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

Abstract

The invention discloses a video clip recommendation method and device, a storage medium and electronic equipment. The method comprises the following steps: acquiring video content when a target user watches a first video and executes a preference behavior; a first video feature corresponding to the video content is obtained, and the first video content is used for representing key elements in the video content; obtaining an association relationship between the first video feature and a second video feature, the second video feature being used for representing a key element in the key video clip; according to the association relationship and the popularity degree of the key video clip, obtaining the preference degree of the target user for the key video clip in the second video; and under the condition that the preference degree is greater than or equal to a preset threshold value, recommending the key video clip in the second video to the target user. The technical problem that the recommendation accuracy of the video clip is low is solved.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular to a method, device, storage medium and electronic device for recommending video clips. Background Art

[0002] In the video recommendation scenario, users' preferences are usually obtained by analyzing their historical behavior data, and personalized video content is pushed accordingly. However, due to the diversity and complexity of video content, it is often difficult to accurately distinguish whether users prefer the content itself or specific elements in the content by directly analyzing content preferences. In addition, users' preferences for videos may be affected by multiple factors, not just personal interests, which leads to the problem of low accuracy in recommending video clips.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a method, device, storage medium and electronic device for recommending video clips, so as to at least solve the technical problem of low accuracy in recommending video clips.

[0005] According to one aspect of an embodiment of the present application, a method for recommending video clips is provided, comprising: obtaining video content when a target user watches a first video and performs a preferred behavior; obtaining a first video feature corresponding to the above video content, wherein the above first video content is used to characterize key elements in the above video content; obtaining an association relationship between the above first video feature and a second video feature, wherein the above second video feature is used to characterize key elements in key video clips; obtaining the degree of preference of the above target user for the key video clip in the second video based on the above association relationship and the popularity of the above key video clip; and recommending the key video clip in the above second video to the above target user when the above preference degree is greater than or equal to a preset threshold.

[0006] According to another aspect of an embodiment of the present application, a video clip recommendation device is also provided, including: a first acquisition unit, used to acquire the video content when a target user watches a first video and performs a preferred behavior; a second acquisition unit, used to acquire a first video feature corresponding to the above video content, wherein the above first video content is used to characterize the key elements in the above video content; a third acquisition unit, used to acquire the association relationship between the above first video feature and the second video feature, wherein the above second video feature is used to characterize the key elements in the key video clip; a fourth acquisition unit, used to acquire the target user's preference for the key video clip in the second video based on the above association relationship and the popularity of the above key video clip; a recommendation unit, used to recommend the key video clip in the second video to the target user when the above preference degree is greater than or equal to a preset threshold.

[0007] As an optional scheme, the above-mentioned device also includes: a fifth acquisition unit, which is used to obtain the first loss function corresponding to the above-mentioned target user and the second loss function corresponding to the above-mentioned target user before obtaining the above-mentioned target user's preference degree for the key video clip in the second video based on the above-mentioned association relationship and the popularity of the above-mentioned key video clip, wherein the above-mentioned first loss function is used to measure the matching degree between the above-mentioned preference degree and the personal interest of the above-mentioned target user, and the above-mentioned second loss function is used to measure the influence of the above-mentioned target user's herd behavior on the above-mentioned preference degree; the above-mentioned fourth acquisition unit includes: an acquisition module, which is used to obtain the above-mentioned preference degree based on the above-mentioned first loss function, the above-mentioned second loss function, the above-mentioned popularity, and the above-mentioned association relationship.

[0008] As an optional solution, the above-mentioned acquisition module includes: a first construction sub-module, which is used to construct the personal interest representation of the above-mentioned target user based on the historical behavior, preference settings and viewing records of the above-mentioned target user, and obtain the above-mentioned first loss function; a second construction sub-module, which is used to construct the herd behavior representation of the above-mentioned target user based on the interactive behavior of the above-mentioned target user on the social platform and combined with popular videos, and obtain the above-mentioned second loss function.

[0009] As an optional solution, the above-mentioned third acquisition unit includes: an input module, used to input the above-mentioned first video feature and the above-mentioned second video feature into an association model to obtain the feature similarity between the above-mentioned first video feature and the second video feature, wherein the above-mentioned association model is a pre-trained neural network model for analyzing the feature similarity between video features.

[0010] As an optional solution, the above-mentioned device also includes: a feedback module, which is used to obtain the behavioral feedback data of the above-mentioned target user on the key video clip in the above-mentioned second video after the key video clip in the above-mentioned second video is recommended to the above-mentioned target user; an optimization module, which is used to optimize the above-mentioned association model using the above-mentioned behavioral feedback data after the key video clip in the above-mentioned second video is recommended to the above-mentioned target user, so as to obtain an optimized association model.

[0011] As an optional solution, the above-mentioned second acquisition unit includes at least one of the following: an identification module, used to identify the characters in the above-mentioned video content to obtain the above-mentioned first video feature; an analysis module, used to analyze the behavior in the above-mentioned video content to obtain the above-mentioned first video feature; a judgment module, used to judge the emotions in the above-mentioned video content to obtain the above-mentioned first video feature.

[0012] As an optional solution, the above-mentioned first acquisition unit includes at least one of the following: an execution module, used to obtain at least one video segment of the above-mentioned target user watching the above-mentioned first video and performing a specified playback behavior; a preprocessing module, used to preprocess the original data of the above-mentioned at least one video segment, remove invalid or abnormal data points, and obtain the above-mentioned video content.

[0013] According to another aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs the method for recommending video clips as described above.

[0014] According to another aspect of an embodiment of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method for recommending video clips through the computer program.

[0015] In an embodiment of the present application, the video content of a target user when watching a first video and performing a preferred behavior is obtained; a first video feature corresponding to the above video content is obtained, wherein the above first video content is used to characterize key elements in the above video content; an association relationship between the above first video feature and a second video feature is obtained, wherein the above second video feature is used to characterize key elements in a key video clip; based on the above association relationship and the popularity of the above key video clip, the preference degree of the above target user for the key video clip in the second video is obtained; and when the above preference degree is greater than or equal to a preset threshold, the key video clip in the above second video is recommended to the above target user.

[0016] By deeply analyzing the correlation between user behavior data and video features, and taking popularity as a reference factor for recommendation, we can achieve the goal of accurate mapping from user preferences to specific video clips, thereby achieving the technical effect of improving the accuracy of video clip recommendations, and thus solving the technical problem of low accuracy in video clip recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] 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:

[0018] Figure 1 is a schematic diagram of an application environment of an optional video clip recommendation method according to an embodiment of the present application;

[0019] Figure 2 is a schematic diagram of a process of an optional method for recommending video clips according to an embodiment of the present application;

[0020] Figure 3 is a schematic diagram of another optional method for recommending video clips according to an embodiment of the present application;

[0021] Figure 4 is a schematic diagram of an optional video clip recommendation device according to an embodiment of the present application;

[0022] Figure 5 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0024] 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 sequence. 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 an order 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.

[0025] According to one aspect of an embodiment of the present application, a method for recommending a video clip is provided. Optionally, as an optional implementation, the method for recommending a video clip can be applied to, but is not limited to, Figure 1 In the environment shown, the environment may include, but is not limited to, a user device 102 and a server 112 . The user device 102 may include, but is not limited to, a display 104 , a processor 106 , and a memory 108 . The server 112 includes a database 114 and a processing engine 116 .

[0026] The specific process can be as follows:

[0027] Step S102, the user device 102 obtains the video content when the target user watches the first video and performs the preferred behavior;

[0028] Step S104, sending the video content to the server 112 via the network 110;

[0029] Steps S106-S110: the server 112 obtains the first video feature corresponding to the video content through the processing engine 116, and further obtains the association relationship between the first video feature and the second video feature; obtains the target user's preference for the key video segment in the second video according to the association relationship and the popularity of the key video segment; and when the preference is greater than or equal to a preset threshold, uses the key video segment in the second video as a video segment to be recommended to the target user;

[0030] Step S112 , sending the key video segment in the second video to the user device 102 via the network 110 , the user device 102 displays the key video segment in the second video on the display 104 via the processor 106 , and stores the key video segment in the second video in the memory 108 .

[0031] remove Figure 1In addition to the examples shown, the terminal device can be a terminal device configured with a target client, which can include but is not limited to at least one of the following: a mobile phone (such as an Android phone, an iOS phone, etc.), a laptop, a tablet computer, a PDA, a MID (Mobile Internet Devices), a PAD, a desktop computer, a smart TV, etc. The target client can be a video client, an instant messaging client, a browser client, an education client, etc. The network can include but is not limited to: a wired network, a wireless network, wherein the wired network includes: a local area network, a metropolitan area network and a wide area network, and the wireless network includes: Bluetooth, WIFI and other networks that realize wireless communication. The server can be a single server, or a server cluster composed of multiple servers, or a cloud server. The above is only an example, and no limitation is made to this in this embodiment.

[0032] Optionally, as an optional implementation, as Figure 2 As shown, the method for recommending a video clip may be executed by an electronic device, and the electronic device may be, for example, Figure 1 The user device or server shown in the figure comprises:

[0033] S202, obtaining the video content when the target user watches the first video and performs the preferred behavior;

[0034] S204, obtaining a first video feature corresponding to the video content, wherein the first video content is used to represent a key element in the video content;

[0035] S206, obtaining an association relationship between the first video feature and the second video feature, wherein the second video feature is used to represent a key element in the key video segment;

[0036] S208, obtaining the target user's preference for the key video segment in the second video according to the association relationship and the popularity of the key video segment;

[0037] S210: When the preference level is greater than or equal to a preset threshold, recommend the key video segment in the second video to the target user.

[0038] In an optional embodiment, target user: a specific user that the system focuses on, whose viewing behavior and preferences are the objects of analysis.

[0039] In an optional embodiment, the first video is a video that the user watches and shows a preferred behavior. The preferred behavior is the preference shown by the user when watching the video, such as liking, commenting, sharing, repeat watching, etc. The video content is the specific picture or information when the user performs the preferred behavior in the first video.

[0040] In an optional embodiment, the first video feature: a feature that characterizes key elements (such as characters, scenes, plots, etc.) in the first video.

[0041] In an optional embodiment, the second video feature: a feature characterizing a key element in a key segment in the second video. The association relationship: the similarity or matching degree between the first video feature and the second video feature. The key video segment: a portion of the second video that is considered attractive.

[0042] In an optional embodiment, popularity: the popularity of a key video clip in the overall user group can be reflected by one or more key data, such as a comprehensive calculation based on the user's viewing behavior, interactive data (such as likes, comments, shares, etc.) and other indicators (such as viewing time, viewing completion rate, etc.). The higher the popularity of a video clip, the more popular it is, and the more users are interested in it or give positive feedback.

[0043] In an optional embodiment, preference degree: the user's preference for the key segment in the second video calculated based on the association relationship and popularity. Preset threshold: a key value for determining whether to recommend to the user, the preference degree must reach or exceed this value.

[0044] It should be noted that this embodiment describes the core process of a personalized recommendation system based on user behavior analysis and video feature matching. The system first captures the preference behavior shown by the user when watching a certain video (first video), and then analyzes the specific content and characteristics of the video when these preference behaviors occur (first video characteristics). Next, the system performs an association analysis on these characteristics and the key segment features (second video features) in another video (second video). Finally, combining this association analysis with the popularity of key video segments, the user's preference for specific key segments in the second video is evaluated, and when the preference reaches or exceeds a certain threshold, these key segments are recommended to the user.

[0045] To further illustrate, it is optional to assume that when the target user A is watching a science fiction movie (first video), he is particularly interested in the scene where a certain character (such as the robot protagonist) appears, and he frequently likes and comments. The system recognizes these preference behaviors and extracts the characteristics of the scene (such as robot design, action special effects, etc.) as the first video features. Subsequently, the system finds key clips containing similar characters or scenes in other science fiction movies (second video) and extracts their features as the second video features. By comparing the two sets of features, the system finds a high degree of correlation, and combined with the popularity of the key clip among the majority of users, calculates the degree of preference of user A for this clip. If the degree of preference exceeds the preset threshold, the system will recommend this key clip to user A.

[0046] This recommendation mechanism can be widely used in various video platforms, such as short video sharing, online movies, TV series streaming services, etc., to help the platform more accurately meet the personalized needs of users and improve user stickiness and satisfaction. In addition, with the development of technology, the algorithms for feature extraction and association analysis can be continuously optimized, such as using deep learning models for higher-level feature recognition and sentiment analysis, to further improve the accuracy and timeliness of recommendations.

[0047] Through the embodiments provided in this application, the correlation between user behavior data and video features is deeply analyzed, and popularity is used as a reference factor for recommendation, thereby achieving the purpose of accurate mapping from user preferences to specific video clips, thereby achieving the technical effect of improving the accuracy of video clip recommendations.

[0048] As an optional solution, before obtaining the target user's preference for the key video clip in the second video according to the association relationship and the popularity of the key video clip, the method further includes: obtaining a first loss function corresponding to the target user and a second loss function corresponding to the target user, wherein the first loss function is used to measure the matching degree between the preference degree and the target user's personal interest, and the second loss function is used to measure the influence of the target user's conformity behavior on the preference degree;

[0049] According to the association relationship and the popularity of the key video clip, the target user's preference for the key video clip in the second video is obtained, including: obtaining the preference according to the first loss function, the second loss function, the popularity, and the association relationship.

[0050] In an optional embodiment, the first loss function is a mathematical function used to quantify the difference or matching degree between the preference degree and the personal interests of the target user. The second loss function is another mathematical function used to quantify the degree of influence of the popularity (i.e., herd behavior) on the preference degree of the target user.

[0051] In an optional embodiment, the matching degree refers to the consistency between the preference degree and the user's personal interests. Herd behavior refers to the behavioral tendency exhibited by the user due to the influence of the outside world (such as popular trends).

[0052] It should be noted that, based on the original process, this section adds a step to consider the influence of user personal interests and herd behavior. Before calculating the target user's preference for the key video clip in the second video, the system first obtains two loss functions: the first loss function is used to evaluate the match between the preference and the target user's personal interests; the second loss function is used to evaluate the impact of the target user's herd behavior (i.e., behavior affected by popularity) on the preference. Ultimately, the calculation of the preference will comprehensively consider the association, the popularity of the key video clip, and these two loss functions.

[0053] The introduction of this step makes the recommendation system more detailed and personalized. It not only considers the user's direct preference behavior, but also deeply explores the user's potential interests and the possibility of being influenced by the outside world. This recommendation strategy that comprehensively considers the user's personal interests and herd behavior can be widely used in various online services, such as social media, e-commerce platforms, news push, etc., to improve the accuracy of recommendations and user satisfaction.

[0054] To further illustrate, it is optional to assume that when the target user B watches videos, he likes science fiction themes (personal interest) and is easily influenced by hot topics and popular trends (herd behavior). When the system recommends the key video clips in the second video to B, it will first analyze his personal interests (such as through historical viewing records, likes and comments, etc.), and construct a first loss function to measure the match between the recommended content and personal interests. At the same time, the system will consider the currently popular video elements or topics and construct a second loss function to measure the changes in B's preferences due to the herd mentality. Finally, the two loss functions, the popularity of the key video clips, and the correlation between the first video features and the second video features are combined to comprehensively calculate B's preference for the key video clips in the second video.

[0055] By introducing the first loss function and the second loss function, the recommendation system can more comprehensively evaluate the user's preference for recommended content. This approach of comprehensively considering the user's personal interests and herd behavior can help reduce the deviation of recommended content and improve the accuracy and user acceptance of recommendations. At the same time, this also provides more optimization space for the recommendation system, such as further improving the personalization and intelligence of recommendations by adjusting the weight of the loss function or introducing more user behavior characteristics. Ultimately, this will help improve user experience, enhance user stickiness, and promote the long-term development of the platform.

[0056] As an optional solution, obtaining a first loss function corresponding to the target user and a second loss function corresponding to the target user includes:

[0057] S1-1, based on the historical behavior, preference settings and viewing history of the target user, construct the personal interest representation of the target user to obtain the first loss function;

[0058] S1-2, based on the target user’s interactive behavior on the social platform and combined with popular videos, construct a representation of the target user’s herd behavior and obtain the second loss function.

[0059] In an optional embodiment, personal interest representation: a model or vector for representing the user's personal interests is constructed based on information such as the target user's historical behavior, preference settings, and viewing history.

[0060] In an optional embodiment, the herd behavior representation is: a model or vector constructed based on the target user's interactive behavior on the social platform and currently popular video content to represent the user's herd behavior tendency.

[0061] It should be noted that this embodiment describes how to obtain the first loss function and the second loss function corresponding to the target user. The first loss function is constructed based on the historical behavior, preference settings, and viewing history of the target user, which reflects the user's personal interests. The second loss function is constructed based on the target user's interactive behavior on the social platform and combined with popular videos, which reflects the user's herd behavior.

[0062] The refinement of this step provides a more accurate and personalized user modeling method for the recommendation system. By constructing individual interest representation and herd behavior representation separately, the system can understand the user's preferences and needs more comprehensively. This method can not only be applied to video recommendations, but can also be extended to other fields, such as music recommendations, news push, product recommendations, etc.

[0063] In addition, with the continuous development of big data and machine learning technology, the construction methods of personal interest representation and herd behavior representation can be more complex and sophisticated. For example, deep learning models can be used to extract high-level features in user behavior, or graph neural networks can be used to capture the complex relationship between users and videos.

[0064] To further illustrate, it is optional to assume that the target user C is a user who likes science fiction movies. His historical behavior includes a large number of science fiction movie viewing records, and his preference settings also clearly indicate his love for science fiction themes. The system can construct C's personal interest representation based on this information, and then obtain the first loss function, which measures the match between the recommended content and C's preference for science fiction movies.

[0065] At the same time, C often participates in discussions about popular movies on social platforms and tends to watch and share currently popular videos. The system can construct a representation of C's herd behavior based on C's interactive behavior on social platforms and the currently popular video content, and then obtain the second loss function. This function measures whether the recommended content is in line with C's herd behavior tendency, that is, whether it is affected by the current popular trend.

[0066] By building personal interest representations based on the target user's historical behavior, preference settings, and viewing history, and building conformity behavior representations based on interactive behaviors and popular videos on social platforms, the recommendation system can more accurately understand the user's preferences and needs. This refined user modeling method helps to improve the accuracy of recommendations and user satisfaction, and reduce the bias and misjudgment of recommended content. At the same time, this also provides more optimization space for the recommendation system, such as by introducing more user behavior features or adopting more advanced machine learning algorithms to further improve the personalization and intelligence of recommendations. Ultimately, this will help improve user experience, enhance user stickiness, and promote the long-term development of the platform.

[0067] As an optional solution, obtaining the association relationship between the first video feature and the second video feature includes:

[0068] The first video feature and the second video feature are input into an association model to obtain feature similarity between the first video feature and the second video feature, wherein the association model is a pre-trained neural network model for analyzing feature similarity between video features.

[0069] In an optional embodiment, the association model is a pre-trained neural network model used to analyze the feature similarity between video features. Feature similarity is a measure of the similarity between two video features in a specific dimension or as a whole.

[0070] It should be noted that this embodiment describes how to obtain the association relationship between the first video feature and the second video feature. Specifically, the process is to input the first video feature and the second video feature into a pre-trained association model, which is a neural network for analyzing the feature similarity between the video features. Through the processing of the model, the similarity score between the two video features can be obtained, and this score represents the association relationship between them.

[0071] The training of the association model is a key step, which usually requires a large amount of video data as a training set, and corresponding labels or similarity scores as supervision information. Through training, the model can learn the inherent laws and similarity measurement methods between video features. In practical applications, in addition to neural network models, other types of machine learning models can also be used to calculate feature similarity, such as support vector machines, random forests, etc.

[0072] To further illustrate, let's assume that there are two videos, video A and video B, and their feature vectors (i.e., the first video feature and the second video feature) are extracted. These feature vectors may contain a variety of information such as color distribution, motion pattern, audio features, etc. in the video. Then, these feature vectors are input into a pre-trained association model, and the model calculates a similarity score, such as 0.85. This score indicates that video A and video B have a high similarity in the feature space, that is, their association relationship is strong.

[0073] By introducing an association model to calculate the feature similarity between the first video feature and the second video feature, the association relationship between the two videos can be evaluated more accurately. This association analysis method based on feature similarity helps to improve the accuracy of recommendations and user satisfaction because it can more deeply explore the degree of match between user preferences and video content. At the same time, the use of association models also provides more optimization space for recommendation systems, such as further improving the personalization and intelligence of recommendations by improving model structure, increasing training data, or introducing more advanced feature extraction technology. Ultimately, this will help improve user experience and platform competitiveness.

[0074] As an optional solution, after recommending the key video segment in the second video to the target user, the method further includes:

[0075] S2-1, obtaining behavioral feedback data of the target user on the key video clip in the second video;

[0076] S2-2, using the behavioral feedback data, optimize the association model to obtain an optimized association model.

[0077] In an optional embodiment, behavioral feedback data refers to data generated by various behaviors performed by the target user on the recommended content (in this case, the key video clip in the second video), such as viewing time, likes, comments, shares, etc. Association model optimization: According to the user's behavioral feedback data, the association model is parameter adjusted, the structure is improved or retrained to improve the performance and accuracy of the model in the recommendation task.

[0078] It should be noted that this embodiment describes how the recommendation system optimizes the association model based on the user's behavioral feedback data after recommending the key video clips in the second video to the target user. Specifically, the system collects user behavioral feedback on the recommended content, such as viewing time, likes, comments, etc., and then uses this data to train and adjust the association model to improve the performance and accuracy of the model.

[0079] The collection and analysis of behavioral feedback data is an important part of the continuous optimization and improvement of the recommendation system. In addition to being used to optimize the association model, this data can also be used to build user portraits, evaluate recommendation effects, and discover potential user groups. In addition, with the continuous development of machine learning technology, more and more advanced algorithms and models are being applied to recommendation systems, such as deep learning and reinforcement learning. These technologies provide more possibilities and means for the optimization of association models.

[0080] To further illustrate, it is optional to assume that after watching the key video clips in the recommended second video, the target user D watches a certain clip for a long time and gives a positive comment. The system collects these behavioral feedback data and uses them to optimize the association model. Through training and adjustment, the model can better learn the preferences and interests of user D, so as to more accurately provide D with content that suits his taste in future recommendations.

[0081] By optimizing the association model using the behavioral feedback data of target users, the recommendation system can continuously improve its performance and accuracy. This optimization not only helps to better meet the personalized needs of users, improve user satisfaction and loyalty, but also helps to improve the overall recommendation effect and competitiveness of the platform. At the same time, continuous model optimization also brings more innovation and development space to the recommendation system, such as introducing more user behavior features, adopting more advanced machine learning algorithms, etc., to further improve the personalization and intelligence level of recommendations. Ultimately, this will help build a more complete, efficient and intelligent recommendation ecosystem.

[0082] As an optional solution, obtaining the first video feature corresponding to the video content includes at least one of the following:

[0083] Identify a person in the video content to obtain a first video feature;

[0084] Analyze the behavior in the video content to obtain a first video feature;

[0085] The emotion in the video content is determined to obtain a first video feature.

[0086] In an optional embodiment, character recognition: through image processing and machine learning technology, the characters appearing in the video are identified and used as part of the video features. Behavior analysis: the actions, activities or events in the video are analyzed to extract information that can describe the behavioral characteristics of the video content. Emotion judgment: using natural language processing or computer vision technology, the emotional tendency conveyed in the video content, such as happiness, sadness, anger, etc., is judged and used as part of the video features.

[0087] It should be noted that this embodiment describes how to obtain the first video feature corresponding to the video content, and specifically mentions three possible methods: identifying characters in the video content, analyzing behaviors in the video content, and judging emotions in the video content. These methods can be used alone or in combination to extract information that can represent the characteristics of the video content.

[0088] To further illustrate, an optional example is character recognition: in a movie, the system can use facial recognition technology to identify the main actors and use this information as video features. In this way, when recommending similar movies, the system can take the actors into consideration.

[0089] Behavior analysis: In a sports video, the system can analyze the athlete's movements, such as shooting, running, jumping, etc., and extract these behavioral features. This helps recommend videos with similar sports scenes. Emotional judgment: In an advertising video, the system can judge the emotional tendency conveyed by the video, such as happiness, warmth, inspiration, etc., and use this information as video features. This helps recommend videos that can resonate with similar emotions in users.

[0090] In addition to the above three methods, other technologies can also be used to obtain video features, such as audio analysis (extracting music, voice and other features), scene recognition (recognizing places, environments, etc. in the video), object detection (recognizing objects, items, etc. in the video), etc. The combination of these technologies can more comprehensively describe the video content and improve the accuracy and personalization of recommendations.

[0091] By using multiple methods to obtain the first video features corresponding to the video content, the recommendation system can more accurately understand the characteristics and attributes of the video content. This helps to improve the accuracy of recommendations and user satisfaction, because the system can more comprehensively consider user preferences and needs. At the same time, rich video features also provide more optimization space and innovation possibilities for the recommendation system, such as introducing more feature dimensions and adopting more advanced feature extraction techniques to further improve the personalization and intelligence of recommendations. Ultimately, this will help build a more complete, efficient and intelligent video recommendation system.

[0092] As an optional solution, obtaining the video content when the target user watches the first video and performs the preferred behavior includes at least one of the following:

[0093] Acquire at least one video segment in which a target user is watching a first video and performing a specified playback behavior;

[0094] Preprocessing is performed on raw data of at least one video clip to remove invalid or abnormal data points to obtain video content.

[0095] In an optional embodiment, the specified playback behavior refers to the special attention or preference for a certain part of the content expressed by the target user through specific playback operations (such as pause, replay, fast forward / rewind, etc.) when watching the video.

[0096] It should be noted that this embodiment describes how to obtain the video content that the target user pays attention to due to performing a preferred behavior (such as a specific playback operation) when watching the first video. This process includes at least two core links: first, the system will record and analyze the specific playback behaviors performed by the target user during the video viewing process, such as pause, replay, etc., and locate the corresponding video clips accordingly; second, the system will pre-process the original data of these video clips to filter out invalid or abnormal data points, so as to obtain purer and more accurate video content.

[0097] To further illustrate, it is optional to assume that when the target user F is watching a documentary, he is particularly interested in the explanation of a historical event, so he pauses and replays this part many times. The system will first identify and record these preference behaviors, and then locate the corresponding video clips. Then, the system will pre-process the raw data of these clips, such as correcting the freeze caused by network problems, deleting invalid frames, etc., so as to obtain the video content that user F is really interested in.

[0098] In addition to playback behavior, the system can also consider other types of user interaction behaviors, such as likes, comments, shares, and viewing time, which can reflect user preferences from different angles. In addition, the preprocessing steps can also be more in-depth and detailed, such as using advanced image processing technology to improve video quality, and using natural language processing to analyze the emotional tendencies in user comments, so as to more comprehensively explore user preferences.

[0099] By accurately acquiring the video content that target users pay attention to due to their preference behavior when watching videos, the recommendation system can have a deeper understanding of users' interests and needs. This method of acquiring video content based on users' actual behavior not only improves the accuracy of recommendations, but also enhances user satisfaction and loyalty. At the same time, the meticulous preprocessing of video content also ensures the accuracy and reliability of the data, providing a solid foundation for subsequent recommendation algorithms. Ultimately, this will promote the construction of a more personalized, intelligent and efficient video recommendation ecosystem.

[0100] As an optional solution, for ease of understanding, the above video clip recommendation method is applied to the recommendation scenario of video hot spots. With the widespread popularization of the Internet and the deep penetration of mobile devices, video content has been closely integrated into people's daily lives and has become an indispensable part. With rich and diverse content resources such as dramas, movies, and variety shows, it has successfully attracted the attention of a large number of users. In this process, users' video content selection and viewing habits are quietly changing. They are no longer satisfied with the traditional passive acceptance mode, but are more inclined to look for those exciting clips that can quickly attract attention and trigger emotional resonance.

[0101] Faced with users' growing and diverse content needs, major video platforms have begun to upgrade their recommendation systems in an effort to provide more personalized and accurate content recommendation services. Traditional recommendation systems mainly screen and sort content based on users' historical viewing behavior, interest preferences, and basic information about the series (such as director, actor, type, etc.). However, with the rapid growth of content resources and the increasing diversification of user needs, the limitations of traditional recommendation systems in terms of in-depth understanding of content and mining of user interests have gradually become apparent.

[0102] This embodiment innovatively proposes a multi-dimensional fusion framework to address the shortcomings of traditional recommendation systems in processing hot and explosive content. This framework achieves more personalized and accurate content recommendations by accurately associating users' real-time behavioral characteristics with character attributes, event behaviors, and emotional experiences in explosive content. At the same time, this embodiment also introduces a correction model for users' interests in explosive content to dynamically balance users' personal interests and herd behavior, ensuring that the recommended explosive content not only meets users' immediate needs, but also guides users to discover new areas of interest.

[0103] In an optional embodiment, when watching a series (including TV series, movies, variety shows, etc.), the audience often encounters multiple fascinating events, behaviors or emotional climaxes. These explosive points as the core of the plot can deeply touch the audience's emotions and stimulate their enthusiasm for watching. However, the user's interest characteristics and emotional state are not static, but change with the development of the plot and the fluctuation of personal psychological state. Traditional modeling methods are powerless to deal with this dynamic nature. It is difficult to effectively capture the user's immediate emotional fluctuations when watching, and it is even more difficult to accurately associate these changes with the emotional and behavioral characteristics of the explosive points in the series. This modeling limitation makes it difficult for the recommendation system to accurately understand the user's current feelings and emotional fluctuations, and thus cannot make recommendations that meet the user's immediate interests.

[0104] In this regard, this embodiment collects and preprocesses the real-time behavior data of users when watching videos, including operations such as playing, fast forwarding, replaying, and dragging progress bars. Using the video analysis platform, this embodiment labels the clips or short films of users' clicks, playing, fast forwarding, and replaying with features such as characters, behaviors, and emotions, and quantizes these features. This step is to convert the user's real-time behavior into a computable data form for subsequent analysis and modeling.

[0105] Next, this embodiment also vectorizes the characters, behaviors, and emotional features of the hot spots prepared in advance. Then, based on deep learning algorithms (such as neural networks, LSTM, etc.), this embodiment builds an association model. The input of the model is the user behavior feature vector and the hot spot feature vector, and the output is the similarity score between the two. By training the model, it can learn the potential correlation between user behavior and hot spots, and can predict the hot spots that users may be interested in based on their real-time behavior characteristics.

[0106] Finally, according to the user's feedback on the recommended hot spots, this embodiment updates the model weights in real time. This step is to ensure that the model can continue to learn and adapt to changes in user interests, so as to more accurately capture the user's current interests and needs.

[0107] In an optional embodiment, the hot spot, as the most attractive and impactful part of the video, can quickly attract the audience's attention in a short period of time, whether it is a plot reversal, visual impact or emotional climax. Users' love for hot spots may come from their interest in the content itself, or may be affected by group effects or social influences. When users see that a hot spot is widely circulated on social media and receives a large number of likes or recommendations, even if the content does not completely meet their personal preferences, they may choose to watch and like it due to "herd mentality" or "social identity." This phenomenon shows that user behavior is not only based on personal interests, but also affected by external social factors. Therefore, how to balance personal interests and herd behavior to optimize recommendation strategies has become an important challenge.

[0108] In this regard, during the modeling process, this embodiment divides the representation of users and items into two parts: one is based on personal interests, and the other is the representation of conformity behavior. For these two parts, this embodiment constructs the interest loss function and the conformity loss function respectively. The specific formulas are as follows:

[0109]

[0110]

[0111] Among them, J(θ) is the interest loss function, L(y,f(x)) is the conformity loss function, m usually represents the number of samples, that is, the number of data points involved in training, y represents the true interest state of each sample, h(x) represents the model's predicted value of the user's interest state, x represents the feature data input into the model, x_a represents the feature vector of a specific project, which is used to calculate the predicted value of the success probability of the project, i represents the sample index, and y_i represents the true label of the i-th sample (such as the popularity of the project).

[0112] When constructing interest loss and conformity loss, this embodiment incorporates the concept and technology of causal inference. For example, if a video is not popular but still receives active interaction from users, it often means that users are indeed very interested in the video. This can be verified by reverse logic: if a video is not popular and users do not show interest, then it is unlikely to receive positive feedback.

[0113] In addition, in order to reflect the user's feedback behavior, this embodiment also designs a click loss function. Specifically, this embodiment uses a common pairwise comparison loss method for processing. This method can compare the user's preference for different videos and adjust the model's recommendation strategy accordingly. By comprehensively considering personal interests, herd behavior and user feedback, the model of this embodiment can more comprehensively understand the user's needs and preferences, thereby providing more accurate and personalized recommendation services.

[0114] It should be noted that in order to optimize the integration of user interest features and hot content features and balance personal interests and herd behavior, this embodiment adopts a series of refined strategies to build an efficient and accurate hot recommendation system.

[0115] First, this embodiment establishes an efficient data collection system to capture the user's behavior data such as playing, fast forwarding, replaying, dragging, etc. when watching videos in real time to ensure the integrity and accuracy of the data. These data are pre-processed through an advanced video analysis platform, including data cleaning and feature annotation, and user behavior is converted into vectorized representation for subsequent calculation and analysis. At the same time, this embodiment prepares feature data related to hot spots and performs vectorized processing to conduct in-depth correlation analysis with user behavior characteristics.

[0116] After the data is prepared, this embodiment uses a neural network to build an association model, which takes the user behavior feature vector and the hot spot feature vector as input and outputs the similarity score between the two. By adjusting the optimization algorithm and the loss function, this embodiment continuously trains the model to improve its prediction performance, ensuring that the hot spot content that users are interested in can be accurately recommended in actual applications.

[0117] In order to balance personal interests and herd behavior, this embodiment designs a comprehensive user representation model. This model not only builds personal interest representation based on the user's historical behavior, preference settings and viewing history, but also forms herd behavior representation by analyzing the user's interaction, likes, sharing and other behaviors on social media, combined with popular videos and popular trends. In model training, this embodiment uses a comprehensive loss function system to adjust the weights of the interest loss function and the herd loss function to achieve a precise balance between the two, respecting the user's individual interests and reasonably integrating the herd behavior factor.

[0118] In addition, the recommendation system of this embodiment has the ability to update and adjust in real time. The system will quickly adjust the recommendation strategy based on the user's real-time behavior data and market trends to ensure that the recommended content is always in sync with the user's current interests and needs. At the same time, this embodiment also establishes a feedback update mechanism to continuously optimize the model weights based on the user's actual feedback on the recommended content, thereby improving the accuracy and personalization of the recommendations.

[0119] In summary, by building an efficient data collection system, refined data preprocessing, advanced association models, a comprehensive user representation model, and the ability to update and dynamically adjust in real time, the hot spot recommendation system of this embodiment can respect the user's individual interests while reasonably incorporating the herd behavior factor to provide users with personalized and attractive recommendation content.

[0120] To further illustrate, the optional Figure 3 As shown in Figure 1, it depicts the whole process of how the video feature analysis platform integrates user behavior data to optimize video content recommendation. First, the platform collects a wide range of user behavior data, including user character traits, behavior patterns, emotional reactions, and popular trends. These data are the basis for subsequent analysis.

[0121] Subsequently, the data is input into the video feature analysis platform, which performs in-depth processing through four core modules: first, analyzing user preferences, including user preferences, popularity prediction and model recall, in order to accurately grasp user interests; second, evaluating the status of video hot spots, considering the video, collection and security status, to ensure the quality and compliance of recommended content; third, using advanced models such as Wide&Deep, combining deep learning with breadth learning, to improve the accuracy and diversity of recommendations; fourth, identifying similar IPs, including similar characters, actions and emotions, to enhance the personalization and relevance of recommendations.

[0122] Next, the platform generates feature vectors based on the above analysis. These vectors highly summarize the complex relationship between videos and users and are the key to subsequent recommendation optimization. By continuously monitoring users' playback, fast-forward, replay, drag and other behaviors, the platform can obtain user feedback in real time and dynamically adjust the recommendation system.

[0123] Finally, based on user feedback, the platform continuously optimizes the feed flow, accurately locates and generates multiple hot spots, and ensures that the recommended videos are both in line with user interests and highly attractive and worth watching. This process not only improves the accuracy of the recommendation system, but also significantly enhances user satisfaction and stickiness.

[0124] Through the embodiments provided in this application, the purpose is achieved, thereby realizing the technical effect.

[0125] It is understandable that in the specific implementation of this application, related data such as user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0126] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0127] According to another aspect of the embodiments of the present application, a video segment recommendation device for implementing the above-mentioned video segment recommendation method is also provided. Figure 4 As shown, the device comprises:

[0128] A first acquisition unit 402 is used to acquire the video content when the target user watches the first video and performs the preferred behavior;

[0129] The second acquisition unit 404 is used to acquire a first video feature corresponding to the video content, wherein the first video content is used to represent a key element in the video content;

[0130] The third acquisition unit 406 is used to acquire the correlation relationship between the first video feature and the second video feature, wherein the second video feature is used to represent the key element in the key video segment;

[0131] The fourth acquisition unit 408 is used to acquire the target user's preference for the key video segment in the second video according to the association relationship and the popularity of the key video segment;

[0132] The recommendation unit 410 is configured to recommend the key video segment in the second video to the target user when the preference degree is greater than or equal to a preset threshold.

[0133] For a specific embodiment, reference may be made to the example shown in the method for recommending video clips described above, which will not be described in detail in this example.

[0134] As an optional solution, the device further includes: a fifth acquisition unit, which is used to acquire a first loss function corresponding to the target user and a second loss function corresponding to the target user before acquiring the target user's preference for the key video segment in the second video according to the association relationship and the popularity of the key video segment, wherein the first loss function is used to measure the matching degree between the preference degree and the target user's personal interest, and the second loss function is used to measure the influence of the target user's herd behavior on the preference degree;

[0135] The fourth acquisition unit 408 includes: an acquisition module, which is used to acquire the preference degree according to the first loss function, the second loss function, the popularity, and the association relationship.

[0136] For a specific embodiment, reference may be made to the example shown in the method for recommending video clips described above, which will not be described in detail in this example.

[0137] As an optional solution, obtain modules including:

[0138] A first construction submodule is used to construct a personal interest representation of the target user based on the historical behavior, preference settings and viewing history of the target user to obtain a first loss function;

[0139] The second construction submodule is used to construct a representation of the target user's herd behavior based on the target user's interactive behavior on the social platform and in combination with popular videos to obtain a second loss function.

[0140] For a specific embodiment, reference may be made to the example shown in the method for recommending video clips described above, which will not be described in detail in this example.

[0141] As an optional solution, the third obtaining unit 406 includes:

[0142] The input module is used to input the first video feature and the second video feature into the association model to obtain the feature similarity between the first video feature and the second video feature, wherein the association model is a pre-trained neural network model for analyzing the feature similarity between video features.

[0143] For a specific embodiment, reference may be made to the example shown in the method for recommending video clips described above, which will not be described in detail in this example.

[0144] As an optional solution, the device also includes:

[0145] A feedback module, for obtaining feedback data of a behavior performed by the target user on the key video segment in the second video after the key video segment in the second video is recommended to the target user;

[0146] The optimization module is used to optimize the association model by using the behavior feedback data after recommending the key video clips in the second video to the target user, so as to obtain the optimized association model.

[0147] For a specific embodiment, reference may be made to the example shown in the method for recommending video clips described above, which will not be described in detail in this example.

[0148] As an optional solution, the second obtaining unit 404 includes at least one of the following:

[0149] An identification module, used to identify people in the video content and obtain a first video feature;

[0150] An analysis module, used for analyzing behaviors in video content to obtain a first video feature;

[0151] The judgment module is used to judge the emotion in the video content and obtain the first video feature.

[0152] For a specific embodiment, reference may be made to the example shown in the method for recommending video clips described above, which will not be described in detail in this example.

[0153] As an optional solution, the first acquisition unit 402 includes at least one of the following:

[0154] An execution module, used for acquiring at least one video segment in which a target user is watching a first video and performing a specified playback behavior;

[0155] The preprocessing module is used to preprocess the original data of at least one video clip to remove invalid or abnormal data points to obtain video content.

[0156] For a specific embodiment, reference may be made to the example shown in the method for recommending video clips described above, which will not be described in detail in this example.

[0157] According to another aspect of the embodiment of the present application, an electronic device for implementing the above-mentioned video clip recommendation method is also provided. The electronic device can be but is not limited to being Figure 1 The user device 102 or the server 112 shown in FIG. 1 is used as an example to illustrate the electronic device as the user device 102. Figure 5 As shown, the electronic device includes a memory 502 and a processor 504. The memory 502 stores a computer program, and the processor 504 is configured to execute the steps in any of the above method embodiments through the computer program.

[0158] In an optional embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0159] In an optional embodiment, the processor may be configured to perform the following steps through a computer program:

[0160] S1, obtaining the video content when the target user watches the first video and performs the preferred behavior;

[0161] S2, obtaining a first video feature corresponding to the video content, wherein the first video content is used to characterize a key element in the video content;

[0162] S3, obtaining an association relationship between the first video feature and the second video feature, wherein the second video feature is used to characterize a key element in the key video segment;

[0163] S4, obtaining the target user's preference for the key video segment in the second video according to the association relationship and the popularity of the key video segment;

[0164] S5: When the preference level is greater than or equal to a preset threshold, recommend the key video segment in the second video to the target user.

[0165] Alternatively, a person skilled in the art may understand that: Figure 5 The structure shown is for illustration only. Figure 5 The structure of the electronic device is not limited. Figure 5 More or fewer components (such as network interfaces, etc.) as shown in, or with Figure 5 Different configurations shown.

[0166] Among them, the memory 502 can be used to store software programs and modules, such as the program instructions / modules corresponding to the method and device for recommending video clips in the embodiments of the present application. The processor 504 executes various functional applications and data processing by running the software programs and modules stored in the memory 502, that is, to implement the above-mentioned method for recommending video clips. The memory 502 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 502 may further include a memory remotely located relative to the processor 504, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. Among them, the memory 502 may be specifically, but not limited to, used to store information such as the first video and the second video. As an example, if Figure 5As shown, the memory 502 may include, but is not limited to, the first acquisition unit 402, the second acquisition unit 404, the third acquisition unit 406, the fourth acquisition unit 408, and the recommendation unit 410 in the video clip recommendation device. In addition, other module units in the video clip recommendation device may also be included but are not limited to, which will not be described in detail in this example.

[0167] Optionally, the transmission device 506 is used to receive or send data via a network. Specific examples of the network may include wired networks and wireless networks. In one example, the transmission device 506 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers via a network cable so as to communicate with the Internet or a local area network. In one example, the transmission device 506 is a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0168] In addition, the electronic device further includes: a display 508 for displaying information such as the first video and the second video; and a connection bus 610 for connecting various module components in the electronic device.

[0169] According to one aspect of the present application, a computer program product is provided, the computer program product comprising a computer program / instruction, the computer program / instruction comprising a program code for executing the method shown in the flow chart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium. When the computer program is executed by a central processing unit, various functions provided by the embodiments of the present application are executed.

[0170] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0171] It should be noted that the computer system of the electronic device is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0172] According to one aspect of the present application, a computer-readable storage medium is provided, and a processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the above-mentioned various optional implementations.

[0173] The above is only a preferred implementation 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 method for recommending video clips, characterized in that: include: Obtaining the video content when the target user watches the first video and performs the preferred behavior; Acquire a first video feature corresponding to the video content, wherein the first video content is used to characterize a key element in the video content; Acquire an association relationship between the first video feature and a second video feature, wherein the second video feature is used to characterize a key element in a key video segment; Obtaining the target user's preference for the key video segment in the second video according to the association relationship and the popularity of the key video segment; When the preference degree is greater than or equal to a preset threshold, the key video segment in the second video is recommended to the target user.

2. The method according to claim 1, characterized in that Before obtaining the target user's preference for the key video segment in the second video according to the association relationship and the popularity of the key video segment, the method further includes: obtaining a first loss function corresponding to the target user and a second loss function corresponding to the target user, wherein the first loss function is used to measure the matching degree between the preference degree and the target user's personal interest, and the second loss function is used to measure the influence of the target user's conformity behavior on the preference degree; The step of obtaining the target user's preference for the key video segment in the second video based on the association relationship and the popularity of the key video segment includes: obtaining the preference based on the first loss function, the second loss function, the popularity and the association relationship.

3. The method according to claim 2, characterized in that The obtaining of a first loss function corresponding to the target user and a second loss function corresponding to the target user includes: Based on the historical behavior, preference settings, and viewing history of the target user, construct a personal interest representation of the target user to obtain the first loss function; Based on the interactive behavior of the target user on the social platform and in combination with popular videos, a herd behavior representation of the target user is constructed to obtain the second loss function.

4. The method according to claim 1, characterized in that: The obtaining the association relationship between the first video feature and the second video feature includes: The first video feature and the second video feature are input into an association model to obtain feature similarity between the first video feature and the second video feature, wherein the association model is a pre-trained neural network model for analyzing feature similarity between video features.

5. The method according to claim 4, characterized in that After recommending the key video segment in the second video to the target user, the method further includes: Acquire behavioral feedback data of the target user on the key video segment in the second video; The association model is optimized using the behavior feedback data to obtain an optimized association model.

6. The method according to any one of claims 1 to 5, characterized in that The obtaining of the first video feature corresponding to the video content includes at least one of the following: Identify characters in the video content to obtain the first video feature; Analyzing the behavior in the video content to obtain the first video feature; The emotion in the video content is determined to obtain the first video feature.

7. The method according to any one of claims 1 to 5, characterized in that The obtaining of the video content when the target user watches the first video and performs the preferred behavior includes at least one of the following: Acquire at least one video segment in which the target user watches the first video and performs a specified playback behavior; The original data of the at least one video clip is preprocessed to remove invalid or abnormal data points to obtain the video content.

8. A video clip recommendation device, characterized in that: include: A first acquisition unit, used to acquire video content when a target user watches a first video and performs a preferred behavior; A second acquisition unit is used to acquire a first video feature corresponding to the video content, wherein the first video content is used to represent a key element in the video content; A third acquisition unit is used to acquire an association relationship between the first video feature and a second video feature, wherein the second video feature is used to represent a key element in a key video segment; a fourth acquisition unit, configured to acquire the target user's preference for the key video segment in the second video according to the association relationship and the popularity of the key video segment; A recommendation unit is used to recommend the key video clip in the second video to the target user when the preference degree is greater than or equal to a preset threshold.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method described in any one of claims 1 to 7 when executed by an electronic device.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 7 through the computer program.

Images