Video recommendation method and device, electronic equipment and nonvolatile storage medium

Abstract

The invention discloses a video recommendation method and device, electronic equipment and a nonvolatile storage medium. The method comprises the steps that user behavior data are collected, a user video preference matrix is determined according to the user behavior data, and the user video preference matrix is used for representing the degree of interest of different users on different types of videos; determining a clustering center according to the user video preference matrix in a mode of mapping a determination process of the clustering center to a foraging behavior of the spotted serow by adopting a spotted serow optimization algorithm; the users are clustered based on the clustering centers, a clustering cluster corresponding to each clustering center is determined, a clustering result is obtained, and video preference feature types corresponding to the users in the same clustering cluster are the same; and according to a clustering result, performing video recommendation on the user. According to the method and the device, the technical problems that a video recommendation result is not accurate enough and a recommendation effect is relatively poor in the prior art are solved.

Description

Technical Field

[0001] The present application relates to the field of big data analysis technology, and more specifically, to a video recommendation method, device, electronic device, and non-volatile storage medium. Background Art

[0002] Continuous breakthroughs in artificial intelligence and deep learning technologies have provided powerful tools for data processing and analysis. Against this backdrop, telecom network attack methods are also evolving. Fraudulent activities using AI technologies, such as face-swapping and voice-sounding, pose a serious threat to public property safety.

[0003] In order to meet this challenge and satisfy the public's demand for telecommunications security, anti-fraud publicity can be carried out by recommending anti-fraud videos to users. However, the video recommendation technology in related technologies has poor recommendation effects and has technical problems such as insufficient accuracy and personalization of recommendations.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] The embodiments of the present application provide a video recommendation method, device, electronic device, and non-volatile storage medium to at least solve the technical problems in the related art of inaccurate video recommendation results and poor recommendation effects.

[0006] According to one aspect of an embodiment of the present application, a video recommendation method is provided, comprising: collecting user behavior data, and determining a user video preference matrix based on the user behavior data, wherein the user video preference matrix is ​​used to characterize the degree of interest of different users in different types of videos; using a spotted hyena optimization algorithm, determining cluster centers based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of spotted hyenas, wherein each cluster center corresponds to a type of video preference feature of the user; clustering the users based on the cluster centers, determining the cluster clusters corresponding to each cluster center, and obtaining clustering results, wherein the video preference feature types corresponding to users in the same cluster cluster are the same; and recommending videos to the users based on the clustering results, wherein the types of videos recommended to users in different clusters are different.

[0007] Optionally, a spotted hyena optimization algorithm is used to determine the cluster center based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of the spotted hyena. The method includes: initializing the spotted hyena population using a chaotic mapping algorithm, wherein the position of each spotted hyena is represented as a set of candidate solutions for the cluster center, and the chaotic mapping algorithm includes: Liebovitch chaotic mapping; determining the fitness value corresponding to the position of the spotted hyena based on the user video preference matrix, wherein the fitness value is used to characterize the quality of the candidate solution of the cluster center in terms of whether the user can be divided into different clusters; based on the fitness value corresponding to the position of the spotted hyena, the cluster center is optimized and adjusted by simulating the process of the spotted hyena surrounding, hunting and attacking the prey to obtain the target solution of the cluster center.

[0008] Optionally, by imitating the process of spotted hyenas surrounding, hunting and attacking their prey, the cluster center is optimized and adjusted, including: in the surrounding stage, determining the current optimal solution of the cluster center, and updating the positions and fitness values ​​of other spotted hyenas based on the position of the spotted hyena with the current optimal solution, wherein the current optimal solution is determined based on the fitness value; in the hunting stage, dividing the spotted hyenas in the population into different role types, and adjusting the positions and fitness values ​​of the spotted hyenas according to the hunting strategies corresponding to each role type, wherein the role types include: leader, follower, and apprentice, the spotted hyena in the leader role is used to update its position by performing a global search, the spotted hyena in the follower role is used to update its position by performing a local search under the guidance of the leader, and the spotted hyena in the apprentice role is used to update its position by performing a random search; in the attacking stage, a nonlinear weight factor and a Levy flight search mechanism are used to adjust the positions and fitness values ​​of the spotted hyenas, wherein the nonlinear weight factor is used to balance the capabilities of the global search and the local search, and the Levy flight search mechanism is used to randomly introduce long-distance jumps and short-distance searches in the search and optimization process to avoid falling into local optimality.

[0009] Optionally, the method also includes: in the process of optimizing and adjusting, using the Cauchy mutation strategy and the reverse learning strategy to perturb the current optimal solution of the cluster center to adjust the position and fitness value of the spotted hyena corresponding to the current optimal solution, wherein the Cauchy mutation strategy is used to perturb the current optimal solution by generating Cauchy distribution random variables, and the reverse learning strategy is used to perturb by constructing a reverse sister solution corresponding to the current optimal solution.

[0010] Optionally, users are clustered based on cluster centers, and the cluster clusters corresponding to each cluster center are determined, and the clustering results obtained include: determining the cluster center as the initial cluster center of the fuzzy C-means clustering algorithm; determining the membership of each user to each cluster center based on the user video preference matrix, obtaining the membership matrix, and updating the cluster center based on the membership matrix; iteratively repeating the above process of determining the membership matrix and updating the cluster center until the objective function converges, wherein the objective function is used to characterize the clustering quality of clustering based on the current membership matrix and cluster center; and determining the clustering results based on the final membership matrix and cluster center.

[0011] Optionally, the method also includes: obtaining an abnormal behavior feature database, and determining that there is a security risk in the user behavior data when text data appears in the user behavior data and the text similarity between the text and the abnormal behavior-related text exceeds a preset similarity threshold, wherein the abnormal behavior feature database contains multiple abnormal behavior-related texts; or, determining a first behavior feature pattern corresponding to the user behavior data, and determining that there is a security risk in the user behavior data when the pattern similarity between the first behavior feature pattern and the second behavior feature pattern is lower than a second preset similarity threshold, wherein the second behavior feature pattern is a behavior feature pattern corresponding to the user's historical user behavior data; and when it is determined that there is a security risk in the user behavior data, sending an alarm message to the Internet of Things terminal device corresponding to the user.

[0012] Optionally, the method also includes: determining the time period corresponding to when collecting user behavior data, and the geographical location corresponding to the Internet of Things terminal device used by the user; determining the target abnormal behavior type corresponding to the geographical location within the time period, wherein the target abnormal behavior type is the type corresponding to the abnormal behavior that has occurred in the geographical location within the time period, and the frequency of occurrence exceeds a preset frequency threshold; obtaining a target video corresponding to the target abnormal behavior type, and recommending the target video to the user's Internet of Things terminal device, wherein the target video is used to remind the user to guard against abnormal behavior of the target abnormal behavior type.

[0013] According to another aspect of an embodiment of the present application, a video recommendation device is also provided, including: an acquisition module for acquiring user behavior data, and determining a user video preference matrix based on the user behavior data, wherein the user video preference matrix is ​​used to characterize the degree of interest of different users in different types of videos; an optimization module for using a spotted hyena optimization algorithm to determine cluster centers based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of spotted hyenas, wherein each cluster center corresponds to a type of video preference feature of the user; a clustering module for clustering users based on the cluster centers, determining the cluster clusters corresponding to each cluster center, and obtaining clustering results, wherein the video preference feature types corresponding to users in the same cluster cluster are the same; a recommendation module for recommending videos to users based on the clustering results, wherein the types of videos recommended to users in different clusters are different.

[0014] According to another aspect of the embodiments of the present application, an electronic device is provided, including: a memory and a processor, wherein the processor is configured to run a program stored in the memory, wherein the video recommendation method is executed when the program is run.

[0015] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided. The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the video recommendation method by running the computer program.

[0016] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, which implements the steps of the video recommendation method when executed by a processor.

[0017] In an embodiment of the present application, user behavior data is collected and a user video preference matrix is ​​determined based on the user behavior data, wherein the user video preference matrix is ​​used to characterize the degree of interest of different users in different types of videos; a spotted hyena optimization algorithm is used to determine cluster centers based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of spotted hyenas, wherein each cluster center corresponds to a type of video preference feature of the user; users are clustered based on the cluster centers, and cluster clusters corresponding to each cluster center are determined to obtain clustering results, wherein the video preference feature types corresponding to users in the same cluster cluster are the same; based on the clustering results, videos are recommended to users, wherein the types of videos recommended to users in different cluster clusters are different. By utilizing different Internet of Things devices to collect user behavior data for processing, users are clustered based on the improved spotted hyena optimization algorithm, and personalized video recommendations are made based on the clustering results, thereby achieving the purpose of improving the accuracy of personalized video recommendations, thereby solving the technical problems in the related art of inaccurate video recommendation results and poor recommendation effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 This is a hardware structure block diagram of a computer terminal (or electronic device) for implementing a method for video recommendation provided in an embodiment of the present application;

[0020] Figure 2 This is a schematic diagram of a video recommendation method according to an embodiment of the present application;

[0021] Figure 3 Schematic diagram of the architecture of an anti-fraud video recommendation system provided according to an embodiment of the present application;

[0022] Figure 4 This is a flow chart of a fuzzy C-means clustering recommendation algorithm based on an improved spotted hyena optimization algorithm provided in an embodiment of the present application;

[0023] Figure 5 3 is a structural diagram of a video recommendation device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] The video recommendation technology in the related art has poor recommendation effect and has technical problems such as insufficient accuracy and personalization of recommendations. In order to solve the above problems, the embodiments of the present application provide relevant solutions, which are described in detail below.

[0027] According to an embodiment of the present application, a method embodiment of video recommendation is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or electronic device) for implementing a video recommendation method. Figure 1As shown, the computer terminal 10 (or electronic device) may include one or more (illustrated as 102a, 102b, ..., 102n in the figure) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0029] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single independent processing module, or may be incorporated in whole or in part into any of the other components of the computer terminal 10 (or electronic device). As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0030] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the video recommendation method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above-mentioned video recommendation method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0031] The transmission device 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0032] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or electronic device).

[0033] In the above operating environment, the embodiment of the present application provides a video recommendation method. Figure 2 is a schematic diagram of a video recommendation method flow according to an embodiment of the present application, such as Figure 2 As shown, the method includes the following steps:

[0034] Step S202: collecting user behavior data and determining a user video preference matrix based on the user behavior data, wherein the user video preference matrix is ​​used to represent the degree of interest of different users in different types of videos;

[0035] Step S204: using the spotted hyena optimization algorithm, the cluster centers are determined based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of the spotted hyena, where each cluster center corresponds to a video preference feature type of the user;

[0036] Step S206: clustering the users based on the cluster centers, determining the clusters corresponding to each cluster center, and obtaining clustering results, wherein the video preference feature types corresponding to users in the same cluster are the same;

[0037] Step S208 : Recommending videos to users based on the clustering results, wherein different types of videos are recommended to users in different clusters.

[0038] Through the above steps, by utilizing different IoT devices to collect user behavior data for processing, users are clustered based on the improved spotted hyena optimization algorithm, and personalized video recommendations are made based on the clustering results, thereby achieving the goal of improving the accuracy of personalized video recommendations, thereby solving the technical problems of inaccurate video recommendation results and poor recommendation effects in related technologies.

[0039] The following further introduces the video recommendation method in steps S202 to S208 of the embodiment of the present application.

[0040] In the embodiment of the present application, the above-mentioned video recommendation method is illustrated by taking the recommendation of anti-fraud videos as an example.

[0041] In this embodiment, the video recommendation method in steps S202 to S208 can be combined with the following Figure 3 The anti-fraud video recommendation system shown in the following is executed. Figure 3 The video recommendation system in Figure 3 As shown in the figure, the system consists of an Internet of Things device module, a data collection and storage module, a telecommunications anti-fraud module, a user reminder module, an anti-fraud short video source module, a user anti-fraud model generation module, a recommendation module and a recommendation effect evaluation module. The system can use different Internet of Things devices to collect user-related data (for example, user behavior data) for processing. At the same time, based on the spotted hyena optimization algorithm, it uses a variety of methods including the lion group optimization algorithm to improve it, and applies it to the fuzzy C-means clustering algorithm to alleviate the cold start problem of the video recommendation process and improve the accuracy of user video recommendations. The details are introduced below.

[0042] The IoT device module is the source of user IoT data collection, including but not limited to: smart gateways / routers, smart home control centers, smart home energy management systems and terminal devices. Through these devices, user-related IoT data, such as user behavior data, can be collected, as shown below.

[0043] 1) The data that the smart gateway / router can collect includes but is not limited to: Device connection records: Recording information about devices connected to the home network, including device name, MAC address, IP address, etc.; Network traffic monitoring: Real-time monitoring of data traffic in the home network, including upload and download speeds, traffic size, flow direction, etc.

[0044] Users can view a list of devices connected to their home network through the smart router's management interface, allowing them to promptly identify and address unknown or unauthorized device connections. By monitoring network traffic, smart routers can identify unusual communication patterns, such as large amounts of data traffic to unknown external servers or frequent data transfers, which may be associated with telecom fraud. If the router detects a large amount of data traffic to unknown external servers, this may indicate that the home network is being used for illegal activities.

[0045] 2) The data that can be collected by the smart home control center (including smart door locks, smart cameras, smart sensors, etc.) includes but is not limited to: device connection status: recording the connection status and online time of all smart home devices; user operation records: records of user operations on devices through the control center, such as setting scheduled tasks and adjusting device parameters.

[0046] The Home Control Center contains the connection status and operation records of home devices. If the device connection status is abnormal or there are abnormal behaviors in the operation records (such as frequent changes to device settings, unauthorized remote access, etc.), there may be a risk of fraud.

[0047] 3) Smart home energy management systems can monitor energy consumption in the home in real time. If the energy consumption of certain appliances (such as TVs and computers) suddenly increases during a certain period of time, even though the household members are not using the appliances, this may indicate that someone has illegally entered the home and is using these appliances. This abnormal energy consumption pattern may be associated with fraudulent activity, as fraudsters may be exploiting home devices for illegal activities.

[0048] 4) Terminal devices, such as smartphones / smartwatches, can check call and text message logs to analyze for unusual call or text message behavior. Frequent text messages or calls from unknown numbers, or unusually short calls, could be signs of fraud. Users can also identify potential fraud risks by observing unusual phone behavior during use. For example, if the phone suddenly becomes unusually slow, battery drains rapidly, or unknown applications and icons appear, these could be signs of malware or fraud.

[0049] In an embodiment of the present application, the data of home Internet of Things devices can be collected by the NodeMCU module in the data collection and storage module and transmitted to the cloud server for storage. Specifically, the NodeMCU device can collect data of home Internet of Things devices, including router data, camera data (if supported or compatible with the camera), and some terminal devices (mobile phones, on the premise that they provide relevant data interfaces or can obtain data by other means) and other home Internet of Things devices, and then send these data to the cloud server for storage. Among them, the NodeMCU device establishes a connection with the cloud server via Wi-Fi, and uses data transmission protocols such as HTTP and MQTT to upload the processed data to the cloud server. After the server receives the data uploaded by the NodeMCU device, it stores it in a distributed database (such as MongoDB, Cassandra, etc.) or a relational database (such as MySQL, PostgreSQL, etc.) for subsequent data analysis and processing. At the same time, the cloud server provides interfaces such as RESTful API or GraphQL, allowing home users, third-party applications or service providers to access and query data stored on the cloud server after authentication.

[0050] Afterwards, the telecommunications anti-fraud module can analyze and process this data. When the risk of telecommunications fraud is identified, a message will be sent to the user reminder module to promptly remind the user. After the user reminder module collects reminders of possible fraud risks, it will promptly remind the user by calling and sending text messages. The specific steps are as follows.

[0051] In some embodiments of the present application, the method also includes the following steps: obtaining an abnormal behavior feature database, and determining that there is a security risk in the user behavior data when text data appears in the user behavior data and the text similarity between the text and the abnormal behavior-related text exceeds a preset similarity threshold, wherein the abnormal behavior feature database contains multiple abnormal behavior-related texts; or, determining a first behavior feature pattern corresponding to the user behavior data, and determining that there is a security risk in the user behavior data when the pattern similarity between the first behavior feature pattern and the second behavior feature pattern is lower than a second preset similarity threshold, wherein the second behavior feature pattern is the behavior feature pattern corresponding to the user's historical user behavior data; when it is determined that there is a security risk in the user behavior data, sending an alarm message to the Internet of Things terminal device corresponding to the user.

[0052] The initial stage involves collecting abnormal information (e.g., information on telecom fraud). This involves building a database of abnormal behavior features. This collection of abnormal information allows for the compilation of text related to abnormal behavior (e.g., text related to telecom fraud), including instances of fraudulent calls, language used to describe telecom fraud, media reports, and news about telecom fraud. During the data collection process, web crawler technology and search engines are used to collect text data on internet-based phone fraud. Corresponding rules are also created and features extracted. This paper uses natural language processing techniques to extract characteristic keywords from fraudulent material. Once a user's phone is answered and a call is received, requests begin to follow. After using voice confirmation technology to convert the caller's voice into text, the program uses the fraud detection criteria established in the previous stage to determine whether the call is fraudulent. The details are as follows.

[0053] During the abnormal information collection phase, text data is preprocessed, including removing unused characters and irrelevant values. To eliminate duplicates, text similarity analysis is used. If two text segments have an 80% similarity, one is considered a duplicate. Only fraud-related text is retained. Next, machine learning algorithms analyze and select the data. Specifically, the text data is vectorized to convert it into input that the mechanism knowledge algorithm can understand. First, keywords are extracted from all data using TF-IDF technology. The conversion rule assigns each word to a feature, then segments the text and eliminates stop words. If the keyword it represents appears in any text, the value of this feature is set to 1; otherwise, it is set to 0. In subsequent stages, machine learning techniques are used to train a model using these vectors. Feature extraction is then performed to generate a list of keywords (text related to abnormal behavior) extracted from the text data related to telecommunications fraud.

[0054] When new user behavior data is collected, the system matches and compares the textual data contained in the data with a generated keyword list. If the words contained in the data have a high degree of similarity or match with words in the keyword list, the system further analyzes the context and combination of these words to determine whether the data involves telecommunications fraud. Once the system determines that the data involves telecommunications fraud, it immediately triggers an early warning mechanism, promptly alerting users to the risk through SMS, email, or in-app notifications, and may provide relevant preventative advice or measures.

[0055] The system also meticulously identifies and analyzes other types of data generated by IoT devices, such as device operating status, usage habits, and environmental changes. For example, if the system detects unusual access or control of a home camera, even if the behavior doesn't directly constitute telecom fraud, it will immediately issue a security warning and advise the user to check their camera settings or take necessary network security measures. Similarly, if router data experiences unusual fluctuations, the system can quickly identify and alert the user to potential cyberattack risks.

[0056] On the other hand, after collecting the user's IoT data, the user's geographic location and other information can be combined to generate a user anti-fraud model. Specifically, the user anti-fraud model generation module contains IoT device data and geographic location data. Based on the data collected by IoT devices, a user personal model is generated. First, data is collected from smart gateways / routers, smart home control centers, smart home energy management systems, and terminal devices. These data may include: network traffic data (such as websites visited, applications used), device usage data (such as switching time, frequency of use), and energy consumption data (such as power consumption, power consumption time). At the same time, remove noise and outliers to ensure data quality. Standardize data of different dimensions to the same range. Extract features from the preprocessed data to generate a user personal model. Assign different weights according to the importance of the features. The formula is as follows:

[0057]

[0058] Among them, F is the fused feature, w i is the weight of the i-th feature, f i is the i-th feature.

[0059] In the telecommunications anti-fraud short video recommendation method and system, in addition to generating a personal model through IoT devices, geolocation technology can also be used to accurately identify the user's geographic location, thereby pushing anti-fraud videos closely related to the types and characteristics of fraud cases in that area. This precise push method ensures that users receive anti-fraud information that is most relevant to their living environment, improving the pertinence and practicality of the information. The following are the steps for generating a user anti-fraud model:

[0060] Step 1. Location feature extraction: for user u x , at its geographical location L x On the , collect the anti-fraud short video category set S that the user watched x , which is used as the user's initial anti-fraud preference feature P at this location L1 =(L x ,S x ).

[0061] Step 2. Cross-location feature integration: Repeat Step 1 for different geographical locations visited by the user within the set time range to summarize and integrate the user u x Based on the anti-fraud short video viewing data at these locations, we can construct the anti-fraud interest feature set P of users in different geographical locations. L2 ,...,P LM .

[0062] Step 3. Time-location preference analysis: Comprehensively analyze the behavior data of users watching anti-fraud short videos in different time periods and geographical locations. x When watching a specific type of anti-fraud short video multiple times in the same location and different time periods, the user's overall preference for the type can be evaluated by calculating the average attention or interest; if the user's viewing behavior of a certain type of short video is inconsistent in different time periods, a comprehensive evaluation can be conducted by combining the time weight or the latest behavior data. This process aims to build a global feature of the user's anti-fraud interest based on geographic location

[0063] Step 4. Build a global anti-fraud user profile: Repeat Steps 1 to 3 for all mobile users, extracting and integrating their anti-fraud interest data to construct a global user anti-fraud profile matrix P based on geographic location. This matrix details each user's preference for various anti-fraud short videos at different locations.

[0064] In summary, we can get the user anti-fraud model as

[0065] In addition, the anti-fraud short video source module in the system can collect different types of anti-fraud short videos and pass them to the recommendation module for use. In this embodiment, the sources of the anti-fraud short video source module are diversified to ensure the richness, authority and timeliness of the content, including but not limited to: official anti-fraud agencies, telecom operators and Internet companies, media organizations and self-media creators, victims and anti-fraud volunteers, etc. These sources together constitute a rich resource library of anti-fraud short videos, providing strong support for anti-fraud propaganda.

[0066] Based on various types of videos, the recommendation module in the system can make personalized recommendations based on the interests and preferences of different users. In this embodiment, the fuzzy C-means clustering algorithm can be used based on the improved spotted hyena optimization algorithm to recommend anti-fraud short videos to users. The following is a detailed introduction.

[0067] Figure 4 is a flow chart of a fuzzy C-means clustering recommendation algorithm based on an improved spotted hyena optimization algorithm provided in an embodiment of the present application, such as Figure 4As shown, in this embodiment, a hybrid strategy method including the Lion Group Optimization Algorithm is used to improve the Spotted Hyena Optimization Algorithm. Furthermore, the improved Spotted Hyena Optimization Algorithm is used to improve the Fuzzy C-means Clustering Algorithm to cluster users, thereby achieving more accurate video recommendations. The following first introduces the optimization principle and mechanism of the Spotted Hyena Algorithm.

[0068] The optimization principle of the spotted hyena optimization algorithm comes from the grassland status of spotted hyenas on the African grasslands and the process of their hunting prey.

[0069] During the encirclement phase, spotted hyenas rely on vision to quickly identify and locate their prey and surround them. The collaboration between individual spotted hyenas, based on social status and mutual trust, allows for a more agile and efficient encirclement. The hyena closest to the target prey is then considered the optimal solution. The remaining spotted hyenas adjust their positions based on this optimal solution, ultimately achieving the global optimal solution. The mathematical model is shown below:

[0070] D h =|B·P p (X)-P(X)|

[0071] P(x+1)=P p (x)-E·D h

[0072] Among them, D h is the distance between the prey and the spotted hyena, B is the swing factor, P p is the current prey position, P is the current hyena position, P(x+1) represents the position of the hyena in the next iteration, x is the number of iterations, and E is the convergence factor. The specific formulas for B and E are as follows:

[0073] B=2·rd1

[0074] E=2h·rd2-h

[0075] h=5-(iter *5 / X max )

[0076] Among them, rd1 and rd2 are random numbers in the interval [0,1], h is the control factor, h decreases linearly from 5 to 0, and the value range is [0,5], iter=1,2,...,X max , X max is the maximum number of iterations.

[0077] During the hunting phase, spotted hyenas group together based on a trustworthy hierarchy and their ability to accurately identify prey. During this process, the population defines the current optimal search individual, and the remaining spotted hyena search individuals are divided into a cluster, which collectively converges on the optimal search individual to update their position. This behavior is specifically defined as follows:

[0078] D h =|B·P h -P K |

[0079] P k =P h -E·D h

[0080] C h =P K +P K+1 +...+P K+N

[0081] Among them, P h The first optimal position of the spotted hyena is defined, and the positions of other spotted hyenas are P k , C h is a cluster of N optimal solutions, where N is the number of spotted hyenas, and N is calculated as follows:

[0082] N=count nos (P h ,P h+1 ,P h+2 ,...,(P h +M))

[0083] Where M is a random vector in the range [0.5, 1], nos defines the number of feasible solutions and calculates all candidate solutions. This process is similar to the optimal solution of the search space.

[0084] During the final stages of the hunt, the spotted hyena continues to update its position and attack its prey. The control factor h decreases linearly from 5 to 0 during the iterations, and the convergence factor E also decreases as h changes. When |E| < 1, the spotted hyena attacks its prey. This behavior is defined as follows:

[0085] P(x+1)=C h / N

[0086] Among them, P(x+1) maintains the optimal solution, C h Maintain the optimal solution cluster.

[0087] During the search for prey (global exploration), the spotted hyena is allowed to update the best search cluster C hWhen |E|>1, the hyenas will disperse from each other, search from places far away from the prey, and continue to look for and attack more suitable prey. This hunting behavior of spotted hyenas ensures the global search capability of the algorithm.

[0088] The spotted hyena optimization algorithm is inspired by the behavior of spotted hyenas when foraging, and searches for the optimal solution by simulating the cooperation and competition of spotted hyenas. However, the spotted hyena optimization algorithm in related technologies still has some shortcomings, including: 1) Low solution accuracy: The algorithm may not find the solution accurate enough to meet the needs of actual problems due to problems with the search strategy or parameter settings during the search for the optimal solution. 2) Slow convergence: The algorithm may converge slowly, and it may take a long time to find a solution close to the optimal solution. This is especially obvious when dealing with large-scale or complex problems, and may not meet the requirements of real-time or efficiency. 3) Easy to fall into local optimality: The spotted hyena optimization algorithm may rely too much on the current best solution during the search process, resulting in premature convergence to the local optimal solution and failure to find the global optimal solution.

[0089] To improve these deficiencies of the spotted hyena optimization algorithm in related art, the present invention improves the spotted hyena optimization algorithm in related art by using a hybrid strategy including the lion group optimization algorithm.

[0090] In an embodiment of the present application, the hunting mechanism of the spotted hyena optimization algorithm is improved using a lion group optimization algorithm. By introducing a hierarchical structure of leaders, followers, and apprentices, the algorithm can form a clearer search hierarchy. The leader (lion king / spotted hyena leader) is responsible for global search and discovers the potential area of ​​the optimal solution; the followers (lioness / female spotted hyenas) conduct local searches under the guidance of the leader and conduct detailed searches of the potential areas; the apprentices (lion cubs / spotted hyena cubs) explore new solution spaces through random searches to increase the diversity of the population. At the same time, the spotted hyena leader, with its high fitness value and search ability, can quickly discover the potential area of ​​the global optimal solution, thereby guiding the entire population to the optimal solution. Female spotted hyenas, searching under the guidance of the spotted hyena leader, can more accurately locate the optimal solution and reduce unnecessary searches. The random search of the spotted hyena cubs can explore new solution spaces and help the population avoid falling into local optimality.

[0091] The Lion Pride Optimization Algorithm (LPAO) is based on the theory of lion prides in nature. By observing and analyzing the diverse identities and behaviors of lion prides, and summarizing their collaborative hunting processes, the new swarm intelligence optimization algorithm, the Lion Pride Algorithm, was developed. Lion prides have a clear division of labor, primarily consisting of three groups: king lions, adult lionesses, and cubs. The corresponding mathematical model is as follows:

[0092] Suppose an optimization problem f(x) has a solution space dimension of D. The number of lions in the lion group is N. The maximum number of iterations of the algorithm is T. Each lion in the lion group can be represented by a D-dimensional vector:

[0093] x i =(x i1 ,x i2 ,...,x il ),1≤i≤N

[0094] Next, we will introduce the main update methods and parameter settings of the Lion Group Algorithm.

[0095] 1) Adult lion scale factor

[0096] The ratio of the number of adult lions (including both lion kings and lionesses) to the total number of lions in the pride, N, is the scaling factor for adult lions, denoted as β. Clearly, β∈(0,1). This scaling factor primarily adjusts the algorithm's global optimization and local search biases. The behavior of young lions is more complex and diverse than that of adult lions. Increasing the number of young lions can increase population diversity and variability, improving the algorithm's global detection capabilities and helping it escape local optima. Adult lions, on the other hand, prioritize local optimization. β is generally chosen to be less than 0.5.

[0097] 2) Maximum step length

[0098] The step size of individual movement should be appropriate. If the step size is too small, the convergence speed will be slow, making it difficult for the algorithm to converge to the optimal point. If the step size is too large, the displacement will be too fast, which may lead to insufficient search accuracy and miss the opportunity to find the optimal point. The step size is defined as follows:

[0099]

[0100] in, and They respectively represent the minimum mean and maximum mean of each vector dimension of the lion group in the area that does not exceed the lion's activity range.

[0101] During the optimization process, the update formulas for the lion king, young lion and lioness are different, as follows.

[0102] 1) Lion King update method

[0103] The lion king is the individual lion with the best objective function value in the entire population. Since the lion king occupies the optimal position in the current population, there is no need to conduct a large-scale search. Instead, based on the current optimal position, a small-scale search is performed in its own neighborhood to explore whether there is a better solution near itself. The update formula for the lion king is:

[0104] x k+1 =g k(1+γ||p k -g k ||)

[0105] Among them, x k+1 represents the global optimal individual of the offspring obtained by the lion king himself, g k is the global optimal individual of the kth generation population, p k represents the historical optimal position of the individual lion after k iterations. γ is a random number generated according to a normal distribution: γ~N(0,1).

[0106] 2) Lioness's update method

[0107] The lioness group is mainly responsible for hunting and teaching cubs through cooperation. During hunting, each lioness will randomly choose another lioness to cooperate with. The update formula of the lioness is:

[0108]

[0109] in, is the historical optimal position of the i-th lion after the k-th iteration cycle, is the historical optimal position of another lioness except the i-th lion, α f The perturbation factor is a dynamic parameter that decreases with the number of update iterations t and gradually approaches zero. Its function is to gradually transition the lioness's movement distance from large to small, taking into account the algorithm's needs for large-scale exploration and small-step local development.

[0110] 3) How to update the young lion

[0111] There are three ways to update a lion cub. The first formula involves following the lion king and exploring near the global optimal position. The second formula shows that it follows the lioness, obtaining her position information and learning to hunt. The third formula, drawing on the concept of elite reverse learning, involves some cubs moving away from the lion king. This increases the randomness of individual movements and the diversity of displacement directions within the population, preventing the lion group from becoming trapped in a local optimal position. The specific formulas are as follows:

[0112]

[0113] in, Represents the individual historical optimal position of the i-th lion cub after the k-th update iteration; is the historical optimal position of the lioness group at the kth generation; g -k It represents the reverse position of the optimal position of the population, as shown in the following formula:

[0114]

[0115] Similarly, a linear dynamic disturbance factor α is added to the update process of the lion cub, which decreases with time. c , which plays the role of lengthening or compressing the distance, making the lion cub's moving steps gradually change from large to small.

[0116] After improving the spotted hyena optimization algorithm by combining the lion group optimization algorithm, the specific update mechanism in the hunting phase of this application embodiment is shown in the following formula:

[0117] Spotted Hyena Leader (Leader Role): P h =g k (1+γ||p k -g k ||)

[0118] Female spotted hyena (follower character):

[0119] Spotted Hyena Cub (Apprentice Role):

[0120] The following combination Figure 4 The specific process of applying the improved spotted hyena optimization algorithm to user clustering in fuzzy C-means clustering is explained.

[0121] First, based on the collected user behavior data, a user video preference matrix (i.e., user-anti-fraud video rating matrix) can be constructed, as follows.

[0122] In the scenario of telecommunications anti-fraud, the matrix of the embodiment of the present application is implemented based on the user-anti-fraud short video rating matrix, which divides the videos in the anti-fraud short video library into categories such as online part-time job order fraud, impersonating customer service fraud, false investment and financial management fraud, etc., and scores are achieved based on the user's interaction with short videos of different categories.

[0123] Specifically, a user-anti-fraud short video matrix can be constructed with users as rows and anti-fraud short videos as columns. The elements in the matrix represent the user's rating or interest in the anti-fraud short video, with higher scores indicating greater interest or greater compliance with user needs. Based on the user's feature scores and the anti-fraud short video's dimension scores, the user's rating of the anti-fraud short video is calculated. This can be achieved through methods such as weighted summation and similarity calculation. The matrix is ​​also normalized to ensure that all ratings are within a range of 1-10. Principal component analysis is used to reduce the matrix's dimensionality to reduce the number of features while retaining key information.

[0124] In the embodiments of this application, the following variables can be defined:

[0125] U represents the user set, U={u1,u2,...,u n}; C represents the category set of anti-fraud short videos, C={c1,c2,...,c n}, where (c1 represents online part-time job fraud, c2 represents fraud by impersonating customer service, etc.); R u,c represents the rating of category c by user u at location l; v u,c represents the set of all short videos under category c that user u has watched; w i represents the weight of the i-th interactive behavior. (Interactive behaviors include: rating, commenting, and forwarding); a u,v,i It represents the number or degree of user u’s i-th interactive behavior on short video v (belonging to category c).

[0126] In order to comprehensively consider the geographical location factor and user interaction behavior, the score of user u for category c is calculated as the sum of weighted scores at multiple geographical locations. Then the score of user u for category c is R u,c It can be calculated as:

[0127]

[0128] Where L is the set of geographical locations visited by user u. l is the weight of the geographic location l, which is determined by factors such as the length of time the user stays at the location and the frequency of visits.

[0129] To calculate the similarity between two users u1 and u2, the cosine similarity calculation formula can be used. The cosine similarity formula is as follows:

[0130]

[0131] Among them, C is the collection of anti-fraud short video categories and are the scores of category c using u1 and u2 respectively.

[0132] This formula calculates the cosine of the angle between two user rating vectors. A larger value indicates that the two users are more similar.

[0133]

[0134] Among them, similar_users(u) is the set of users most similar to user u (you can sort them according to the similarity score and select the top k users), sim(u,u') is the similarity score between user u and user u', R u',v The rating given by user u' to this category.

[0135] The fuzzy C-means clustering algorithm, based on an improved spotted hyena optimization algorithm, determines cluster centers by mapping the cluster center determination process to the foraging behavior of spotted hyenas. The algorithm randomly initializes c cluster centers and encodes them as individual locations in the spotted hyena optimization algorithm. The improved spotted hyena optimization algorithm then searches the dataset to determine the optimal cluster center locations. These optimal locations serve as the initial centers for the fuzzy C-means clustering algorithm, achieving accurate dataset partitioning and outputting a membership matrix and cluster centers. The detailed process is as follows.

[0136] First, initialize the parameters of the fuzzy C-means clustering algorithm, including the number of user clusters, population size N, the threshold ε for stopping clustering iteration, and the fitness value f.

[0137] Then, the improved spotted hyena optimization algorithm is used to map the cluster center determination process to the foraging behavior of spotted hyenas. The specific steps are as follows.

[0138] In some embodiments of the present application, a spotted hyena optimization algorithm is adopted to determine the cluster center based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of the spotted hyena, including: initializing the spotted hyena population using a chaotic mapping algorithm, wherein the position of each spotted hyena is represented as a set of candidate solutions for the cluster center, and the chaotic mapping algorithm includes: Liebovitch chaotic mapping; determining the fitness value corresponding to the position of the spotted hyena based on the user video preference matrix, wherein the fitness value is used to characterize the quality of the candidate solution of the cluster center in terms of whether it can divide users into different clusters; based on the fitness value corresponding to the position of the spotted hyena, the cluster center is optimized and adjusted by simulating the process of the spotted hyena surrounding, hunting and attacking the prey to obtain the target solution of the cluster center.

[0139] Specifically, the embodiment of the present application adds an initialization phase to the improved spotted hyena optimization algorithm. The original population of the spotted hyena optimization algorithm in the related art appears randomly, which leads to an imbalanced distribution of the original population and makes it impossible to balance the richness and uniformity of the population. In the embodiment of the present application, the Liebovitch mapping is proposed to improve the algorithm, which has higher search efficiency than the traditional mapping. Its mathematical model is as follows:

[0140]

[0141] Among them, P1 and P2 are two control parameters, X k represents the result of the kth iteration, that is, the input value of the mapping, X k+1 Represents the result of the k+1th iteration, that is, the output value of the mapping.

[0142] In the embodiment of the present application, the strategy of the newly added initialization phase can be used to initialize the spotted hyena population and calculate the position and fitness value f of the spotted hyena. F Then, in the subsequent stages of encirclement, hunting, and attack, the position and fitness value of the spotted hyena are updated according to different strategies, as follows.

[0143] In some embodiments of the present application, by imitating the process of spotted hyenas surrounding, hunting and attacking prey, the cluster center is optimized and adjusted, including: in the encirclement stage, determining the current optimal solution of the cluster center, and updating the positions and fitness values ​​of other spotted hyenas based on the position of the spotted hyena with the current optimal solution, wherein the current optimal solution is determined based on the fitness value; in the hunting stage, dividing the spotted hyenas in the population into different role types, and adjusting the positions and fitness values ​​of the spotted hyenas according to the hunting strategies corresponding to each role type, wherein the role types include: leader, follower, and apprentice, the spotted hyena in the leader role is used to update its position by performing a global search, the spotted hyena in the follower role is used to update its position by performing a local search under the guidance of the leader, and the spotted hyena in the apprentice role is used to update its position by performing a random search; in the attack stage, a nonlinear weight factor and a Levy flight search mechanism are used to adjust the position and fitness value of the spotted hyena, wherein the nonlinear weight factor is used to balance the capabilities of the global search and the local search, and the Levy flight search mechanism is used to randomly introduce long-distance jumps and short-distance searches in the search and optimization process to avoid falling into the local optimum.

[0144] Specifically, during the hunting phase, the improved hunting mechanism can be used to divide the population and update the positions and fitness values ​​f of the corresponding spotted hyena leaders (leader role), female spotted hyenas (follower role) and spotted hyena cubs (apprentice role). FL 、f FC 、f FY .

[0145] During the attack phase, the improved optimization algorithm uses a nonlinear weighting factor and Levy flight search mechanism to adjust the spotted hyena's position and fitness. Specifically, the nonlinear adaptive weighting factor balances the algorithm's global and local optimization capabilities, which is crucial for improving the algorithm's convergence speed and performance. The nonlinear adaptive weighting factor is introduced into the algorithm's position update method during the attack process, as shown in the following formula:

[0146]

[0147] Among them, w(k) represents the weight factor of the kth iteration, k represents the current weight number, k max Indicates the maximum number of iterations of the algorithm.

[0148] The nonlinear weight factor consists of two phases: a linear decay phase and a nonlinear decay phase. In the early stages of an iteration, the linear decay rate Δk1 remains constant, ensuring the continuity of the algorithm's population diversity and improving search efficiency and accuracy in the early stages. As the number of iterations increases, the weight factor transitions from the linear decay phase to the nonlinear decay phase. In the later stages of an iteration, the nonlinear decay rate Δk2 gradually decreases, exhibiting a certain degree of volatility. This variability in Δk2 further helps ensure diversity in the algorithm's population in the later stages and enhances the algorithm's ability to escape local optima.

[0149] The salient features of the Levy flight search mechanism are its random walk and jumping properties. This property, when embedded in the algorithm, can increase the diversity of the algorithm population and the flexibility of the search space. When the algorithm is trapped near the local optimum, it can use the jumping characteristics of Levy flight to jump out of the local optimum. At the same time, it can use the random walk of Levy flight to continuously expand the search space and improve the convergence efficiency and convergence accuracy of the algorithm. Therefore, the Levy flight search strategy is introduced in the position update method to perturb the position and improve the algorithm's ability to jump out of the local optimum and the search efficiency. The specific formula is as follows:

[0150] L evy (s,ζ)≈s -ζ ,ζ∈(1,3]

[0151] Among them, s is the random search step size; ζ is a random number in (1, 3]

[0152] The embodiment of the present application can combine the characteristics of nonlinear adaptive weight factors and Levy flight mechanism. In the algorithm, the position update during the attack process adopts a nonlinear adaptive Levy flight search method. This can increase the possibility of the algorithm escaping the local optimum and further improve the convergence accuracy of the algorithm. This strategy can maintain stability for a long time in both the early and late iterations, allowing the spotted hyena to achieve good results in both the early global optimization and the later local optimization. The inertia weight is applied to the spotted hyena position update formula:

[0153] P(x+1)=C h / N·w(k)·L evy

[0154] In addition, a new Cauchy reverse learning mutation stage is added to the improved spotted hyena optimization algorithm in the embodiment of the present application, as follows.

[0155] In some embodiments of the present application, the method also includes the following steps: in the process of optimizing and adjusting, the Cauchy mutation strategy and the reverse learning strategy are used to perturb the current optimal solution of the cluster center to adjust the position and fitness value of the spotted hyena corresponding to the current optimal solution, wherein the Cauchy mutation strategy is used to perturb the current optimal solution by generating a Cauchy distribution random variable, and the reverse learning strategy is used to perturb by constructing a reverse sister solution corresponding to the current optimal solution.

[0156] Specifically, in the later stages of the algorithm, individuals cluster as the number of iterations increases. The optimal individual, lacking the ability to escape local minima, is prone to becoming trapped in local optima. Therefore, a Cauchy anti-learning mutation strategy is introduced to perturb the current optimal solution, enhancing the diversity and global search capabilities of the population and preventing the algorithm from being trapped in local minima. The standard Cauchy distribution function has a relatively small peak value. After the algorithm undergoes Cauchy mutation, it can spend less time on local search, improving its optimization performance. Therefore, the algorithm is integrated with Cauchy mutation, leveraging its strong perturbation capability to manipulate the current optimal solution and accelerate the population's approach to the optimal position. The standard Cauchy distribution function and the mathematical model for updating the optimal solution are shown below.

[0157]

[0158] X newbest =X best +X best ×Cauchy(0,1)

[0159] Where f(x) is the probability density function of the Cauchy distribution, x is a random variable, X newbest is the new optimal solution after Cauchy mutation, X best represents the optimal solution of the current iteration, and Cauchy(0,1) is a random variable with Cauchy distribution.

[0160] Reverse learning is to construct the reverse solution of the current feasible solution in the solution space, and guide individuals to find the best solution by comparing the two, thereby increasing the diversity of the population. The mathematical model of the reverse learning strategy is as follows:

[0161] X'=K1(Ub+Lb)-X best

[0162] X newbest =K2(X best -X')

[0163] Among them, X' is the optimal solution X of the population at time t best The reverse solution, K1, K2 are random numbers between [0,1], X newbest is the updated population position, Ub and Lb represent the upper and lower bounds.

[0164] According to random probability, different mutation strategies are selected to alternately perturb the optimal individuals, helping the population quickly escape the local extreme value area and accelerating the convergence of the algorithm. Based on the above strategy, the Cauchy anti-learning mutation strategy is established, and its mathematical model is shown in the formula.

[0165]

[0166] P is a random number between [0, 1]. When P ≥ 0.5, after obtaining the optimal individual, the population is perturbed through Cauchy mutation, increasing the algorithm's search range and helping it quickly find the global optimal solution. When P < 0.5, the algorithm perturbs the population using a reverse learning strategy to expand the population's mining range. By randomly selecting mutation strategies, the algorithm's optimization speed and global convergence ability are improved to a certain extent.

[0167] The embodiment of the present application improves the spotted hyena optimization algorithm by using a hybrid strategy that includes a new initialization stage, an improved hunting stage using a lion group optimization algorithm, an improved prey attack stage using a nonlinear adaptive Levy flight search, and a new Cauchy reverse learning mutation stage. The algorithm is used to obtain the optimal solution, that is, after the algorithm reaches the maximum number of iterations, the optimal value can be output, that is, the initial cluster center of the fuzzy C-means clustering algorithm is determined.

[0168] Afterwards, the users can be clustered using the fuzzy C-means clustering algorithm based on the initial cluster centers determined by the improved spotted hyena optimization algorithm, as follows.

[0169] In some embodiments of the present application, users are clustered based on cluster centers, and the cluster clusters corresponding to each cluster center are determined, and the clustering results are obtained, including the following steps: determining the cluster center as the initial cluster center of the fuzzy C-means clustering algorithm; determining the membership of each user to each cluster center based on the user video preference matrix, obtaining the membership matrix, and updating the cluster center based on the membership matrix; iteratively repeating the above process of determining the membership matrix and updating the cluster center until the objective function converges, wherein the objective function is used to characterize the clustering quality of clustering based on the current membership matrix and cluster center; and determining the clustering results based on the final membership matrix and cluster center.

[0170] Specifically, the fuzzy C-means clustering algorithm is used to cluster the correlation matrix R m×n Perform fuzzy clustering and divide the user set X into c clusters. Cluster center C = {C1, C2, ..., C c}, the user's membership to the cluster center is μ ij In order to obtain better clustering results, set the constraints to The clustering result is the membership matrix U corresponding to the user (n) = (μ ij) n×c The clustering process is the process of minimizing the objective function value, which can be expressed as the following formula.

[0171]

[0172] Among them, m is a parameter that describes the degree of blur.

[0173] μ ij It can be expressed as the following formula, that is, the degree of membership of user i to the jth cluster.

[0174]

[0175] C j is the cluster center of user cluster j, and the formula is as follows.

[0176]

[0177] d ij is the distance between each user and the cluster center, d ij =||x i -c j ||,x i is the feature vector corresponding to the i-th user.

[0178] When applying the fuzzy C-means clustering algorithm to cluster users, the cluster center matrix C and the membership matrix U are updated according to the above-mentioned related formulas, and the objective function value is calculated and compared until J(T)-J(T+1)<ε. The iteration is stopped and the user cluster membership matrix U is output. Then, based on the final output membership matrix, the nearest similar users can be divided and the predicted scores are calculated, and the final Top-N recommendation results are output, thereby recommending different types of videos to different users.

[0179] In addition, as an optional implementation, the method also includes the following steps: determining the time period corresponding to when collecting user behavior data, and the geographical location corresponding to the Internet of Things terminal device used by the user; determining the target abnormal behavior type corresponding to the geographical location within the time period, wherein the target abnormal behavior type is the type corresponding to the abnormal behavior that has occurred in the geographical location within the time period, and the frequency of occurrence exceeds a preset frequency threshold; obtaining a target video corresponding to the target abnormal behavior type, and recommending the target video to the user's Internet of Things terminal device, wherein the target video is used to remind the user to guard against abnormal behavior of the target abnormal behavior type.

[0180] By combining the user's geographic location with the abnormal behavior types in a specific time period, and focusing on the abnormal behavior types that occur frequently in characteristic time periods and locations, we can ensure that the recommended videos are highly relevant to the actual risks faced by the user.

[0181] After completing the video recommendation, you can also use the recommendation effect evaluation module in the system to evaluate the effectiveness of the video recommendation. During the evaluation, in addition to focusing on the performance of the recommendation system, it is also necessary to conduct a comprehensive assessment based on the user's knowledge mastery, behavior changes, and reporting and feedback. The details are as follows.

[0182] Specifically, during the assessment, a questionnaire containing questions related to anti-fraud knowledge can be designed. The questions should cover common fraud tactics, fraud identification techniques, and response strategies. Users who have watched the anti-fraud short video are selected as the sample to ensure representativeness. Data is collected through online or paper questionnaires, and user responses are statistically analyzed to understand the user's level of anti-fraud knowledge.

[0183] In addition, we can also record the user's behavioral data before and after watching the anti-fraud short video, such as the number of times they clicked on suspicious links, the length of time they answered fraudulent calls, etc. We can compare and analyze the user's behavioral data before and after watching the anti-fraud short video to see if there is any obvious phenomenon of reducing the risk of being deceived.

[0184] Recommendation algorithms are continuously optimized and adjusted based on the results of questionnaire surveys, behavioral changes, reports, and feedback. For example, for content quality issues that receive significant user feedback, short video screening and recommendation strategies can be optimized; for user groups with less pronounced behavioral changes, the relevance and interest of recommended content can be adjusted.

[0185] This application solution collects users' device usage data, network behavior data, etc. through the Internet of Things gateway, and uses relevant technologies to process and analyze the data to identify potential fraudulent behavior and make personalized video recommendations, thereby improving the user experience.

[0186] According to an embodiment of the present application, an embodiment of a video recommendation device is also provided. Figure 5 Schematic diagram of a video recommendation device according to an embodiment of the present application. Figure 5 As shown, the device includes:

[0187] The acquisition module 50 is used to collect user behavior data and determine a user video preference matrix based on the user behavior data, wherein the user video preference matrix is ​​used to represent the degree of interest of different users in different types of videos;

[0188] An optimization module 52 is configured to determine cluster centers based on a user video preference matrix by using a spotted hyena optimization algorithm to map the cluster center determination process to the foraging behavior of a spotted hyena, wherein each cluster center corresponds to a video preference feature type of the user;

[0189] A clustering module 54 is configured to cluster users based on cluster centers, determine clusters corresponding to each cluster center, and obtain clustering results, wherein users in the same cluster have the same video preference feature type;

[0190] The recommendation module 56 is configured to recommend videos to users based on the clustering results, wherein different types of videos are recommended to users in different clusters.

[0191] Optionally, a spotted hyena optimization algorithm is used to determine the cluster center based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of the spotted hyena. The method includes: initializing the spotted hyena population using a chaotic mapping algorithm, wherein the position of each spotted hyena is represented as a set of candidate solutions for the cluster center, and the chaotic mapping algorithm includes: Liebovitch chaotic mapping; determining the fitness value corresponding to the position of the spotted hyena based on the user video preference matrix, wherein the fitness value is used to characterize the quality of the candidate solution of the cluster center in terms of whether the user can be divided into different clusters; based on the fitness value corresponding to the position of the spotted hyena, the cluster center is optimized and adjusted by simulating the process of the spotted hyena surrounding, hunting and attacking the prey to obtain the target solution of the cluster center.

[0192] Optionally, by imitating the process of spotted hyenas surrounding, hunting and attacking their prey, the cluster center is optimized and adjusted, including: in the surrounding stage, determining the current optimal solution of the cluster center, and updating the positions and fitness values ​​of other spotted hyenas based on the position of the spotted hyena with the current optimal solution, wherein the current optimal solution is determined based on the fitness value; in the hunting stage, dividing the spotted hyenas in the population into different role types, and adjusting the positions and fitness values ​​of the spotted hyenas according to the hunting strategies corresponding to each role type, wherein the role types include: leader, follower, and apprentice, the spotted hyena in the leader role is used to update its position by performing a global search, the spotted hyena in the follower role is used to update its position by performing a local search under the guidance of the leader, and the spotted hyena in the apprentice role is used to update its position by performing a random search; in the attacking stage, a nonlinear weight factor and a Levy flight search mechanism are used to adjust the positions and fitness values ​​of the spotted hyenas, wherein the nonlinear weight factor is used to balance the capabilities of the global search and the local search, and the Levy flight search mechanism is used to randomly introduce long-distance jumps and short-distance searches in the search and optimization process to avoid falling into local optimality.

[0193] Optionally, the optimization module 52 is also used to: in the process of optimization adjustment, use the Cauchy mutation strategy and the reverse learning strategy to perturb the current optimal solution of the cluster center to adjust the position and fitness value of the spotted hyena corresponding to the current optimal solution, wherein the Cauchy mutation strategy is used to perturb the current optimal solution by generating a Cauchy distribution random variable, and the reverse learning strategy is used to perturb the current optimal solution by constructing a reverse sister solution corresponding to the current optimal solution.

[0194] Optionally, users are clustered based on cluster centers, and the cluster clusters corresponding to each cluster center are determined, and the clustering results obtained include: determining the cluster center as the initial cluster center of the fuzzy C-means clustering algorithm; determining the membership of each user to each cluster center based on the user video preference matrix, obtaining the membership matrix, and updating the cluster center based on the membership matrix; iteratively repeating the above process of determining the membership matrix and updating the cluster center until the objective function converges, wherein the objective function is used to characterize the clustering quality of clustering based on the current membership matrix and cluster center; and determining the clustering results based on the final membership matrix and cluster center.

[0195] Optionally, the video recommendation device is also used to: obtain an abnormal behavior feature database, and determine that there is a security risk in the user behavior data when text data appears in the user behavior data and the text similarity between the text and the abnormal behavior-related text exceeds a preset similarity threshold, wherein the abnormal behavior feature database contains multiple abnormal behavior-related texts; or determine a first behavior feature pattern corresponding to the user behavior data, and determine that there is a security risk in the user behavior data when the pattern similarity between the first behavior feature pattern and the second behavior feature pattern is lower than a second preset similarity threshold, wherein the second behavior feature pattern is the behavior feature pattern corresponding to the user's historical user behavior data; when it is determined that there is a security risk in the user behavior data, an alarm message is sent to the Internet of Things terminal device corresponding to the user.

[0196] Optionally, the video recommendation device is also used to: determine the time period corresponding to when collecting user behavior data, and the geographical location corresponding to the Internet of Things terminal device used by the user; determine the target abnormal behavior type corresponding to the geographical location within the time period, wherein the target abnormal behavior type is the type corresponding to the abnormal behavior that has occurred in the geographical location within the time period, and the frequency of occurrence exceeds a preset frequency threshold; obtain a target video corresponding to the target abnormal behavior type, and recommend the target video to the user's Internet of Things terminal device, wherein the target video is used to remind the user to guard against abnormal behavior of the target abnormal behavior type.

[0197] It should be noted that the various modules in the above-mentioned video recommendation device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0198] It should be noted that the video recommendation device provided in this embodiment can be used to perform Figure 2 The video recommendation method shown, therefore, the relevant explanations of the above video recommendation method are also applicable to the embodiments of the present application and will not be repeated here.

[0199] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the following video recommendation method by running the computer program: collecting user behavior data, and determining a user video preference matrix based on the user behavior data, wherein the user video preference matrix is ​​used to characterize the degree of interest of different users in different types of videos; using a spotted hyena optimization algorithm, by mapping the cluster center determination process to the foraging behavior of a spotted hyena, to determine the cluster center based on the user video preference matrix, wherein each cluster center corresponds to a type of video preference feature of the user; clustering the users based on the cluster center, determining the cluster cluster corresponding to each cluster center, and obtaining a clustering result, wherein the video preference feature types corresponding to users in the same cluster cluster are the same; and recommending videos to the users based on the clustering result, wherein the types of videos recommended to users in different clusters are different.

[0200] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the video recommendation method described in each embodiment of the present application: collecting user behavior data, and determining a user video preference matrix based on the user behavior data, wherein the user video preference matrix is ​​used to characterize the degree of interest of different users in different types of videos; using a spotted hyena optimization algorithm, by mapping the cluster center determination process to the foraging behavior of a spotted hyena, to determine the cluster centers based on the user video preference matrix, wherein each cluster center corresponds to a type of video preference feature of the user; clustering the users based on the cluster centers, determining the cluster clusters corresponding to each cluster center, and obtaining clustering results, wherein the video preference feature types corresponding to users in the same cluster cluster are the same; and recommending videos to the users based on the clustering results, wherein the types of videos recommended to users in different clusters are different.

[0201] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0202] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0203] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0204] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0205] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0206] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0207] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A video recommendation method, characterized in that: include: Collecting user behavior data and determining a user video preference matrix based on the user behavior data, wherein the user video preference matrix is ​​used to represent the degree of interest of different users in different types of videos; A spotted hyena optimization algorithm is used to determine cluster centers based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of a spotted hyena, wherein each cluster center corresponds to a video preference feature type of the user; Clustering the users based on the cluster centers, determining a cluster corresponding to each cluster center, and obtaining a clustering result, wherein the video preference feature types corresponding to the users in the same cluster are the same; Based on the clustering result, videos are recommended to the user, wherein the types of videos recommended to the users in different clusters are different.

2. The video recommendation method according to claim 1, characterized in that The spotted hyena optimization algorithm is used to map the cluster center determination process to the foraging behavior of spotted hyenas. The cluster centers are determined based on the user video preference matrix. The method includes: A chaotic mapping algorithm is used to initialize the spotted hyena population, wherein the position of each spotted hyena is represented as a set of candidate solutions for cluster centers. The chaotic mapping algorithm includes: Liebovitch chaotic mapping; Determining, based on the user video preference matrix, a fitness value corresponding to the position of the spotted hyena, wherein the fitness value is used to represent the quality of the candidate solution of the cluster center in terms of whether it can divide users into different clusters; Based on the fitness value corresponding to the position of the spotted hyena, the cluster center is optimized and adjusted by simulating the process of the spotted hyena surrounding, hunting and attacking prey to obtain a target solution for the cluster center.

3. The video recommendation method according to claim 2, characterized in that By simulating the process of spotted hyenas surrounding, hunting and attacking prey, the cluster center is optimized and adjusted including: In the encirclement phase, the current optimal solution of the cluster center is determined, and the positions and fitness values ​​of other spotted hyenas are updated according to the position of the spotted hyena of the current optimal solution, wherein the current optimal solution is determined according to the fitness value; During the hunting phase, the spotted hyenas in the population are divided into different role types, and the positions and fitness values ​​of the spotted hyenas are adjusted according to the hunting strategies corresponding to each role type, wherein the role types include: leader, follower, and apprentice. The spotted hyenas in the leader role are used to update their positions by performing a global search, the spotted hyenas in the follower role are used to update their positions by performing a local search under the guidance of the leader, and the spotted hyenas in the apprentice role are used to update their positions by performing a random search. During the attack phase, a nonlinear weight factor and a Levy flight search mechanism are used to adjust the position and fitness value of the spotted hyena, wherein the nonlinear weight factor is used to balance the capabilities of global search and local search, and the Levy flight search mechanism is used to randomly introduce long-distance jumps and short-distance searches during the search and optimization process to avoid falling into local optimality.

4. The video recommendation method according to claim 3, wherein: The method further comprises: In the process of optimizing and adjusting, the Cauchy mutation strategy and the reverse learning strategy are used to perturb the current optimal solution of the cluster center to adjust the position and fitness value of the spotted hyena corresponding to the current optimal solution, wherein the Cauchy mutation strategy is used to perturb the current optimal solution by generating Cauchy distribution random variables, and the reverse learning strategy is used to perturb by constructing the reverse sister solution corresponding to the current optimal solution.

5. The video recommendation method according to claim 1, wherein: Clustering the users based on the cluster centers, determining the clusters corresponding to each cluster center, and obtaining clustering results include: Determining the cluster center as the initial cluster center of the fuzzy C-means clustering algorithm; Determining the membership of each user to each cluster center based on the user video preference matrix to obtain a membership matrix, and updating the cluster centers based on the membership matrix; Iteratively repeating the above process of determining the membership matrix and updating the cluster centers until the objective function converges, wherein the objective function is used to characterize the clustering quality of clustering based on the current membership matrix and cluster centers; The clustering result is determined based on the final membership matrix and the cluster centers.

6. The video recommendation method according to claim 1, wherein: The method further comprises: Obtaining an abnormal behavior feature database, and determining that the user behavior data presents a security risk if text data having a text similarity with abnormal behavior-related text exceeding a preset similarity threshold appears in the user behavior data, wherein the abnormal behavior feature database contains a plurality of abnormal behavior-related texts; Alternatively, determining a first behavior characteristic pattern corresponding to the user behavior data, and determining that the user behavior data has a security risk if the pattern similarity between the first behavior characteristic pattern and a second behavior characteristic pattern is lower than a second preset similarity threshold, wherein the second behavior characteristic pattern is a behavior characteristic pattern corresponding to the historical user behavior data of the user; When it is determined that there is a security risk in the user behavior data, an alarm message is sent to the IoT terminal device corresponding to the user.

7. The video recommendation method according to claim 1, wherein: The method further comprises: Determine the time period corresponding to when the user behavior data was collected, and the geographical location corresponding to the IoT terminal device used by the user; Determine a target abnormal behavior type corresponding to the geographic location within the time period, wherein the target abnormal behavior type is a type corresponding to abnormal behavior that has occurred at the geographic location within the time period and has an occurrence frequency exceeding a preset frequency threshold; A target video corresponding to the target abnormal behavior type is obtained, and the target video is recommended to the Internet of Things terminal device of the user, wherein the target video is used to remind the user to guard against abnormal behavior of the target abnormal behavior type.

8. A video recommendation device, characterized in that: include: An acquisition module is used to collect user behavior data and determine a user video preference matrix based on the user behavior data, wherein the user video preference matrix is ​​used to represent the degree of interest of different users in different types of videos; an optimization module for determining cluster centers based on the user video preference matrix by mapping the cluster center determination process to the foraging behavior of a spotted hyena using a spotted hyena optimization algorithm, wherein each cluster center corresponds to a type of video preference feature of the user; A clustering module, configured to cluster users based on the cluster centers, determine a cluster corresponding to each cluster center, and obtain a clustering result, wherein the video preference feature types corresponding to the users in the same cluster are the same; A recommendation module is configured to recommend videos to the user based on the clustering result, wherein the types of videos recommended to the users in different clusters are different.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is configured to run a program stored in the memory, wherein the video recommendation method according to any one of claims 1 to 7 is executed when the program is run.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the video recommendation method according to any one of claims 1 to 7 by running the computer program.

11. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the video recommendation method according to any one of claims 1 to 7 are implemented.

Images