm to obtain the initial screening related content items corresponding to the approximate feature vector.

4. The method according to claim 3, characterized in that, The intelligent service matrix module includes: a QOE optimization algorithm bitrate adaptation unit, a containerized deployment unit, and a tiered degradation strategy unit; The instruction is provided to the pre-built intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and provide feedback to the target user, including: The content description information corresponding to the target content to be recommended is obtained, and the bitrate of the target content to be recommended is converted by the QOE optimization algorithm bitrate adaptation unit to obtain the target content to be recommended bitrate adapted. The content description information includes video quality score, video stuttering penalty, and bitrate switching penalty; the QOE optimization algorithm bitrate adaptation unit contains the QOE optimization algorithm bitrate adaptation method formula, which is... ,in, For the current adapted bitrate; Rate the video quality; This represents the stuttering penalty coefficient. This is the penalty coefficient for bitrate switching; For video quality; For the duration of buffering; This refers to the number of times the device has switched. The parameter server of the containerized deployment unit performs second-level hot updates on the target bitrate-adapted content to be recommended, and combines the non-core feature disabling method of the hierarchical degradation strategy unit to provide feedback on the target bitrate-adapted content to the target user.

5. A content recommendation device based on IPTV interactive scenarios, characterized in that, The content recommendation system based on IPTV interactive scenarios is executed by a data acquisition layer, a calculation and analysis layer, and a recommendation engine service layer. The user description data acquisition module is used to acquire user description data corresponding to the target user by building a multi-dimensional three-dimensional perception network on the data acquisition layer. The initial screening related content item determination module is used in the recall stage of the calculation and analysis layer to process the data of each dimension in the user description data through a pre-built dual-tower model to obtain low-dimensional dense user feature vectors and low-dimensional dense content feature vectors, and to obtain the initial screening related content items through a pre-set vector space approximation search algorithm. The target content to be recommended module is used to calculate and sort the initially screened relevant content items in the ranking stage of the calculation and analysis layer through a pre-optimized reinforcement learning ranking algorithm module to obtain the target content to be recommended. The target content to be recommended feedback module is used to instruct the pre-built intelligent service matrix module in the recommendation engine service layer to optimize the target content to be recommended and provide feedback to the target user. The user description data comprises multi-dimensional data, including basic user attribute data, interaction behavior sequence data, environmental context data, and content feature data. Specifically, the basic user attribute data includes static age profile data and static geographic profile data; the interaction behavior sequence data includes dynamic click behavior, dynamic hover behavior, dynamic swipe behavior, dynamic sharing behavior, and dynamic skip behavior; the environmental context data includes current device network status scenario information and current device model scenario information; and the content feature data includes video bitrate metadata and video category metadata. The target content to be recommended module includes: a real-time state perception description parameter acquisition submodule, used to acquire the real-time state perception description parameters corresponding to the target user by instructing the state perception unit; a current reward value determination submodule, used to sort the initial screening related content items according to the acquired real-time state perception description parameters by instructing the action decision unit, and to acquire the current interaction behavior sequence data corresponding to the target user in real time, and calculate the current reward value by using the reward value calculation formula in the reward feedback unit; a current update adjustment strategy determination submodule, used to instruct the reward feedback unit to update the adjustment strategy according to the current reward value by using a dynamic strategy optimization calculation method to obtain the current update adjustment strategy; and a target content to be recommended submodule, used to obtain the target content to be recommended according to the current update adjustment strategy. The current reward value determination submodule is used to: calculate the current reward value based on the real-time acquired current interaction behavior sequence data of the target user using the reward value calculation formula in the reward feedback unit. The formula for calculating the reward value is as follows: ,in, , , , and These are the weight parameters; The click reward value corresponding to the aforementioned dynamic click behavior; The hovering reward value corresponding to the hovering dynamic behavior; The sharing reward value corresponding to the aforementioned sharing dynamic behavior; This is a calculation item for incentives for long-tail content; To skip the penalty calculation item; Rewards for long-tail content; To skip the penalty value; The current update adjustment strategy determination submodule is used to: perform update calculations on the current reward value using a dynamic strategy optimization calculation method to obtain the current update adjustment strategy. Among them, the dynamic strategy optimization calculation method is as follows: ;in, The learning rate is the policy rate. Let s be the gradient of the logarithmic probability of choosing action a, where a is the action in the user interaction and s is the current real-time state of the user. The expected reward value corresponding to the current user's real-time status; It is the entropy coefficient; For policy entropy; Among them, click dynamic behavior is when a user explicitly clicks on content, and click dynamic behavior is a strong positive signal; hover dynamic behavior is when a user stays for more than 3 seconds but does not click, and hover dynamic behavior is a weak positive signal; share dynamic behavior is when a user shares content to a social media platform, which is an extremely strong positive signal; for the long-tail content incentive calculation item, when the recommended content is niche or low-exposure content, the long-tail content reward value is set to 1, otherwise the long-tail content reward value is 0; the skip penalty calculation item is for users to quickly swipe past.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a content recommendation method based on an IPTV interactive scenario as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute a content recommendation method based on an IPTV interactive scenario as described in any one of claims 1-4.

Images