Social cache retrieval system fused with hotspot analysis

Abstract

The invention relates to the technical field of data caching, in particular to a social cache retrieval system fused with hotspot analysis, comprising: a popularity modeling unit for acquiring a hotspot event set and performing popularity modeling on each hotspot event in the hotspot event set to obtain popularity features of the hotspot events; and the social association unit is used for acquiring a user set and constructing a user-content bipartite graph according to the interaction behavior between the hot event set and the user set. Through the popularity modeling unit and the propagation calculation unit, popularity modeling can be performed on the hot events according to user behaviors and historical data, and the popularity prediction model is used for predicting the popularity of the hot events at the future moment, so that the events about to become hot spots can be identified in advance, and the accuracy of the hot events is improved. And distributing more cache resources for the events according to the prediction result.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of data caching, and particularly relates to a social caching retrieval system fusing hot spot analysis. BACKGROUND

[0002] Traditional systems usually do not predict and analyze the hotness of hot events, so they cannot identify events that will become hot in advance. As a result, when a hot event suddenly breaks out, the traditional system may not be able to load it into the cache in time, resulting in delays or cache misses when users request. In addition, the traditional system often uses fixed strategies to store data, such as a first-come-first-served cache strategy, or randomly deletes data when the cache capacity is limited. This approach does not take into account the heat and social relevance of events, and cannot optimize the caching of hot events, which may result in slow response to popular content. Moreover, the traditional system usually only manages the cache based on the most frequently accessed data, ignoring the influence of social networks and user behavior on content demand. For example, the system does not adjust the cache content based on user social relationships and behavior analysis, so it cannot determine which events will attract widespread attention in social networks, resulting in a low cache hit rate. In addition, because the traditional system does not manage and prioritize cache content, many unpopularity or irrelevant events still occupy cache space, which reduces the utilization efficiency of cache space and does not optimize resources, resulting in low performance. SUMMARY

[0003] The technical problem to be solved by the application is to overcome the above-mentioned shortcomings of the prior art, and to provide a social caching retrieval system fusing hot spot analysis.

[0004] The technical solution adopted to solve the above technical problems is: a social caching retrieval system fusing hot spot analysis, comprising:

[0005] a heat modeling unit, configured to obtain a set of hot events, and model the heat of each hot event in the set of hot events to obtain the heat characteristics of the hot event;

[0006] a social relevance unit, configured to obtain a set of users, construct a user-content bipartite graph according to the interaction behaviors between the set of hot events and the set of users, and determine the global social relevance of each hot event in the set of hot events according to the user-content bipartite graph;

[0007] a propagation calculation unit, configured to perform heat prediction on the heat feature of the hot event according to a pre-trained heat prediction model to obtain a future time heat feature of the hot event, and perform content propagation on the future time heat feature of the hot event according to an infectious disease model to obtain a diffusion probability of the hot event;

[0008] a cache calculation unit, configured to define a cache utility function according to the global social correlation degree, the future time heat feature and the diffusion probability of the hot event, and determine a cache utility value of each hot event in the hot event set according to the cache utility function.

[0009] Preferably, the system further comprises:

[0010] a social cache unit, configured to determine a hot event list and a non-hot event list according to the cache utility value of the hot event, store the hot event list into a local cache, and store the non-hot event list into each cache node in a cache cluster.

[0011] a search request unit, configured to obtain a search request, match the search request with the hot event in the local cache, and if no match is successful, send a query request to a cache node with the highest cache utility value within a preset hop number according to a distributed hash table.

[0012] Preferably, the heat modeling on each hot event in the hot event set to obtain the heat feature of the hot event comprises:

[0013] counting the hot event access times according to a preset time sliding window;

[0014] obtaining a total forwarding quantity of the hot event of the user set according to the preset time sliding window, and determining a social influence weight of the user according to the total forwarding quantity of the hot event of the user set;

[0015] obtaining an interaction behavior of the user within the time sliding window, and defining an interaction behavior intensity of the user within the time sliding window according to the interaction behavior of the user within the time sliding window, wherein the interaction behavior comprises likes, comments and forwarding, and the interaction behavior intensity is a weighted sum of the interaction behaviors;

[0016] determining the heat feature of the hot event according to the hot event access times, the social influence weight of the user interacting with the hot event and the interaction behavior intensity of the user interacting with the hot event.

[0017] Preferably, the calculation formula of the social influence weight of the user is as follows:

[0018]

[0019] wherein, represents the kth user u k in the user set U i represents the social influence weight of the ith hot event c k in the hot event set to the kth user u i in the user set, SC(U k ,c i ) represents the forwarding quantity of the ith hot event c i in the hot event set to the user set U i ;

[0020] The calculation formula of the hotness feature of the hot event is as follows:

[0021]

[0022] wherein, H i (t) represents the hotness feature of the hot event in the time sliding window t, λ represents an adjustable optimization weight, N access (c i ,t) represents the hot event access times of the user to the ith hot event c i in the hot event set, represents the total hot event access times of the user to the hot event set, I(u k ,c i ,t) represents the interaction behavior intensity of the user to the ith hot event c i in the hot event set in the time sliding window t, represents the maximum value function.

[0023] Preferably, the user-content bipartite graph comprises a user node set, a hot event node set, a connection edge set and an edge weight set, the user node corresponds to a user, the hot event node corresponds to a hot event, the connection edge is used for connecting the user node and the hot event node, when the user interacts with the hot event, the connection edge exists between the corresponding user node and the hot event node, and the edge weight of the connection edge is defined according to the interaction behavior intensity of the user and the hot event.

[0024] Preferably, the global social correlation degree of each hot event in the hot event set is determined according to the user-content bipartite graph, comprising:

[0025] An initial influence value is assigned to each node in the user-content bipartite graph, and the initial influence value is calculated according to the following formula:

[0026]

[0027] wherein PR (0) (k) represents the initial influence value of the kth node in the user-content bipartite graph, and N represents the number of nodes in the user-content bipartite graph;

[0028] A steady-state influence of each node in the user-content bipartite graph is solved according to a random walk simulation, and the steady-state influence is calculated according to the following formula:

[0029]

[0030] wherein PR (l+1) (k) represents the steady-state influence of the node in the l+1th iteration, d represents a damping factor, the damping factor represents a probability that the user continues the random walk, and the remaining 1-d represents a probability that the user randomly jumps to any node, B(k) represents a source node set of the node, M bk represents a transition probability of the node b to the node k, and A bk represents an edge weight value of the node b to the node k in the user-content bipartite graph;

[0031] The above operation is repeated until the steady-state influences after two iterations are less than a preset influence threshold value, so as to obtain a final influence threshold value;

[0032] The final steady-state influences of all nodes are normalized according to a maximum value and a minimum value of the final steady-state influences of the nodes, so as to obtain a global social correlation degree of the nodes, i.e., a global social correlation degree of the hot event.

[0033] Preferably, the heat prediction model adopts an LSTM network, and a calculation formula of the diffusion probability of the hot event is as follows:

[0034]

[0035] wherein P spread (c i ) represents the diffusion probability of the hot event, β represents a preset density coefficient, γ represents a preset attenuation factor, S i represents the global social correlation degree of the hot event, represents a heat feature at a future moment.

[0036] Preferably, a calculation formula of the cache utility value of the hot event is as follows:

[0037]

[0038] wherein, U i represents the cache utility value of the hot event, and a represents a preset weight factor, Size(c i ) represents the data size of the hot event.

[0039] The beneficial effects of the present application are as follows: (1) The present application can model the hotness of hot events according to user behavior and historical data through the hotness modeling unit and the propagation calculation unit, and predict the hotness of hot events at future time points using the hotness prediction model, so as to identify events that will become hot in advance, and allocate more cache resources to these events according to the prediction results, and the social association unit constructs a user-content bipartite graph based on the interaction between users and hot events, and calculates the global social association degree, so that the system can identify events that not only have hotness in time, but also have high association degree in the social network and can trigger a large number of user attention, thereby more accurately allocating the cache priority of hot events; (2) The retrieval request unit of the present application can quickly match with the locally cached hot event list, reduce query delay, and quickly return the result for the stored hot event, improve the user's retrieval experience, and when the local cache is not matched successfully, the system sends a query request to the cache node with the highest cache utility value in the cache cluster through the distributed hash table, to ensure the speed and efficiency of obtaining hot events from remote cache nodes, which avoids the bottleneck of frequently obtaining data from the database or other slow storage; (3) The present application defines a cache utility function according to the global social association degree, future time hotness characteristics and diffusion probability of the hot event through the cache calculation unit, which can accurately evaluate the cache value of each hot event, and this process can ensure that the hot events most likely to be retrieved are stored in the local cache first, improve the cache utilization rate and hit rate, and the social cache unit stores the hot events and non-hot events separately according to the cache utility value, the hot events are stored in the local cache, and the non-hot events are stored in the cluster cache, which can reduce the waste of cache space and ensure that the hot events can quickly respond to user requests, and by storing the hot event list in the local cache and storing the non-hot events in different nodes of the cache cluster, this distributed cache mode makes the system have good scalability, and as the hot events change continuously, the system can flexibly adjust the cache strategy and allocate the load between multiple cache nodes as needed, thereby adapting to the changes in user demand. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The system architecture diagram of the overall system in an embodiment of the present application.

[0041] Reference signs: 1, heat modeling unit; 2, social association unit; 3, propagation calculation unit; 4, cache calculation unit; 5, social cache unit; 6, retrieval request unit. DETAILED DESCRIPTION

[0042] Embodiment one, as shown in the figure, the present application proposes a kind of social cache retrieval system of fusion hot spot analysis, comprising: Figure 1

[0043] Heat modeling unit 1, heat modeling unit 1 is used to obtain hot event set, the heat of each hot event in hot event set is modeled, to obtain the heat characteristics of hot event;

[0044] Social association unit 2, social association unit 2 is used to obtain user set, according to the interactive behavior between hot event set and user set User-content bipartite graph is constructed, according to user-content bipartite graph determines the global social association degree of each hot event in hot event set;

[0045] Propagation calculation unit 3, propagation calculation unit 3 is used to carry out heat forecast according to the heat characteristics of hot event of pre-trained heat prediction model to hot event, to obtain the future time heat characteristics of hot event, according to infectious disease model, the future time heat characteristics of hot event is carried out content propagation, to obtain the diffusion probability of hot event;

[0046] Cache calculation unit 4, cache calculation unit 4 is used to define cache utility function according to the global social association degree, future time heat characteristics and diffusion probability of hot event, according to cache utility function determines the cache utility value of each hot event in hot event set.

[0047] ​In the present application, the heat refers to the degree of attention of an event, content or topic, which is usually measured by user interaction (such as clicking, commenting, sharing), and heat modeling is used to predict the degree of attention of an event or content by analyzing historical data and user behavior; the set of hot events refers to the set of events or topics that are widely discussed and focused on in the social platform at present; the bipartite graph is a graph structure, the nodes in the graph can be divided into two categories, users and contents. On social media platforms, there is a relationship between users and contents, such as user's like or comment behavior on a certain content, and the edge of the bipartite graph connects the user and the content they interact with (hot events); the heat prediction model is a machine learning or statistical model that aims to predict the future heat of hot events through historical data, for example, by predicting the heat change of an event in the future through user's historical behavior data and current heat features; the infectious disease model is inspired by the transmission model of infectious diseases in epidemiology, and the common one is SIR model (susceptible, infected, recovered), in social media, the spread of hot events is similar to the spread of infectious diseases, users spread the content to other users through interaction (such as forwarding, commenting), and the model calculates the diffusion probability of hot events by simulating this spread process, that is, the degree to which events will be focused on by more users in the future; the diffusion probability refers to the probability that a hot event will be focused on by more users at a certain time in the future.

[0048] In an optional embodiment, the system further comprises:

[0049] The social cache unit 5 is configured to determine the hot event list and the non-hot event list according to the cache utility value of the hot event, store the hot event list in the local cache, and store the non-hot event list in each cache node in the cache cluster.

[0050] The retrieval request unit 6 is configured to obtain a retrieval request, match the retrieval request with the hot events in the local cache, and if the matching is unsuccessful, send a query request to the cache node with the highest cache utility value within the preset hop number according to the distributed hash table.

[0051] It should be noted that the cache utility value is an indicator for measuring whether a hot event should be cached, which is usually based on the popularity of the hot event, the diffusion probability, user engagement and other factors. The higher the utility value, the more likely the event will be frequently accessed in the future, and therefore it is more worthy of being cached. The hot event list refers to a collection of hot events with high cache utility values. Since these events have received a lot of attention and are likely to continue to be queried or discussed by users, they are stored in the local cache for quick access. The cache utility value of the non-hot event list is lower, meaning that they have lower attention or are less likely to be accessed in the future. Therefore, they are not stored in the local cache but are scattered in the cache cluster to reduce the burden of the local cache. The local cache refers to the cache stored on a single server or device, which is used to quickly respond to user requests. Hot events are preferentially stored in the local cache due to their high access frequency, thereby improving retrieval efficiency. The cache cluster is a distributed cache system composed of multiple cache nodes. Non-hot events are stored in this cluster, which can reduce the burden of individual cache nodes and ensure system scalability and efficiency. The retrieval request refers to the query request issued by the user, who wants to obtain certain information or events. In this system, the retrieval request is a query for hot event information. Hot event matching in the local cache refers to when the user issues a query, the retrieval request unit first searches the local cache for hot events that meet the conditions. If the query is successful, the system will directly return the data in the cache, thereby improving response speed. The distributed hash table (DHT) is a distributed storage system that can scatter data across multiple nodes and provide efficient lookup and access mechanisms. In this system, DHT is used to distribute and manage cache data by hashing values to locate storage nodes. The preset hop count refers to the maximum number of nodes that the system allows to query when using the distributed hash table. The smaller the hop count, the faster the query response time. Conversely, the query time may be longer. The preset hop count is a parameter used to limit the query range to ensure query efficiency. The cache node with the highest cache utility value refers to the node with the highest cache utility value if there is no matching hot event in the local cache. The retrieval request unit will send a request to other cache nodes according to the distributed hash table. It will select the node with the highest cache utility value to query in order to improve the probability of retrieval success. The node with the highest utility value is selected because these nodes store events that are more likely to be currently popular, thereby increasing the chance of successful matching.

[0052] In the second embodiment, the social cache retrieval system based on hot event analysis is provided. Compared with the first embodiment, the second embodiment further comprises: modeling the hotness of each hot event in the hot event set to obtain the hotness feature of the hot event, including:

[0053] counting the hot event access times according to the preset time sliding window;

[0054] obtaining the total forwarding quantity of the hot event of the user set according to the preset time sliding window, and determining the social influence weight of the user according to the total forwarding quantity of the hot event of the user set;

[0055] obtaining the interaction behavior of the user in the time sliding window, and defining the interaction behavior intensity of the user in the time sliding window according to the interaction behavior of the user in the time sliding window, wherein the interaction behavior includes likes, comments and forwarding, and the interaction behavior intensity is the weighted sum of the interaction behavior;

[0056] determining the hotness feature of the hot event according to the hot event access times, the social influence weight of the user interacting with the hot event and the interaction behavior intensity of the user interacting with the hot event.

[0057] In the embodiment, the time sliding window is a dynamic time range used to collect and analyze data in a specific time period, and the size and sliding step of the time window are preset. As time goes on, the window will move forward continuously, so as to contain new data and exclude old data. The sliding window is widely used in processing and analyzing data with time sequence, such as the access times of hot events and the behavior data of users; the forwarding quantity reflects the propagation intensity of the user to the event, and the higher the forwarding quantity is, the wider the propagation range of the event is, which may have greater influence on other users.

[0058] In an optional embodiment, the calculation formula of the social influence weight of the user is as follows:

[0059]

[0060] wherein, represents the kth user u k in the user set U i , SC(u k ,c i ) represents the social influence weight of the kth user u k in the user set U to the ith hot event c i in the hot event set, SC(U,c i ) represents the total forwarding quantity of the user set U to the ith hot event c i in the hot event set;

[0061] The calculation formula of the heat feature of the hotspot event is as follows:

[0062]

[0063] H (t) = λ * N (t) + (1-λ) * I (t), where H i (t) represents the heat feature of the hotspot event in the time sliding window t, λ represents an adjustable optimization weight, N access (c i ,t) represents the number of times of access of the user to the i-th hotspot event c i in the set of hotspot events, I (t) represents the total number of times of access of the user to the set of hotspot events, I k (c i ,t) represents the interaction behavior intensity of the user to the i-th hotspot event c i in the set of hotspot events in the time sliding window t, max represents a maximum value function.

[0064] In an optional embodiment, the user-content bipartite graph includes a user node set, a hotspot event node set, a connection edge set and an edge weight set, the user node corresponds to a user, the hotspot event node corresponds to a hotspot event, the connection edge is used to connect the user node and the hotspot event node, when the user interacts with the hotspot event, the connection edge exists between the corresponding user node and the hotspot event node, and the edge weight of the connection edge is defined according to the interaction behavior intensity of the user and the hotspot event.

[0065] It should be noted that in the user-content bipartite graph, the structure of the graph is composed of two types of nodes (user nodes and hotspot event nodes) and the connection edges therebetween, the connection edge represents the interaction relationship between the user and the hotspot event, and the edge weight reflects the intensity of the interaction behavior. The interaction behavior intensity is the source of the edge weight, and the weight of the edge is defined through the behavior (such as likes, comments and forwards) of the user and the event, so as to reveal the attention and participation degree of the user to the event. This graph model is usually used for social network analysis, user behavior prediction and hotspot event propagation research.

[0066] In an optional embodiment, the global social correlation degree of each hotspot event in the set of hotspot events is determined according to the user-content bipartite graph, including:

[0067] An initial influence value is assigned to each node in the user-content bipartite graph, and the calculation formula of the initial influence value is as follows:

[0068]

[0069] wherein, PR (0)(k) represents the initial influence value of the kth node in the user-content bipartite graph, and N represents the number of nodes in the user-content bipartite graph;

[0070] The steady-state influence of each node in the user-content bipartite graph is solved according to the random walk simulation, and the calculation formula of the steady-state influence is as follows:

[0071]

[0072] Wherein, PR (l+1) (k) represents the steady-state influence of the node at the l+1th iteration, d represents a damping factor, and the damping factor represents the probability of the user continuing to randomly walk, and the remaining 1-d is the probability of randomly jumping to any node, B(k) represents a source node set of the node, and M bk represents the transition probability of the node b to the node k, and A bk represents the edge weight value of the node b to the node k in the user-content bipartite graph;

[0073] The above operation is repeated until the steady-state influence after two iterations is less than a preset influence threshold, so as to obtain a final influence threshold;

[0074] The final steady-state influence of the node is normalized according to the maximum value and the minimum value of the final steady-state influence of all nodes, so as to obtain the global social correlation degree of the node, that is, the global social correlation degree of the hot event.

[0075] It should be noted that the initial influence value is the influence distribution of each node (whether it is a user node or a content node) without iteration; the steady-state influence refers to the influence value of the node in the stable state after multiple iterations, and when the random walk model is used to calculate the influence of the node in the network, multiple iterations are usually required to converge to a steady-state value, which reflects the long-term influence of the node; the source node set refers to the set of all nodes connected to the current node, and for a certain node, its source node set contains all other nodes connected to it.

[0076] In an optional embodiment, the heat prediction model adopts an LSTM network, and the calculation formula of the diffusion probability of the hot event is as follows:

[0077]

[0078] Wherein, P spread (c i ) represents the diffusion probability of the hot event, β represents a preset density coefficient, γ represents a preset attenuation factor, S i represents the global social correlation degree of the hot event, represents the heat feature at the future moment.

[0079] In an alternative embodiment, the formula for calculating the cache utility value of a hotspot event is as follows:

[0080]

[0081] wherein U i represents the cache utility value of a hotspot event, a represents a preset weight factor, Size(c i ) represents the data size of the hotspot event.

[0082] The embodiments of the present application are described in detail above with reference to the accompanying drawings, but the present application is not limited thereto, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present application.

Claims

1. A social caching retrieval system integrating hotspot analysis, characterized in that, include: A heat modeling unit (1) is used to obtain a set of hot events and perform heat modeling on each hot event in the set of hot events to obtain the heat characteristics of the hot events. Social association unit (2), the social association unit (2) is used to obtain a user set, construct a user-content bipartite graph based on the interaction behavior between the hot event set and the user set, and determine the global social association degree of each hot event in the hot event set based on the user-content bipartite graph; The propagation calculation unit (3) is used to perform heat prediction on the heat characteristics of the hot event based on the pre-trained heat prediction model, so as to obtain the heat characteristics of the hot event at future time, and to perform content propagation on the heat characteristics of the hot event at future time according to the infectious disease model, so as to obtain the diffusion probability of the hot event. The cache calculation unit (4) is used to define a cache utility function based on the global social relevance, future time heat characteristics and diffusion probability of the hot event, and to determine the cache utility value of each hot event in the hot event set based on the cache utility function.

2. The social caching retrieval system integrating hotspot analysis according to claim 1, characterized in that, The system also includes: Social caching unit (5), the social caching unit (5) is used to determine the list of hot events and the list of non-hot events according to the caching utility value of the hot events, store the list of hot events in the local cache, and store the list of non-hot events in each cache node in the cache cluster; The retrieval request unit (6) is used to obtain a retrieval request and match the retrieval request with the hot events in the local cache. If no match is found, a query request is sent to the cache node with the highest cache utility value within a preset number of hops according to the distributed hash table.

3. The social caching retrieval system integrating hotspot analysis according to claim 2, characterized in that, Perform popularity modeling on each of the hot events in the set of hot events to obtain the popularity characteristics of the hot events, including: The number of visits to the hot events is counted according to a preset time sliding window. The total number of reposts of hot events in the user set is obtained according to a preset time sliding window, and the social influence weight of the user is determined according to the total number of reposts of hot events in the user set. The user's interactive behavior within the time sliding window is obtained, and the intensity of the user's interactive behavior within the time sliding window is defined based on the user's interactive behavior within the time sliding window. The interactive behavior includes liking, commenting, and forwarding, and the intensity of the interactive behavior is a weighted sum of the interactive behaviors. The popularity characteristics of a hot topic are determined based on the number of times the hot topic is accessed, the social influence weight of the users who interact with the hot topic, and the intensity of the interactive behavior of the users who interact with the hot topic.

4. The social caching retrieval system integrating hotspot analysis according to claim 3, characterized in that, The formula for calculating the user's social influence weight is as follows: in, Represents the k-th user u in the user set U. k For the i-th hot event c in the set of hot events i Social influence weight, SC(u k ,c i ) represents the k-th user u in the user set. k For the i-th hot event c in the set of hot events i The number of forwards, SC(U,c i ) represents the user set U and the i-th hot event c in the hot event set. i Total number of reposts; The formula for calculating the popularity characteristics of the trending events is as follows: Among them, H i (t) represents the popularity characteristics of hot events within a sliding time window t, λ represents the adjustable optimization weight, and N access (c i (t) represents the user's response to the i-th hot event c in the set of hot events. i Number of visits to trending events I(u) represents the total number of times a user accesses the hot events in the hot event set. k ,c i (t) represents the user's view of the i-th hot event c in the hot event set within the time sliding window t. i The intensity of interactive behavior, This represents the function for finding the maximum value.

5. A social caching retrieval system integrating hotspot analysis according to claim 4, characterized in that, The user-content bipartite graph includes a set of user nodes, a set of hot event nodes, a set of connecting edges, and a set of edge weights. The user nodes correspond to users, and the hot event nodes correspond to hot events. The connecting edges are used to connect the user nodes and the hot event nodes. When a user interacts with a hot event, there is a connecting edge between the corresponding user node and the hot event node. The edge weights of the connecting edges are defined according to the intensity of the interaction between the user and the hot event.

6. A social caching retrieval system integrating hotspot analysis according to claim 5, characterized in that, The global social relevance of each hot event in the hot event set is determined based on the user-content bipartite graph, including: An initial influence value is assigned to each node in the user-content bipartite graph, wherein the formula for calculating the initial influence value is as follows: Among them, PR (0) (k) represents the initial influence value of the k-th node in the user-content bipartite graph, and N represents the number of nodes in the user-content bipartite graph; The steady-state influence of each node in the user-content bipartite graph is determined using random walk simulation, and the formula for calculating the steady-state influence is as follows: Among them, PR (l+1) (k) represents the steady-state influence of a node in the (l+1)th iteration, d represents the damping factor, and the damping factor represents the probability that the user continues the random walk. The remainder (1-d) is the probability of randomly jumping to any node. B(k) represents the set of source nodes of the node, and M bk Let represent the transition probability from node b to k, and A bk Let $\mathbf$ represent the edge weight from node $b$ to node $k$ in the user-content bipartite graph. Repeat the above operation until the steady-state influence after two iterations is less than the preset influence threshold, so as to obtain the final influence threshold. The final steady-state influence of a node is normalized based on the maximum and minimum values ​​of the final steady-state influence of all nodes to obtain the global social correlation of the node, which is the global social correlation of the hot topic event.

7. A social caching retrieval system integrating hotspot analysis according to claim 6, characterized in that, The heat prediction model uses an LSTM network, and the formula for calculating the spread probability of the hot topic event is as follows: Among them, P spread (c i ) represents the diffusion probability of a hot topic event, β represents the preset density coefficient, γ represents the preset decay factor, and S i This indicates the overall social relevance of trending events. This indicates the future popularity characteristics.

8. A social caching retrieval system integrating hotspot analysis according to claim 7, characterized in that, The formula for calculating the caching utility value of the hot events is as follows: Among them, U i This represents the caching utility value of hot events, α represents the preset weighting factor, and Size(c i ) indicates the data size of hot events.

Images