A hierarchical preference-aware video content caching method

Abstract

This invention relates to a hierarchical preference-aware video content caching method, belonging to the field of wireless communication. The method includes: S1: constructing a hierarchical user personal preference feature model based on video type preferences and content rankings for that type; S2: constructing a pre-caching model that maximizes the system hit rate; S3: combining request probabilities and latency information under different user access methods to construct a latency-minimizing model, selecting an appropriate caching strategy for the pre-cached content to minimize the average latency of user requests. This invention can effectively reduce the average latency of user requests and has broad application prospects.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication and relates to a hierarchical preference-aware video content caching method. Background Technology

[0002] In video transmission, Quality of Experience (QoE) places stricter demands on reducing network latency. Under these data demands, backhaul link transmission faces numerous challenges. To address this, Akamai has proposed wireless caching, which migrates video buffering from the backbone Content Delivery Network (CDN) to edge network nodes, such as user terminals and service base stations (BS). For large-scale video transmission, wireless video caching alleviates the bottleneck of limited backhaul link capacity and effectively improves service latency.

[0003] To further increase the storage capacity of edge caches, a promising approach is to deploy small base stations (SBSs) that can offload large amounts of data from macro base stations (MBSs). SBSs, with their smaller storage capacity, form a heterogeneous cellular network together with MBSs. Compared to traditional cellular networks, the dense small-cell access of heterogeneous cellular networks avoids signal blind spots and allows for cost-effective integration of existing WiFi and cellular technologies, improving the performance of peak data traffic control strategies. However, during peak traffic periods, backhaul link capacity demands extremely high data traffic, while the limited storage capacity of edge nodes restricts the storage of massive amounts of video content, making the deployment of high-capacity backbone networks for small cells very costly. Therefore, solutions based solely on heterogeneous cellular networks are insufficient to effectively address peak traffic demands; it is necessary to rationally utilize cache entity storage space and appropriately manage cache resources to maximize the caching efficiency of limited cache capacity.

[0004] To address this issue, Mobile Network Operators (MNOs) proactively cache popular content on Service Provider Base (SBS) networks (SBSs) to provide services closer to users. This not only reduces peak network traffic but also significantly lowers latency for users requesting data content. However, on the one hand, in proactive caching at edge nodes, video popularity can vary over time, depending on user characteristics and mobility patterns, making accurate prediction of popularity extremely difficult. Some studies assume the probability of a user requesting certain videos is known, failing to consider the global or local nature of popular content distribution. Instead, they uniformly use global popularity across all regions, which is unsuitable for real-world applications. On the other hand, most proactive caching strategies consider homogeneous preference models, where all users independently and randomly request content based on the same popularity. This fails to consider the significant impact of user content preferences on caching system performance in the context of multimedia content distribution, especially in wireless social networks where different regional groups have vastly different preferences for video content. Such homogeneous processing is inaccurate for the heterogeneity of user content preferences.

[0005] Therefore, for video content in cloud servers, it is necessary not only to consider the overall popularity of the content, but also to consider the different preferences of users in different regions. It is necessary to conduct preference analysis on users in the region, design caching strategies based on user content preferences, and complete the placement of content in different regions. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a hierarchical preference-aware video content caching method. A hierarchical user content preference prediction model is designed based on LSTM. This model uses historical user request information to predict the probability of requesting unknown content. The hierarchical prediction model depends on the user's preference for a type and the preference for content within that type. LSTM is used to predict the preference for content within a type, thereby obtaining preference information. Using this preference information, probability models and latency models are constructed for different user request methods. For the objective function of maximizing the hit rate within a region, bilateral exchange is used to filter out the content that needs to be placed in the pre-cached set, thus obtaining the pre-cached content for edge nodes while ensuring the maximum hit rate. For the obtained pre-cached set with the maximum hit rate, it is necessary to further determine which edge node these contents should be placed on, so that each SBS can obtain the minimum average latency for user content retrieval under collaborative caching. The Minimize Latency Collaboration Caching (MLCC) algorithm iteratively alternates between optimal collaborative delivery and cache placement to obtain a caching strategy that minimizes latency.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A hierarchical preference-aware video content caching method, comprising the following steps:

[0009] S1: Construct a hierarchical user personal preference feature model based on video type preferences and content ranking of that type;

[0010] S2: Construct a pre-caching model that maximizes the system hit rate;

[0011] S3: Combining request probability and latency information under different user access methods, construct a latency minimization model, select an appropriate caching strategy for the pre-cached content, and minimize the average latency of user requests.

[0012] Optionally, S1 specifically includes the following steps:

[0013] S11: Divide the content I in the content library into G types. User u has their own preference for each type of content. Define a preference function r(u,g) to represent user u's preference for type g:

[0014]

[0015] Among them, h u For user u's historical access content information, p(g|h u ) is the conditional feature probability, which is the probability that the historical information contains content of type g; p(g) is the unconditional feature probability, which represents the probability that the entire content set contains content of type g.

[0016] S12: The probability α of user u's preference for content i i,u The preference for content i is jointly determined by the content type preference and the content preference within that type, P(f). Therefore, the preference of user u for content i is defined through a hierarchical structure based on type preferences:

[0017]

[0018] Where r(u,g) represents user u's preference for type g; P(f) represents content i's preference for that type; χ represents the weighting factor of the ranking within the type on the probability of user preference; α i,u The value of is between [0,1]. The larger the value, the greater the probability that user u requests content i.

[0019] S13: A user's preference for video content is determined by both their preference for that content type and the content's ranking within that type. Specifically, user u's preference for type g depends on the number of videos in that type that have been rated by the user, expressed as a probability p(g) of requesting a set of rated videos containing type g, and the probability p(g|h) of requesting a set of videos containing type g in the user's historical information. u The user u's preference P(f) for content in type g cannot be directly obtained, so we consider using LSTM for time series prediction.

[0020] The LSTM network structure includes input gate x t Forgotten Gate t Output gate y t These gates decide whether to retain or discard data based on the importance of the data in the sequence; within time t, the LSTM has L inputs x(t) = [g u (t),g u (t+1),...,g u [(t+L-1)], where L is the time window, defined as the number of input data samples for each unit of the LSTM; the ideal output of the LSTM is y(t) = g u (t+L), where t=1,2,...,T d T d It is the total number of data samples; define g. u (t+L) represents the prediction output of the LSTM at time t. The optimization objective during prediction is to optimize the weights and biases of the LSTM to minimize the root mean square error (RMSE), where RMSE is defined as:

[0021]

[0022] Where ω represents the number of training iterations.

[0023] Optionally, S2 specifically includes the following steps:

[0024] S21: When the content requested by a user is not on the edge server, it needs to be retrieved from the cloud server. If this content is frequently requested by users, each request will have to be transmitted from the cloud server to MBS before being sent to the user, resulting in significant backhaul transmission redundancy. The caching system needs to meet the user's edge cache hit rate requirement, i.e., the probability that the content requested by the user is cached by the edge server. The pre-caching objective function that maximizes the hit rate within the MBS coverage area is defined as:

[0025] P1:

[0026] stC1:

[0027] C2:

[0028] Among them, optimization variables The pre-cached set is created by selecting high-hit-rate content from cloud servers for the MBS coverage area and placing it into the pre-cached set. In this way, the hit rate within the MBS coverage area is improved, while the content selection difficulty of the caching strategy is reduced; α i,u β represents the probability of user u's preference for content i, obtained by LSTM prediction; i Represents a pre-cached set binary variable, β i =1 That is, content i belongs to the pre-cached set. β i =0 C1 indicates that each piece of content can only appear once in the pre-cache set; C2 indicates the constraint on the capacity of the pre-cache set, s i The size of content i;

[0029] S22: To solve for the pre-cached content sets of MBS and SBSn in P1 use This represents the complement set of the pre-cached content, i.e. Given all content stored in the cloud server; utilize swap blocking pairs to complete a bilateral swap of the pre-cached set, and define a swap blocking pair as follows: Given a pair (i,j), where and Then (i,j) satisfies When the time is a blocking pair for swapping, That is, when swapping content i in the pre-cache set and content j in the non-pre-cache set, the sets Hit rate If an additional pair (i,j) is added, then (i,j) is called a blocking pair. A blocking pair means that (i,j) is preventing the pre-cached set from moving towards a higher hit rate. The positions of (i,j) in the set need to be swapped until no more blocking pairs are found, thus improving the hit rate and yielding the pre-caching strategy.

[0030] Optionally, S3 specifically includes the following steps:

[0031] S31: During the content request phase, if the SBS has cached the content, the user u retrieves the requested content directly from the local SBS via the wireless communication link, taking into account the user's content preferences and the caching status of the content on the local SBS. i,n The probability that user u obtains content i from SBSn / MBS is:

[0032] p u,n,i =α i,u c i,n

[0033] Where, α i,u c represents the probability that user u prefers content i; i,n Let i be a binary cache variable for content, and n = 0;

[0034] When a user cannot obtain the requested content from SBS and MBS, the local SBS retrieves the requested content from other SBSs. The probability that SBSn obtains the requested content i from the collaborating SBSm is:

[0035]

[0036] Among them, M n,m This represents the set of cooperative SBSm around SBSn that satisfy the transmission conditions, and the channel condition between SBSm and MBS is log(1+SNR). m,0 When )>E, m∈M n,m E represents the minimum uplink capacity requirement; This is used to determine whether the cooperating SBSm (n≠m) of SBSn has cached content i. If the value is 'i', it means that the SBSm cached content 'i'; otherwise, it is not cached

[0037] S32: By optimizing cache placement during the caching stage, users can independently request their preferred content; latency is used as a performance indicator for user content requests, and transmission latency is calculated for different request methods.

[0038] When a user retrieves requested content from the local SBS, the requested content is transmitted via the wireless link. The data rate from SBSn to user u is calculated as follows:

[0039]

[0040] Among them, B SBS The channel bandwidth from SBS to the user; g n,u It is the channel gain between SBSn and user u, g n,u Depend on Given, where h n,u It is a random variable that follows a Gaussian distribution (0,1). It is the path loss constant, d n,u P is the distance between SBSn and user u, where ν is the path loss exponent; SBS This is the transmission power of SBS; It is co-channel interference from other SBSs; σ 2 It is the power of additive white Gaussian noise;

[0041] The latency for user u to download content i from local SBSn is:

[0042]

[0043] Among them, s i,u It is the size of the video content i requested by user u;

[0044] Similarly, the downlink transmission rate from MBS to SBSn is:

[0045]

[0046] Among them, B MBS The channel bandwidth from MBS to SBS; g 0,n This is the channel gain of MBS; P MBS This is the maximum transmit power of the MBS;

[0047] When a user fails to request content i on the local SBS, but content i is stored in the MBS, the latency for user u to download the requested content i from the MBS is given as:

[0048]

[0049] When the collaborating SBSm provides requested content to the user, the content transmission is completed through the MBS relay. The latency of obtaining content from the collaborating SBSm includes the uplink latency from the collaborating SBSm to the MBS, the downlink latency from the MBS to the local SBS, and the transmission latency from the local SBS to the user. First, the uplink data rate from the collaborating SBSm to the MBS is defined as:

[0050]

[0051] The shortest uplink latency from the cooperative SBSm to the MBS is:

[0052]

[0053] in, Used to indicate whether the collaborative SBSm caches content i, n≠m, V is a constant, so that SBSs that do not contain the requested content i are ignored;

[0054] The latency for a user to retrieve cached content i from the collaborative SBSm is given as:

[0055]

[0056] S33: Based on the latency expressions and request probabilities of the three request methods, the average latency for users to obtain content is:

[0057]

[0058] After obtaining the pre-cache set with the highest hit rate, the cache placement objective function that minimizes latency is defined as:

[0059] P2:

[0060] stC1:

[0061] C2:

[0062] Among them, optimization variables This refers to the caching strategy, including the caching status of content i in SBS; when n=0, it is stored in MBS; C1 represents the binary cache variable c. i,n Constraints; C2 represents the cache capacity constraints for SBS and MBS, M n M0 and M0 represent the storage capacities of SBSn and MBS, respectively;

[0063] S34: Decouple problem P2 into two sub-problems: the cache placement problem and the delivery optimization problem, simplifying the optimal delivery problem to P3:

[0064] P3:

[0065] The goal of problem P3 is to find the optimal cooperating SBS m for SBS n. In solving the optimal delivery problem, using the prior information c from the caching strategy, the optimal delivery is dynamically decided based on the transmission rate and channel state of its cooperating SBS, thus completing the cooperative sharing of content i. For a given set M... n,m In each cache decision, the SBS calculates the uplink data rate and transmission delay t from the SBS to the MBS. m If the solution delay in this iteration is greater than that in the previous iteration, then let m... n,i =1, otherwise m n,i =0, until the collaborative SBS with the minimum delivery delay is found. This is the optimal delivery strategy;

[0066] S35: Under the premise of a priori optimal delivery strategy, a fixed delivery strategy m will be used to optimize the caching strategy; for the caching strategy under a known delivery strategy, it simplifies to problem P4:

[0067] P4:

[0068] stC1:

[0069] C2:

[0070] A caching strategy based on particle swarm optimization is adopted, based on a known pre-cache set. Minimize the average latency for users to access content, given delivery strategy m;

[0071] A feasible solution to the particle representation caching strategy is denoted by c, where each particle has a position denoted by x. c This means that the fitness function is the average time delay that needs to be optimized:

[0072]

[0073] At the same time, the local optimal solution for each particle is defined as follows: The optimal solution obtained by particle swarm optimization According to particle swarm optimization and Update the velocity v of each particle c via v c The position of each particle is updated to obtain the next generation of the particle swarm, and the velocity of the particles is v. c and position x c The updated formula is:

[0074]

[0075]

[0076] Where a is the coefficient of the iteration speed in iteration t; ψ1 and ψ2 are the weights of the local optimum and the global optimum; ε1 and ε2 are two uniform random variables in [0,1]; the algorithm iterates until the fitness converges, thereby obtaining the global optimum;

[0077] S36: To obtain a caching strategy that minimizes latency, delivery and cache placement are jointly optimized using an alternating iterative method. In the ηth iteration, the delivery strategy m(η-1) obtained in the η-1th iteration is used to obtain the optimized caching strategy c(η) according to the particle swarm optimization algorithm. Based on the obtained c(η), the updated delivery strategy m(η) is determined using the optimal delivery algorithm until the maximum number of iterations is reached.

[0078] The beneficial effects of this invention are as follows:

[0079] To address the challenge of unpredictable user preferences, this invention employs a hierarchical content preference method based on LSTM to predict user historical information. It derives user request probability information from the user's preference type and the preferred content within that type. Based on this request information and the probability of requests under different request methods, a latency-minimizing objective function is constructed. This is followed by utilizing edge pre-caching to maximize the hit rate, resulting in a pre-cached content set that maximizes the hit rate. Finally, considering collaborative scenarios, a collaborative caching strategy that minimizes user content retrieval latency is derived. Compared to traditional strategies, this invention effectively improves user experience and provides more flexible management of edge caching resources. It demonstrates superior performance in user-centric heterogeneous cellular networks, effectively addressing the challenge of meeting personalized user preference needs.

[0080] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0081] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0082] Figure 1 This is a network system architecture diagram of the present invention;

[0083] Figure 2 This is a schematic diagram of the collaborative cache delivery method of the present invention. Detailed Implementation

[0084] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0085] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0086] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0087] Please see Figures 1-2 This invention provides a hierarchical preference-aware video content caching method, specifically comprising the following steps:

[0088] Step 1: Construct a hierarchical user personal preference feature model based on video type preferences and content ranking for that type;

[0089] Step 1.1: Divide the content in the content library into G types, using... This indicates that user u has its own preferences for each type of content, and a preference function r(u,g) is defined to represent user u's preference for type g:

[0090]

[0091] Among them, h u For user u's historical access content information, p(g|h u ) is the conditional feature probability, which is the probability that the historical information contains content of type g; p(g) is the unconditional feature probability, which represents the probability that the entire content set contains content of type g.

[0092] Step 1.2: The probability α of user u's preference for content i i,u The preference for content i is jointly determined by the content type preference and the content preference within that type, P(f). Therefore, the preference of user u for content i is defined through a hierarchical structure based on type preferences:

[0093]

[0094] Where r(u,g) represents user u's preference for type g; P(f) represents content i's preference within that type; and χ represents the weighting factor of the ranking within the type on the probability of user preference. α i,u The value of is between [0,1]. The larger the value, the greater the probability that user u requests content i.

[0095] Step 1.3: A user's preference for video content is determined by both their preference for that content type and the content's ranking within that type. Specifically, user u's preference for type g depends on the number of videos rated by the user within that type, which can be calculated using the request probability p(g) of the set of rated videos containing type g, and the request probability p(g|h) of the set of videos containing type g in the user's historical information. u The user u's preference P(f) for content in type g cannot be directly obtained, so we consider using LSTM for time series prediction.

[0096] The LSTM network structure includes input gate x t Forgotten Gate t Output gate y t These gates can decide whether to retain or discard data based on its importance in the sequence. Within time t, the LSTM has L inputs x(t) = [g...]. u (t),g u (t+1),...,g u [(t+L-1)], where L is the time window, which can be determined as the number of input data samples for each unit of the LSTM. Meanwhile, the ideal output of the LSTM is y(t) = g u (t+L), where t=1,2,...,T d T d It represents the total number of data samples. Define g. u (t+L) represents the prediction output of the LSTM at time t. The optimization objective during prediction is to optimize the weights and biases of the LSTM to minimize the root mean square error (RMSE), where RMSE is defined as:

[0097]

[0098] Where ω represents the number of training iterations.

[0099] Step 2: Construct a pre-caching model that maximizes the system hit rate;

[0100] Step 2.1: When the content requested by a user is not on the edge server, it needs to be retrieved from the cloud server. If this content is frequently requested, each request will require transmission from the cloud server to the MBS before being sent to the user, resulting in significant backhaul redundancy. Therefore, the caching system needs to meet the user's edge cache hit rate requirement, i.e., the probability that the content requested by the user is cached on the edge server. Therefore, the pre-caching objective function that maximizes the hit rate within the MBS coverage area is defined as:

[0101] P1:

[0102] stC1:

[0103] C2:

[0104] Among them, optimization variables The pre-cached set is created by selecting high-hit-rate content from cloud servers for the MBS coverage area and placing it into the pre-cached set. In this way, the hit rate within the MBS coverage area can be improved, while reducing the difficulty of content selection for caching strategies; α i,u β represents the probability of user u's preference for content i, which can be predicted by LSTM; i Represents a pre-cached set binary variable, β i =1 That is, content i belongs to the pre-cached set. β i =0 C1 indicates that each piece of content can only appear once in the pre-cache set; C2 indicates the constraint on the capacity of the pre-cache set, s i The size of content i;

[0105] Step 2.2: To solve for the pre-cached content sets of MBS and SBSn in P1 use This represents the complement set of the pre-cached content, i.e. This refers to all content stored in the cloud server. A two-sided exchange of the pre-cached set is performed using exchange blocking pairs. The definition of an exchange blocking pair is given as follows: Given a pair (i,j), where... and Then (i,j) satisfies When the time is a blocking pair for swapping, That is, when swapping content i in the pre-cache set and content j in the non-pre-cache set, the sets Hit rate If an additional pair (i,j) is added, then (i,j) is called a blocking pair. A blocking pair means that (i,j) is preventing the pre-cached set from moving towards a higher hit rate. The positions of (i,j) in the set need to be swapped (until no more blocking pairs are found), thereby improving the hit rate and yielding the pre-caching strategy.

[0106] Step 3: Combine the request probability and latency information under different user access methods to build a minimum latency model, select an appropriate caching strategy for the pre-cached content, so as to minimize the average latency of user request content.

[0107] Step 3.1: During the content request phase, if the SBS has cached the content, user u can directly retrieve the requested content from the local SBS via the wireless communication link, taking into account the user's content preferences and the caching status of the content on the local SBS. i,n The probability that user u obtains content i from SBSn / MBS (n=0) can be obtained as follows:

[0108] p u,n,i =α i,u c i,n

[0109] Where, α i,u c represents the probability that user u prefers content i; i,n This is a binary cache variable for content i.

[0110] When a user cannot obtain the requested content from SBS and MBS, the local SBS will retrieve the requested content from other SBSs. Therefore, the probability that SBSn obtains the requested content i from the collaborating SBSm can be calculated as follows:

[0111]

[0112] Among them, M n,m This represents the set of cooperative SBSm around SBSn that satisfy the transmission conditions, and the channel condition between SBSm and MBS is log(1+SNR). m,0 When )>E, m∈M n,m E represents the minimum uplink capacity requirement; This is used to determine whether the cooperating SBSm (n≠m) of SBSn has cached content i. If the value is true, it means that SBSm has cached content i; otherwise, it has not been cached.

[0113] Step 3.2: By optimizing cache placement during the caching phase, users can independently request their preferred content. To achieve a better user experience, latency is used as a performance metric for user content requests. Therefore, transmission latency is calculated for different request methods.

[0114] When a user retrieves requested content from the local SBS, the requested content can be transmitted via a wireless link. The data rate from SBS n to user u can be calculated as follows:

[0115]

[0116] Among them, B SBS The channel bandwidth from SBS to the user; g n,u It is the channel gain between SBSn and user u, g n,u Depend on Given, where hn,u It is a random variable that follows a Gaussian distribution (0,1). It is the path loss constant, d n,u P is the distance between SBSn and user u, where ν is the path loss exponent; SBS This is the transmission power of SBS; It is co-channel interference from other SBSs; σ 2 It is the power of additive white Gaussian noise.

[0117] The latency for user u to download content i from local SBSn is:

[0118]

[0119] Among them, s i,u It is the size of the video content i requested by user u.

[0120] Similarly, the downlink transmission rate from MBS to SBSn is:

[0121]

[0122] Among them, B MBS The channel bandwidth from MBS to SBS; g 0,n This is the channel gain of MBS; P MBS This is the maximum transmit power of the MBS.

[0123] When a user does not receive content i from the local SBS, but content i is stored in the MBS, the latency for user u to download the requested content i from the MBS is:

[0124]

[0125] When the collaborating SBS m provides requested content to a user, content transmission needs to be completed via the MBS relay. Therefore, the latency for obtaining content from the collaborating SBS m includes the uplink latency from the collaborating SBS m to the MBS, the downlink latency from the MBS to the local SBS, and the transmission latency from the local SBS to the user. First, the uplink data rate from the collaborating SBS m to the MBS is defined as:

[0126]

[0127] The shortest uplink latency from the cooperative SBSm to the MBS is:

[0128]

[0129] in, V is used to indicate whether the collaborative SBSm (n≠m) caches content i; V is a large constant that makes SBSs that do not contain requested content i ignored.

[0130] Therefore, the latency for a user to retrieve cached content i from the collaborative SBSm is given as:

[0131]

[0132] Step 3.3: Therefore, based on the latency expressions and request probabilities of the three request methods mentioned above, the average latency for users to obtain content is:

[0133]

[0134] After obtaining the pre-cached set with the highest hit rate, in order to implement the caching strategy that minimizes the average user latency, it is necessary to consider not only whether the pre-cached content is placed in MBS or SBS, but also the collaborative delivery of SBS under collaborative caching, so that users can complete content requests in the best delivery manner. Therefore, the cache placement objective function that minimizes latency is defined as follows:

[0135] P2:

[0136] stC1:

[0137] C2

[0138] Among them, optimization variables This refers to the caching strategy, including the caching status of content i in SBS (when n=0, it is stored in MBS); C1 represents the binary cache variable c. i,n Constraints; C2 represents the cache capacity constraints for SBS and MBS, M n M0 and M0 represent the storage capacities of SBSn and MBS, respectively.

[0139] Step 3.4: Since problem P2 is a nested binary linear programming problem containing two objective functions—the outer one being a caching strategy to minimize latency, and the inner one being a delivery strategy for optimal delivery—problem P2 is decoupled into two sub-problems: the cache placement problem and the delivery optimization problem. The optimal delivery problem is then simplified to P3:

[0140] P3:

[0141] The goal of problem P3 is to find the optimal cooperating SBSm for SBSn. Therefore, when solving the optimal delivery problem, by utilizing the prior information c of the caching strategy, the optimal delivery can be dynamically decided based on the transmission rate of its cooperating SBS and the channel state, thereby completing the cooperative sharing of content i. For set M... n,m In each cache decision, the SBS calculates the uplink data rate and transmission delay t from the SBS to the MBS. mIf the solution delay in this iteration is greater than that in the previous iteration, then let m... n,i =1, otherwise m n,i =0, until the collaborative SBS with the minimum delivery delay is found. This is the optimal delivery strategy.

[0142] Step 3.5: Under the premise of a priori optimal delivery strategy, a fixed delivery strategy m will be used to optimize the caching strategy. For a caching strategy with a known delivery strategy, this can be simplified to problem P4:

[0143] P4:

[0144] stC1:

[0145] C2:

[0146] However, the latency varies when content is placed in different edge caching entities; specifically, the content transmission latency depends on whether the collaborative SBS is stored. Therefore, the above problem may not be solvable efficiently in polynomial time. Within the constraints, there are an infinite number of possible values ​​to determine the optimal solution. Therefore, a particle swarm optimization (PSO)-based caching strategy is adopted, given a known pre-cached set... Minimize the average latency for users to access content, given the delivery strategy m.

[0147] A feasible solution to the particle representation caching strategy is denoted by c, where each particle has a position denoted by x. c This means that the fitness function is the average time delay that needs to be optimized:

[0148]

[0149] At the same time, the local optimal solution for each particle is defined as follows: The optimal solution obtained by particle swarm optimization According to particle swarm optimization and Update the velocity v of each particle c via v c The position of each particle is updated to obtain the next generation of the particle swarm, and the velocity of the particles is v. c and position x c The updated formula is:

[0150]

[0151]

[0152] Where a is the coefficient of the iteration speed in iteration t; ψ1 and ψ2 are the weights of the local optimum and the global optimum, respectively; ε1 and ε2 are two uniform random variables in [0,1]. The algorithm iterates until the fitness converges, thus obtaining the global optimum.

[0153] Step 3.6: Since content delivery and cache placement are interdependent, the optimal content delivery method for P4 needs to be considered when solving the cache placement problem of P2, and the caching strategy of P2 needs to be considered when solving the delivery strategy of P4. Therefore, in order to obtain a caching strategy that minimizes latency, it is necessary to use an alternating iterative method to jointly optimize delivery and cache placement. In the ηth iteration, the delivery strategy m(η-1) obtained in the (η-1)th iteration is used to obtain the optimized caching strategy c(η) according to the particle swarm optimization algorithm. Based on the obtained c(η), the optimal delivery algorithm is used to determine the updated delivery strategy m(η) until the maximum number of iterations is reached.

[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A hierarchical preference-aware video content caching method, characterized in that: The method includes the following steps: S1: Construct a hierarchical user personal preference feature model based on video type preferences and content ranking for that type; specifically including the following steps: S11: Divide the content I in the content library into G types. User u has their own preference for each type of content. Define a preference function r(u,g) to represent user u's preference for type g: Among them, h u For user u's historical access content information, p(g|h u ) is the conditional feature probability, which is the probability that the historical information contains content of type g; p(g) is the unconditional feature probability, which represents the probability that the entire content set contains content of type g. S12: The probability α of user u's preference for content i i,u The preference of user u for content i is determined by both content type preference and content preference within that type, P(f). This preference is defined through a hierarchical structure based on type preferences. Where r(u,g) represents user u's preference for type g; P(f) represents content i's preference for that type; χ represents the weighting factor of the ranking within the type on the probability of user preference; α i,u The value of is between [0,1]. The larger the value, the greater the probability that user u requests content i. S13: A user's preference for video content is determined by both their preference for that content type and the content's ranking within that type. Specifically, user u's preference for type g depends on the number of videos in that type that have been rated by the user, expressed as a probability p(g) of requesting a set of rated videos containing type g, and the probability p(g|h) of requesting a set of videos containing type g in the user's historical information. u The user u's preference P(f) for content in type g cannot be directly obtained, so LSTM for time series prediction is used for prediction. The LSTM network structure includes input gate x t Forgotten Gate t Output gate y t These gates decide whether to retain or discard data based on the importance of the data in the sequence; within time t, the LSTM has L inputs x(t) = [g u (t),g u (t+1),...,g u [(t+L-1)], where L is the time window, defined as the number of input data samples for each unit of the LSTM; the ideal output of the LSTM is y(t) = g u (t+L), where t=1,2,...,T d T d It is the total number of data samples; define g. u (t+L) represents the prediction output of the LSTM at time t. The optimization objective during prediction is to optimize the weights and biases of the LSTM to minimize the root mean square error (RMSE), where RMSE is defined as: Where ω represents the number of training iterations; S2: Construct a pre-caching model that maximizes the system hit rate; specifically including the following steps: S21: When the content requested by a user is not on the edge server, it needs to be retrieved from the cloud server. If this content is frequently requested by users, each request will have to be transmitted from the cloud server to MBS before being sent to the user, resulting in significant backhaul transmission redundancy. The caching system needs to meet the user's edge cache hit rate requirement, i.e., the probability that the content requested by the user is cached by the edge server. The pre-caching objective function that maximizes the hit rate within the MBS coverage area is defined as: Among them, optimization variables The pre-cached set is created by selecting high-hit-rate content from cloud servers for the MBS coverage area and placing it into the pre-cached set. In this way, the hit rate within the MBS coverage area is improved, while the content selection difficulty of the caching strategy is reduced; α i,u β represents the probability of user u's preference for content i, obtained by LSTM prediction; i Represents a pre-cached set binary variable, β i =1 That is, content i belongs to the pre-cached set. β i =0 C1 indicates that each piece of content can only appear once in the pre-cache set; C2 indicates the constraint on the capacity of the pre-cache set, s i The size of content i; S22: To solve for the pre-cached content sets of MBS and SBSn in P1 use This represents the complement set of the pre-cached content, i.e. Given all content stored in the cloud server; utilize swap blocking pairs to complete a bilateral swap of the pre-cached set, and define a swap blocking pair as follows: Given a pair (i,j), where and Then (i,j) satisfies When the time is a blocking pair for swapping, That is, when swapping content i in the pre-cache set and content j in the non-pre-cache set, the sets Hit rate If an additional pair (i,j) is added, then (i,j) is called a blocking pair. A blocking pair means that (i,j) is preventing the pre-cached set from moving towards a higher hit rate. The positions of (i,j) in the set need to be swapped until no more blocking pairs are found, thus improving the hit rate and yielding the pre-caching strategy. S3: Combining request probability and latency information under different user access methods, construct a latency minimization model, select an appropriate caching strategy for the pre-cached content, and minimize the average latency of user requests; specifically including the following steps: S31: During the content request phase, if the SBS has cached the content, the user u retrieves the requested content directly from the local SBS via the wireless communication link, taking into account the user's content preferences and the caching status of the content on the local SBS. i,n The probability that user u obtains content i from SBSn / MBS is: p u,n,i =α i,u c i,n Where, α i,u c represents the probability that user u prefers content i; i,n Let i be a binary cache variable for content, and n = 0; When a user cannot obtain the requested content from SBS and MBS, the local SBS retrieves the requested content from other SBSs. The probability that SBSn obtains the requested content i from the collaborating SBS m is: Among them, M n,m This represents the set of cooperative SBSsm around SBS n that satisfy the transmission conditions, and the channel condition log(1+SNR) between SBS m and MBS. m,0 When )>E, m∈M n,m E represents the minimum uplink capacity requirement; This is used to determine whether the cooperating SBS m (n≠m) of SBS n has cached content i. If the value is 'i', it means that the SBSm cached content 'i'; otherwise, it is not cached S32: By optimizing cache placement during the caching stage, users can independently request their preferred content; latency is used as a performance indicator for user content requests, and transmission latency is calculated for different request methods. When a user retrieves requested content from the local SBS, the requested content is transmitted via the wireless link. The data rate from SBSn to user u is calculated as follows: Among them, B SBS The channel bandwidth from SBS to the user; g n,u It is the channel gain between SBSn and user u, g n,u Depend on Given, where h n,u It is a random variable that follows a Gaussian distribution (0,1). It is the path loss constant, d n,u P is the distance between SBSn and user u, where ν is the path loss exponent; SBS This is the transmission power of SBS; It is co-channel interference from other SBSs; σ 2 It is the power of additive white Gaussian noise; The latency for user u to download content i from local SBSn is: Among them, s i,u It is the size of the video content i requested by user u; The downlink transmission rate from MBS to SBS n is: Among them, B MBS The channel bandwidth from MBS to SBS; g 0,n This is the channel gain of MBS; P MBS This is the maximum transmit power of the MBS; When a user fails to request content i on the local SBS, but content i is stored in the MBS, the latency for user u to download the requested content i from the MBS is given as: When the cooperating SBS m provides requested content to the user, the content transmission is completed through the MBS relay. The latency of obtaining content from the cooperating SBS m includes the uplink latency from the cooperating SBS m to the MBS, the downlink latency from the MBS to the local SBS, and the transmission latency from the local SBS to the user. First, the uplink data rate from the cooperating SBS m to the MBS is defined as: The shortest uplink latency from the cooperating SBS m to the MBS is: in, Used to indicate whether the collaborative SBSm caches content i, n≠m, V is a constant, so that SBSs that do not contain the requested content i are ignored; The latency for a user to retrieve cached content i from the collaborative SBS m is given as: S33: Based on the latency expressions and request probabilities of the three request methods, the average latency for users to obtain content is: After obtaining the pre-cache set with the highest hit rate, the cache placement objective function that minimizes latency is defined as: Among them, optimization variables This refers to the caching strategy, including the caching status of content i in SBS; when n=0, it is stored in MBS; C1 represents the binary cache variable c. i,n Constraints; C2 represents the cache capacity constraints for SBS and MBS, M n M0 and M0 represent the storage capacities of SBSn and MBS, respectively; S34: Decouple problem P2 into two sub-problems: the cache placement problem and the delivery optimization problem, simplifying the optimal delivery problem to P3: The goal of problem P3 is to find the optimal cooperating SBSm for SBS n. When solving the optimal delivery problem, using the prior information c from the caching strategy, the optimal delivery is dynamically decided based on the transmission rate and channel state of its cooperating SBSs, thus completing the cooperative sharing of content i. For a given set M... n,m In each cache decision, the SBS calculates the uplink data rate and transmission delay t from the SBS to the MBS. m If the solution delay in this iteration is greater than that in the previous iteration, then let m... n,i =1, otherwise m n,i =0, until the collaborative SBS with the minimum delivery delay is found. This is the optimal delivery strategy; S35: Under the premise of a priori optimal delivery strategy, a fixed delivery strategy m will be used to optimize the caching strategy; for the caching strategy under a known delivery strategy, it simplifies to problem P4: A caching strategy based on particle swarm optimization is adopted, based on a known pre-cache set. Minimize the average latency for users to access content, given delivery strategy m; A feasible solution to the particle representation caching strategy is denoted by c, where each particle has a position denoted by x. c This means that the fitness function is the average time delay that needs to be optimized: At the same time, the local optimal solution for each particle is defined as follows: The optimal solution obtained by particle swarm optimization According to particle swarm optimization and Update the velocity v of each particle c via v c The position of each particle is updated to obtain the next generation of the particle swarm, and the velocity of the particles is v. c and position x c The updated formula is: Where a is the coefficient of the iteration speed in iteration t; ψ1 and ψ2 are the weights of the local optimum and the global optimum; ε1 and ε2 are two uniform random variables in [0,1]; the algorithm iterates until the fitness converges, thereby obtaining the global optimum; S36: To obtain a caching strategy that minimizes latency, delivery and cache placement are jointly optimized using an alternating iterative method. In the ηth iteration, the delivery strategy m(η-1) obtained in the η-1th iteration is used to obtain the optimized caching strategy c(η) according to the particle swarm optimization algorithm. Based on the obtained c(η), the updated delivery strategy m(η) is determined using the optimal delivery algorithm until the maximum number of iterations is reached.

Images