Resource recommendation method and device, electronic equipment and storage medium

Abstract

The invention provides a resource recommendation method and device, electronic equipment and a storage medium, relates to the technical field of artificial intelligence, in particular to the fields of deep learning models, large models and big data, and can be applied to scenes such as intelligent search and intelligent recommendation application. The specific implementation scheme is as follows: according to at least one of an image in a resource, a script text for the resource, a first object associated with the resource and a resource attribute of a reference resource with the same resource category, determining a multi-modal feature for the resource; determining an interaction feature sequence according to historical interaction data generated by a second object for the resource in at least one time period; determining a popularity evaluation value for the resource according to the basic feature, the multi-modal feature and the interaction feature sequence of the resource; and performing resource recommendation according to the popularity evaluation value.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, particularly deep learning models, large models, and big data, and can be used in scenarios such as intelligent search and intelligent recommendation applications. More specifically, the present disclosure provides a resource recommendation method, apparatus, electronic device, storage medium, and computer program product. Background Art

[0002] Users watch short and long dramas in their daily lives. Recommendation systems can improve user experience by recommending resources that match their preferences, but the effectiveness of recommendation systems needs to be improved. Summary of the Invention

[0003] The present disclosure provides a resource recommendation method, apparatus, electronic device, storage medium, and computer program product.

[0004] According to one aspect of the present disclosure, a resource recommendation method is provided, comprising: determining multimodal features for a resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and resource attributes of a reference resource of the same category as the resource; determining an interaction feature sequence based on historical interaction data generated by a second object for the resource in at least one time period; determining a popularity evaluation value for the resource based on the basic features, multimodal features, and interaction feature sequence of the resource; and recommending the resource based on the popularity evaluation value.

[0005] According to another aspect of the present disclosure, a resource recommendation device is provided, comprising: a multimodal feature determination module, a sequence determination module, an evaluation value determination module, and a recommendation module. The multimodal feature determination module is used to determine the multimodal features for a resource based on at least one of the image in the resource, the script text for the resource, the first object associated with the resource, and the resource attributes of a reference resource of the same resource category. The sequence determination module is used to determine the interaction feature sequence based on the historical interaction data generated by the second object for the resource in at least one time period. The evaluation value determination module is used to determine the heat evaluation value for the resource based on the basic features, multimodal features, and interaction feature sequence of the resource. The recommendation module is used to recommend resources based on the heat evaluation value.

[0006] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method provided by the present disclosure.

[0007] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute the method provided by the present disclosure.

[0008] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements the method provided in the present disclosure when executed by a processor.

[0009] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings are used to better understand the present invention and do not constitute a limitation of the present invention.

[0011] Figure 1 is a schematic diagram of an application scenario of the resource recommendation method and device according to an embodiment of the present disclosure;

[0012] Figure 2 is a schematic flow chart of a resource recommendation method according to an embodiment of the present disclosure;

[0013] Figure 3 is a schematic diagram of a resource recommendation method according to an embodiment of the present disclosure;

[0014] Figure 4 is a schematic structural block diagram of a resource recommendation device according to an embodiment of the present disclosure; and

[0015] Figure 5 It is a structural block diagram of an electronic device used to implement the resource recommendation method of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0016] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0017] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0018] In the technical solution disclosed herein, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0019] It should be noted that the current recommendation system lacks the scientific and real-time reasoning ability to determine whether a resource will become a hit resource, and is unable to accurately estimate the future popularity of a resource, resulting in low recommendation accuracy and poor recommendation effects.

[0020] The disclosed embodiments aim to provide a resource recommendation method, which determines multimodal features for a resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and resource attributes of a reference resource of the same resource category. Furthermore, an interaction feature sequence is determined based on historical interaction data generated by a second object for the resource within at least one time period. Subsequently, a heat evaluation value for the resource is determined based on the basic features, multimodal features, and interaction feature sequence of the resource, and resource recommendations are made based on the heat evaluation value.

[0021] It is understandable that multimodal features can characterize the characteristics of resources from the dimensions of images, texts, and the first object, thereby more accurately reflecting the attractiveness of resources to users. In addition, the interaction feature sequence can reflect the attractiveness of resources to users from the interaction between the resource and the second object, and when the historical interaction data is recent data, the interaction feature sequence determined based on the historical interaction data can make the popularity evaluation value have a certain real-time nature. Therefore, the resource recommendation method provided by the embodiment of the present disclosure can enable the recommendation system to have accurate and real-time reasoning capabilities, thereby accurately estimating the attractiveness of resources to users, determining the popularity of resources in the future, improving the recommendation accuracy of the recommendation system, and improving the recommendation effect.

[0022] The technical solution provided by the embodiments of the present disclosure can be applied to the recommendation system to recommend short dramas, long dramas, videos released by authors on the platform, variety shows and other resources to users. It is particularly suitable for short drama content recommendation and traffic distribution scenarios on short video platforms.

[0023] The technical solutions provided by the present disclosure will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Figure 1 Schematic diagram of an application scenario of the resource recommendation method and device according to an embodiment of the present disclosure.

[0025] It should be noted that Figure 1 The examples shown are merely examples of system architectures to which the embodiments of the present disclosure may be applied, to help those skilled in the art understand the technical content of the present disclosure, but do not mean that the embodiments of the present disclosure may not be used in other devices, systems, environments or scenarios.

[0026] like Figure 1As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium for providing communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0027] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Terminal devices 101, 102, and 103 can be various electronic devices with display screens and support web browsing, including but not limited to smartphones, tablet computers, laptop computers, and desktop computers, etc.

[0028] The server 105 may be a server that provides various services, such as a background management server (for example only) that provides support for websites that users browse using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results to the terminal device. In addition, in actual applications, the server 105 may deploy a model to determine the popularity evaluation value of a resource, and determine a target resource that can be recommended to the user from multiple resources based on the popularity evaluation value. For example, the server 105 determines the popularity evaluation value based on the user's resource acquisition request, and then determines the target resource based on the popularity evaluation value, and feeds back the target resource to the terminal device.

[0029] It should be noted that the resource recommendation method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the resource recommendation device provided in the embodiment of the present disclosure can generally be set in the server 105. The resource recommendation method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the resource recommendation device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0030] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0031] Figure 2 is a schematic flowchart of a resource recommendation method according to an embodiment of the present disclosure.

[0032] like Figure 2As shown, the resource recommendation method 200 may include operations S210 to S240.

[0033] In operation S210 , a multimodal feature for a resource is determined based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource.

[0034] Resources can be short plays, long plays, videos published by authors on the platform, variety shows, and other video resources. Resources include images and script text. Images can be extracted from video resources, and script text can be pre-configured text or text converted based on audio in the resources.

[0035] The first object associated with the resource may be a user who participates in creating the resource, such as an actor, director, screenwriter, singer, etc.

[0036] The reference resource has the same category as the resource for which the popularity assessment is to be determined. The category can represent the subject matter of the work, such as ancient style or historical fiction. Resource attributes of the reference resource include, for example, distribution volume, likes, collections, and the relationship between distribution volume and time.

[0037] Feature extraction may be performed on at least one of the image, the script text, the first object, and the resource attributes of the reference resource to extract corresponding features, thereby obtaining multimodal features.

[0038] In operation S220 , an interaction feature sequence is determined based on historical interaction data generated by the second object with respect to the resource within at least one time period.

[0039] The second object may be a user who views the resource, such as a user who follows a TV series, etc. The time period may be a period of time before the current moment. For example, there are three time periods: the past hour, the past hour to the past two hours, and the past two hours to the past three hours.

[0040] Historical interaction data, for example, includes at least one of the following: distribution volume, comment volume, like volume, attention volume, forwarding volume, and average playback time per person. The historical interaction data of each time period can form an interaction data sequence. For example, the interaction data sequence includes: historical interaction data for the past hour, historical interaction data from the past hour to the past two hours, and historical interaction data from the past two hours to the past three hours. The historical interaction data for each time period includes distribution volume, comment volume, and like volume. Feature extraction can be performed on the interaction data sequence to obtain an interaction feature sequence. The acquisition and use of historical interaction data are known and agreed upon by all users, and are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0041] In operation S230 , a popularity evaluation value for the resource is determined based on the basic features, multimodal features, and interaction feature sequence of the resource.

[0042] For example, a resource's basic features, multimodal features, and interactive feature sequences can be input into a pre-trained popularity assessment model, which outputs a popularity assessment value. The popularity assessment model can be a deep learning model, and this embodiment does not limit the popularity assessment model.

[0043] It should be noted that the popularity assessment model is pre-trained. For example, during training, the basic features of the sample resources, the multimodal features of the sample resources, and the interaction feature sequences derived from historical interaction sequence samples can be input into the trained popularity assessment model, which then outputs an estimated evaluation value. The true evaluation value can be pre-configured. A loss value can be calculated based on the difference between the estimated and true evaluation values, and the parameters of the popularity assessment model can be adjusted based on the loss value.

[0044] In operation S240 , resource recommendations are made based on the popularity evaluation values.

[0045] For example, you can sort multiple resources in the database by popularity evaluation value, select resources that are ranked before the predetermined order as target resources, and then display the target resources on the front-end page. Alternatively, you can select resources with popularity evaluation values ​​greater than the evaluation value threshold as target resources and display the target resources on the front-end page. Alternatively, you can add resources with popularity evaluation values ​​greater than the evaluation value threshold to the popular resource pool to increase the resource's exposure.

[0046] According to the technical solution provided by the embodiment of the present disclosure, multimodal features can characterize the characteristics of resources from the dimensions of images, texts, and the first object, thereby more accurately reflecting the attractiveness of resources to users. In addition, the interaction feature sequence can reflect the attractiveness of resources to users from the interaction between the resource and the second object, and when the historical interaction data is recent data, the interaction feature sequence determined based on the historical interaction data can make the popularity evaluation value have a certain degree of real-time performance. Therefore, the resource recommendation method provided by the embodiment of the present disclosure can enable the recommendation system to have accurate and real-time reasoning capabilities, thereby accurately estimating the attractiveness of resources to users, determining the popularity of resources in the future, improving the recommendation accuracy of the recommendation system, and improving the recommendation effect.

[0047] Next, the process of determining the basic characteristics of a resource is described.

[0048] In one example, basic features can be determined based on basic attributes of a resource. Basic attributes include at least one of the following: resource name, resource category, resource identifier, and identifier of a video within the resource. For example, if the resource is a short play, the resource identifier is the short play identifier, and the video identifier within the resource is the identifier of a single episode.

[0049] In another example, the basic feature can be determined according to a posteriori attributes of the resource, wherein the a posteriori attributes include at least one of the following: fast scrolling rate, satisfaction rate, and continuous play rate.

[0050] The fast-scrolling rate can be determined based on the number of times the duration of a single resource playback is less than the first predetermined duration and the total number of times the resource is played. For example, if the duration of a single resource playback by a user is less than 3 seconds, it indicates that a fast-scrolling has occurred. If the number of fast-scrolling events is 3 and the total number of times the resource is played is 10, the fast-scrolling rate can be 30%. It should be noted that the first predetermined duration can be related to the duration of the resource. For example, for a long video, the first predetermined duration can be 3 minutes. For a short video, the first predetermined duration can be 3 seconds.

[0051] The satisfaction rate can be determined based on the number of satisfactory interactions between the second object and the resource and the total number of plays. For example, if the second object plays the resource for longer than the second predetermined duration, such as a user playing a video for longer than 10 minutes, this indicates satisfactory playback. Alternatively, if the second object generates positive interactions with the resource, such as sharing, liking, and collecting, this indicates satisfactory playback. If the number of satisfactory plays is 20 and the total number of resource plays is 100, the satisfaction rate can be 20%.

[0052] The extended play rate can be determined based on the number of extended plays and the total number of plays. Extended play indicates that after playing a video in the resource, the second object continues to play other videos in the resource. For example, if a short series consists of 50 episodes and the user plays episode 10 and then continues to play episode 11, episode 1, or another episode, then extended play has occurred. If the number of extended plays is 15 and the total number of plays for all 50 episodes in the short series is 100, the extended play rate is 15%.

[0053] It can be seen that the above-mentioned posterior attributes characterize the behavior of the second object during the resource playback process. These behaviors can reflect the degree of attraction of the resource to the second object. Therefore, based on these posterior attributes, the heat evaluation value of the resource can be determined more accurately, thereby improving the accuracy of the recommendation.

[0054] In another example, the basic features may be determined according to the basic attributes and a posteriori attributes of the resource.

[0055] In practical applications, a pre-trained feature extraction network can be used to extract features. For example, at least one of the basic attributes and the posterior attributes of a resource is input into the feature extraction network, and the basic features are output.

[0056] The above describes the process of determining the basic characteristics of a resource.

[0057] Next, the process of determining the multimodal characteristics of a resource is described.

[0058] In one example, the multimodal feature includes a script sentiment sub-feature. The script sentiment sub-feature is determined by dividing the script text into multiple text sets based on the scenes of the script text, and then determining the script sentiment sub-feature based on the sentiment categories of each of the multiple text sets.

[0059] For example, a text set may include one or more sentence texts. The text set can be input into a pre-trained model to obtain a text representation. The text representation can then be input into a classification network for multi-classification to obtain the emotional categories of the text set. Emotional categories include, for example, love, dependence, competition, misunderstanding, etc. The above-mentioned pre-trained model is, for example, BERT (Bidirectional Encoder Representations from Transformers, a bidirectional encoder based on Transformers), and the classification network can use a fully connected network. After obtaining multiple emotional categories for each text set, several emotional categories with confidence levels higher than a confidence threshold can be used as the emotional tendencies of the script. Feature extraction of the emotional tendencies of the script can obtain sub-features of the emotional tendencies of the script.

[0060] This example determines the sentiment of a script. It's understandable that a script's sentiment influences user preferences for resources. For example, some users prefer short dramas about romance, while others prefer short dramas about business competition. Therefore, using the script's sentiment sub-feature can more accurately determine popularity assessments.

[0061] In another example, the multimodal feature includes a conflict density sub-feature. This conflict density sub-feature is determined by segmenting the script text into multiple segmented words, then determining keywords representing the conflict content from the multiple segmented words, and then determining the conflict density sub-feature based on the number of keywords and the number of the multiple segmented words.

[0062] For example, the script text is divided into multiple segments through word segmentation processing, and then it is determined whether the segmentation is consistent with the conflict words in the predetermined vocabulary, such as "quarrel", "confrontation", etc. If consistent, the segmentation is determined as a keyword, otherwise it is determined as a non-keyword. Then the ratio between the number of keywords and the number of multiple segmentations included in the script text can be determined, and the ratio represents the frequency of occurrence of conflict words. The frequency of occurrence of conflict words can be used as conflict density. Alternatively, the attention weight of the conflict words can be analyzed based on the attention mechanism, and the product of the frequency of occurrence of the conflict words and the attention weight of the conflict words can be used as the conflict density. The conflict density can then be feature extracted to obtain conflict density sub-features.

[0063] The multimodal features in this example include a conflict density sub-feature. Resources with a moderate conflict density are more likely to attract users. Therefore, the popularity evaluation value can be determined more accurately based on the conflict density sub-feature.

[0064] In another example, the multimodal feature includes a sub-feature of interlude spread. This interlude spread sub-feature is determined by: identifying at least one similar line in a dialogue library that satisfies a similarity relationship with the dialogue text in the script. Then, based on the mapping relationship between the similar line and the resource, the number of resources containing the at least one similar line is determined. Based on the number of resources, the interlude spread sub-feature is then determined.

[0065] For example, the dialogue text is input into a pre-trained model to obtain the textual representation of the dialogue text. A dialogue library can be pre-constructed, and the dialogue library includes dialogue texts from multiple resources. The similarity between the textual representation of the dialogue text in the script text and the textual representation of the dialogue text in the dialogue library can be calculated online or offline. If the similarity is higher than a first similarity threshold, it means that the dialogue text in the resource script text and the dialogue text in the dialogue library satisfy a similarity relationship, and the dialogue text in the dialogue library is a similar dialogue. In addition, a mapping relationship between the dialogue text and the resource in the dialogue library is also pre-constructed, and the dialogue text and the resource have a mapping relationship indicating that the dialogue text is derived from the resource. In this way, the number of resources including similar dialogues can be determined. The ratio between the number of resources including similar dialogues and the predetermined number of resources can be used as the repetition rate between plays, and the predetermined number of resources can be the total number of resources in the resource library.

[0066] It can be seen that the greater the number of resources or the greater the ratio of the number of resources to the total number of resources, the more resources contain lines similar to the lines in the script text, and the more widely the lines are spread. In this example, the number of resources or the repetition rate between plays can be used as the spreadability data of the lines between plays. Feature extraction can be performed on the spreadability data of the lines between plays to obtain the spreadability sub-features of the lines between plays. It can be understood that the wider spread of lines indicates that the line text is easy to attract users. Therefore, based on the spreadability sub-features of the lines between plays, the popularity evaluation value can be more accurately determined.

[0067] In another example, the multimodal feature includes a sub-feature of the spread of lines within a play. This sub-feature is determined by determining the repetition rate between multiple lines in the script based on their similarities, and then determining the sub-feature of the spread of lines based on the repetition rate.

[0068] For example, for any two dialogue texts in a script, if the similarity between the two dialogue texts exceeds a second similarity threshold, the two dialogue texts can be determined to be duplicates. The duplicate dialogue texts can be added to a first set, and all dialogue texts in the script text can be added to a second set. The ratio between the number of dialogue texts in the first set and the number of dialogue texts in the second set can be calculated, representing the repetition rate between the dialogue texts in the script text. Feature extraction can be performed on the repetition rate to obtain a sub-feature of the spread of dialogue texts within the play.

[0069] It is understandable that the repetition rate of lines in the same resource can indicate the quality of the resource. For example, a high repetition rate indicates that the resource has problems such as a rough script and a single theme expression. Therefore, the popularity evaluation value can be determined more accurately based on the propagation characteristics of the lines in the play.

[0070] In another example, the multimodal feature includes a picture rhythm sub-feature. This picture rhythm sub-feature is determined by extracting multiple images from a video in a resource to obtain an image sequence. Then, for any two adjacent images in the image sequence, a determination is made as to whether the similarity between the adjacent images is less than or equal to an image similarity threshold. If so, a shot cut is determined to have occurred; otherwise, no shot cut is determined to have occurred. The picture rhythm sub-feature is then determined based on the number of shot cuts.

[0071] For example, a resource includes multiple videos. Taking a short play as an example, a video can be an episode of the short play. For the videos in the resource, multiple images can be extracted from the video clip between the starting moment and the predetermined moment of the video to obtain an image sequence. The predetermined moment can be 5 seconds, that is, multiple images are extracted from the first 5 seconds of an episode of the short play, and images can be extracted every predetermined number of frames during the extraction process. It should be noted that for short plays, the attractiveness of the video to users mainly depends on the first few seconds of the video clip at the beginning of the video. Therefore, by extracting images from the video clip between the starting moment and the predetermined moment, images that have a greater impact on the popularity can be extracted, thereby accurately evaluating the popularity of the resource.

[0072] For example, after extracting the image sequence, the similarity between two adjacent image frames can be calculated. If so, it means that the image scene has changed significantly, so a shot switch has occurred. In this way, the number of shot switches in the image sequence can be determined.

[0073] Next, the picture rhythm can be determined by the number of shot cuts. For example, the number of shot cuts per unit time can be calculated to obtain the cut frequency, which can be used as the picture rhythm data. Feature extraction can be performed on the picture rhythm data to obtain picture rhythm sub-features.

[0074] This example uses inter-frame similarity fluctuations to reflect the tempo of a video. For example, high fluctuations indicate a fast tempo, allowing for an accurate assessment of the tempo. It's understandable that a slow tempo indicates bland content, while a fast tempo indicates an excessive amount of information presented in a short period of time. Therefore, tempo can impact a resource's appeal to users. Therefore, using the tempo sub-feature allows for a more accurate determination of popularity.

[0075] In another example, a multimodal feature includes an associated sub-feature. The associated sub-feature is determined by obtaining a graph, the graph including multiple nodes, the multiple nodes including a subject node and a cast node. Based on the subject of the resource and the first object, a target subject node and a target cast node are determined from the multiple nodes. Then, based on the target subject node and the target cast node, associated nodes are determined from the graph. Next, based on the associated nodes, associated sub-features are determined.

[0076] A graph can be pre-built based on the genres and cast members of multiple resources. Nodes in the graph include genre nodes and cast nodes. Cast nodes include actor nodes, director nodes, and screenwriter nodes. Associated nodes are connected by edges. For example, if an actor and director participated in the same skit, the actor node representing the actor and the director node representing the director are connected by an edge. For another example, if an actor participated in a comedy skit, the actor node representing the actor and the genre node representing the comedy genre are connected by an edge.

[0077] For a resource whose popularity evaluation value needs to be determined, the subject matter and cast and crew of the resource are known, so at least one of the target subject matter node and the target cast and crew node can be determined from the graph, where the target subject matter node represents the subject matter of the resource, and the target cast and crew node represents the first object associated with the resource.

[0078] Next, associated nodes can be determined from the graph based on at least one of the target subject matter node and the target cast member node. For example, the shortest path between the node and the target subject matter node or the target cast member node can be determined. If the number of edges included in the shortest path is less than or equal to a predetermined number of edges, the node is determined to be an associated node. Alternatively, each node in the graph can be vectorized into a node vector, and then the similarity between the node vector of the node in the graph and the node vector of the target subject matter node or the target cast member node can be calculated. If the similarity is greater than or equal to a node similarity threshold, the node is determined to be an associated node. Feature extraction can be performed on the attributes of the associated nodes to obtain associated sub-features.

[0079] This example first constructs a graph and determines associated nodes based on the graph. The associated nodes can reflect other subject matter and cast and crew information related to the subject matter and cast and crew of the resource. This information can assist in determining the popularity evaluation value of the resource and improve the accuracy of the popularity evaluation value.

[0080] In another example, the multimodal feature includes an influence sub-feature. The influence sub-feature is determined in the following manner: based on at least one of the number of works and the number of fans of the first object, an influence sub-feature that characterizes the influence of the first object is determined. For example, a first ratio between the number of works of the first object and the total number of resources can be calculated, or a second ratio between the number of fans of the first object and the total number of fans of the first object associated with all resources can be calculated. The first ratio and the second ratio can be used as influence data of the first object. Feature extraction can be performed on the influence data to obtain the influence sub-feature. It can be understood that the large number of works and fans of the first object indicates that the first object itself has a certain degree of popularity, so the resources in which the first object participates in the creation are more likely to generate popularity. Therefore, determining the popularity evaluation value based on the influence sub-feature can improve the accuracy of the popularity evaluation value, thereby improving the recommendation effect.

[0081] In another example, the multimodal feature includes a lifecycle sub-feature, which is determined in the following manner: determining the lifecycle sub-feature based on the relationship between the playback volume and time of the reference resource.

[0082] For example, the reference resource and the resource for which popularity evaluation is to be predicted share the same category. The category can represent the subject matter of the work, such as classical Chinese or historical fiction. The lifecycle data of the reference resource can be determined, for example, using time as the horizontal axis and play count as the vertical axis, with the origin of the horizontal axis representing the initial distribution time of the reference resource, thereby determining the relationship between play count and time. Feature extraction can be performed on the lifecycle data to obtain lifecycle sub-features.

[0083] In this example, since the reference resource and the resource to be processed share the same theme, the lifecycle of the reference resource provides a reference for the popularity of the resource to be processed. Therefore, determining the popularity evaluation based on the lifecycle sub-feature can improve the accuracy of the popularity evaluation and, in turn, enhance the effectiveness of recommendations.

[0084] In other examples, multimodal features can also include at least one of textual semantic sub-features and visual semantic sub-features. The script text and multiple images from the resource can be fed into the CLIP (Contrastive Language-Image Pretraining) model to obtain high-dimensional semantic features of the images and visuals, which serve as textual semantic sub-features and visual semantic sub-features. These sub-features contain information such as plot material, plot, and color saturation.

[0085] It should be noted that the multimodal features include at least one of the above-mentioned script emotional tendency sub-features, conflict density sub-features, inter-play dialogue communication sub-features, intra-play dialogue communication sub-features, picture rhythm sub-features, association sub-features, influence sub-features, life cycle sub-features, text semantic sub-features and visual semantic sub-features. After obtaining each sub-feature, the sub-features can be fused by splicing or other means, and the fused features can be used as multimodal features.

[0086] The above describes the process of determining the multimodal characteristics of a resource.

[0087] Figure 3 It is a schematic diagram of the resource recommendation method according to an embodiment of the present disclosure.

[0088] In this embodiment, the interaction data sequence 301 can be pre-constructed, and then the interaction feature sequence 302 can be determined. For example, the historical interaction data of the T period, the T-1 period, and the T-2 period can be constructed as the interaction data sequence 301. The T period can represent the period from the current moment to the past hour, the T-1 period can represent the period from the past 1 hour to the past 2 hours, and the T-2 period can represent the period from the past 2 hours to the past 3 hours. The historical interaction data for each period can include the distribution volume, comment volume, like volume, attention volume, forwarding volume, average playback time per person, etc. of the resource in that period. Next, the features of the interaction data sequence 301 can be extracted using a Transformer network structure or other network structure to obtain the interaction feature sequence 302. In addition, the interaction data sequence 301 can be updated regularly, for example, once every hour, to perform real-time modeling of user behavior feedback to ensure the real-time nature of the interaction data sequence 301, and then dynamically adjust the inference results of the heat evaluation value 310.

[0089] Furthermore, basic features 304 of the resource can be determined based on the basic attributes 303 of the resource, and multimodal features 306 of the resource can be determined based on multimodal data 305, such as images within the resource, script text for the resource, a first object associated with the resource, and resource attributes of reference resources of the same resource category. The method for determining basic features 304 and multimodal features 306 can be found above and will not be further described in this embodiment. Basic features 304 and multimodal features 306 can be determined using data from the T period or the T-1 period.

[0090] Next, feature weights for basic features 304, multimodal features 306, and interaction feature sequence 302 can be determined based on historical interaction data 307 within a predetermined time period. For example, historical interaction data 307 within a predetermined time period can be interaction data from a past period of time, such as interaction data from the past day, past 12 hours, or past hour. For example, historical interaction data 307 within a predetermined time period can be input into a pre-trained gating network 309, which then outputs specific feature weights, such as 0.1, 0.7, and 0.2.

[0091] Next, a popularity evaluation value 310 may be determined based on the resource's basic features 304, multimodal features 306, interactive feature sequence 302, and feature weights. For example, the basic features 304, multimodal features 306, interactive feature sequence 302, and their respective feature weights may be input into a popularity evaluation model 308, which then outputs a popularity evaluation value 310. Alternatively, the basic features 304, multimodal features 306, and interactive feature sequence 302 may be multiplied by their respective feature weights to obtain a plurality of processed features, which are then input into the popularity evaluation model 308, which then outputs a popularity evaluation value 310.

[0092] It should be noted that, in the actual training process, the heat evaluation model 308 and the gating network 309 can be trained simultaneously. For example, the historical interaction data samples are input into the gating network 309 to be trained, and the estimated weights of each feature are output. Then, the basic features of the sample resources, the multimodal features of the sample resources, the interaction feature sequence obtained based on the historical interaction sequence samples, and the respective estimated weights are input into the heat evaluation model 308 to obtain an estimated evaluation value. The loss value is then calculated based on the difference between the estimated evaluation value and the true evaluation value, and the parameters of the heat evaluation model 308 and the gating network 309 are adjusted based on the loss value until the heat evaluation model 308 and the gating network 309 converge.

[0093] In this embodiment, the feature weights of each feature are adjusted based on the second object's interactions over the past period of time, and then the popularity evaluation value 310 is determined. In this way, the second object's interactions over the past period of time can strengthen some features in the basic features 304, multimodal features 306, and interactive feature sequence 302, while weakening other features, thereby making the popularity evaluation value 310 determined based on these features more accurate and improving the accuracy of recommendations.

[0094] Next, the process of recommending resources based on popularity evaluation values ​​is explained.

[0095] In this embodiment, the initial evaluation value of the resource can be adjusted according to the popularity evaluation value to obtain a target evaluation value. Then, resource recommendations are made based on the target evaluation values ​​of the multiple resources.

[0096] For example, the initial evaluation value of the resource can be determined using other evaluation algorithms. For example, the evaluation algorithm can calculate the degree of match between the second object and the resource and use the degree of match as the initial evaluation value. Next, the initial evaluation value of the resource is adjusted using the heat evaluation value. For example, the heat evaluation value and the initial evaluation value are weighted, added, multiplied, or other operations are performed to obtain the operation result, which is used as the target evaluation value. Resource recommendations are then made based on the target evaluation value, for example, recommending resources whose target evaluation value is greater than or equal to a threshold.

[0097] This embodiment adjusts the initial evaluation value based on the popularity evaluation value, which can accurately determine the popularity of the resource, thereby improving the accuracy of resource recommendation.

[0098] The above describes the process of recommending resources based on popularity evaluation values.

[0099] Figure 4 It is a schematic structural block diagram of a resource recommendation device according to an embodiment of the present disclosure.

[0100] like Figure 4 As shown, the resource recommendation device 400 may include a multimodal feature determination module 410 , a sequence determination module 420 , an evaluation value determination module 430 and a recommendation module 440 .

[0101] The multimodal feature determination module 410 is used to determine the multimodal features for the resource based on at least one of the image in the resource, the script text for the resource and the first object associated with the resource, and the resource attributes of the reference resource with the same resource category.

[0102] The sequence determination module 420 is configured to determine an interaction feature sequence based on historical interaction data generated by the second object with respect to the resource within at least one time period.

[0103] The evaluation value determination module 430 is used to determine the popularity evaluation value for a resource based on the basic features, multimodal features, and interaction feature sequence of the resource.

[0104] The recommendation module 440 is used to recommend resources based on the popularity evaluation value.

[0105] According to another embodiment of the present disclosure, a multimodal feature includes a script sentiment sub-feature; and a multimodal feature determination module includes a partitioning sub-module and a sentiment sub-feature determination sub-module. The partitioning sub-module is configured to partition the script into multiple text sets based on the script's scenes. The sentiment sub-feature determination sub-module is configured to determine the script sentiment sub-feature based on the sentiment categories of the multiple text sets.

[0106] According to another embodiment of the present disclosure, a multimodal feature includes a conflict density sub-feature; and a multimodal feature determination module includes a segmentation sub-module, a keyword determination sub-module, and a conflict density sub-feature determination sub-module. The segmentation sub-module is used to segment the script text into multiple segmented words. The keyword determination sub-module is used to determine keywords representing the conflicting content from the multiple segmented words. The conflict density sub-feature determination sub-module is used to determine the conflict density sub-feature based on the number of keywords and the number of multiple segmented words.

[0107] According to another embodiment of the present disclosure, the multimodal feature includes a sub-feature of inter-play dialogue spreadability; and the multimodal feature determination module includes: a similar line determination sub-module, a quantity determination sub-module, and a sub-feature determination sub-module of inter-play dialogue spreadability. The similar line determination sub-module is used to determine, for the dialogue text in the script text, at least one similar line in the dialogue library that satisfies a similarity relationship with the dialogue text. The quantity determination sub-module is used to determine the number of resources that include at least one similar line based on the mapping relationship between similar lines and resources. The sub-feature determination sub-module of inter-play dialogue spreadability is used to determine the sub-feature of inter-play dialogue spreadability based on the number of resources.

[0108] According to another embodiment of the present disclosure, the multimodal features include a sub-feature of the spreadability of lines within a play; and the multimodal feature determination module includes a repetition rate determination sub-module and a sub-module for determining the spreadability of lines within a play. The repetition rate determination sub-module is configured to determine the repetition rate between multiple lines within the script text based on their similarities. The sub-module for determining the spreadability of lines within a play sub-feature is configured to determine the spreadability of lines within a play based on the repetition rate between the multiple lines.

[0109] According to another embodiment of the present disclosure, a multimodal feature includes a picture rhythm sub-feature; and a multimodal feature determination module includes: an extraction sub-module, a switch determination sub-module, and a picture rhythm sub-feature determination sub-module. The extraction sub-module is used to extract multiple images from a video in a resource to obtain an image sequence. The switch determination sub-module is used to determine that a shot cut has occurred when it is determined that the similarity between adjacent images in the image sequence is less than or equal to an image similarity threshold. The picture rhythm sub-feature determination sub-module is used to determine the picture rhythm sub-feature based on the number of shot cuts.

[0110] According to another embodiment of the present disclosure, the extraction submodule includes: an extraction unit for extracting multiple images from a video segment between a start time and a predetermined time of the video in the resource to obtain an image sequence.

[0111] According to another embodiment of the present disclosure, the multimodal feature includes an associated sub-feature; the multimodal feature determination module includes: an acquisition sub-module, a first node determination sub-module, a second node determination sub-module and an associated sub-feature determination sub-module. The acquisition sub-module is used to acquire a graph, the graph includes multiple nodes, and the multiple nodes include subject nodes and cast nodes. The first node determination sub-module is used to determine a target subject node and a target cast node from multiple nodes based on the subject of the resource and the first object. The second node determination sub-module is used to determine an associated node from the graph based on the target subject node and the target cast node. The associated sub-feature determination sub-module is used to determine an associated sub-feature based on the associated node.

[0112] According to another embodiment of the present disclosure, the multimodal feature includes an influence sub-feature; the multimodal feature determination module includes: an influence sub-feature determination sub-module, which is used to determine the influence sub-feature characterizing the influence of the first object based on at least one of the number of works and the number of fans of the first object.

[0113] According to another embodiment of the present disclosure, the multimodal feature includes a life cycle sub-feature; the multimodal feature determination module includes: a life cycle sub-feature determination sub-module, which is used to determine the life cycle sub-feature based on the relationship between the playback volume and time of the reference resource.

[0114] According to another embodiment of the present disclosure, a popularity evaluation value determination module includes: a weight determination submodule and an evaluation value determination submodule. The weight determination submodule is used to determine the feature weights of basic features, multimodal features, and interaction feature sequences based on historical interaction data within a predetermined time period. The evaluation value determination submodule is used to determine the popularity evaluation value of a resource based on its basic features, multimodal features, interaction feature sequences, and feature weights.

[0115] According to another embodiment of the present disclosure, the basic features are determined based on at least one of the basic attributes and a posteriori attributes of the resource. The basic attributes include at least one of the following: resource name, resource category, resource identifier, and identifier of the video in the resource. The a posteriori attributes include at least one of the following: fast scrolling rate, satisfaction rate, and continued playback rate; wherein the fast scrolling rate is determined based on the number of times the duration of a single playback of the resource is less than a predetermined duration and the total number of playbacks of the resource; the satisfaction rate is determined based on the number of satisfactory interactive behaviors generated between the second object and the resource and the total number of playbacks; the continued playback rate is determined based on the number of continued playbacks and the total number of playbacks, and continued playback characterizes that the second object continues to play other videos in the resource after playing a video in the resource.

[0116] According to another embodiment of the present disclosure, the historical interaction data includes at least one of the following: distribution volume, comment volume, like volume, attention volume, forwarding volume, and average playback time per person.

[0117] According to another embodiment of the present disclosure, the recommendation module includes an adjustment submodule and a recommendation submodule. The adjustment submodule is configured to adjust the initial evaluation value of a resource based on the popularity evaluation value to obtain a target evaluation value. The recommendation submodule is configured to recommend resources based on the target evaluation values ​​of multiple resources.

[0118] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, including at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned resource recommendation method.

[0119] According to an embodiment of the present disclosure, the present disclosure further provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the above-mentioned resource recommendation method.

[0120] According to an embodiment of the present disclosure, the present disclosure further provides a computer program product, including a computer program, which implements the above-mentioned resource recommendation method when executed by a processor.

[0121] Figure 51 is a block diagram of an electronic device for implementing the resource recommendation method of an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0122] like Figure 5 As shown, device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. RAM 503 may also store various programs and data required for the operation of device 500. Computing unit 501, ROM 502, and RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to bus 504.

[0123] Various components in device 500 are connected to I / O interface 505, including: an input unit 506, such as a keyboard, mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a magnetic disk, optical disk, etc.; and a communication unit 509, such as a network card, modem, wireless communication transceiver, etc. The communication unit 509 allows device 500 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0124] The computing unit 501 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as the resource recommendation method. For example, in some embodiments, the resource recommendation method may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed onto the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the resource recommendation method described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to perform the resource recommendation method in any other suitable manner (e.g., via firmware).

[0125] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-a-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0126] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0127] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0128] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0129] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0130] Computer systems may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The client and server relationship arises through computer programs running on the respective computers and having a client-server relationship to each other.

[0131] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0132] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A resource recommendation method, comprising: determining a multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource; determining an interaction feature sequence based on historical interaction data generated by the second object with respect to the resource within at least one time period; Determining a popularity evaluation value for the resource based on the basic features of the resource, the multimodal features, and the interaction feature sequence; as well as Based on the popularity evaluation value, resource recommendations are made.

2. The method according to claim 1, wherein The multimodal feature includes a script sentiment tendency sub-feature; and determining the multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource includes: Dividing the script text into a plurality of text sets according to the scenes of the script text; and The script emotion tendency sub-features are determined according to the emotion categories of the multiple text sets.

3. The method according to claim 1, wherein The multimodal feature includes a conflict density sub-feature; and determining the multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource includes: Segmenting the script text into a plurality of segmented words; Determining keywords representing conflicting content from the multiple word segments; and The conflict density sub-feature is determined according to the number of the keywords and the number of the multiple word segments.

4. The method according to claim 1, wherein The multimodal feature includes a sub-feature of interlude communication; and determining the multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource includes: For the dialogue text in the script text, determining at least one similar dialogue in a dialogue library that satisfies a similarity relationship with the dialogue text; Determining the number of resources including the at least one similar line according to the mapping relationship between the similar lines and the resources; and Determine the characteristics of the interlude communication based on the number of resources.

5. The method according to claim 1, wherein The multimodal feature includes a sub-feature of the spreadability of lines in a play; and determining the multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource includes: For a plurality of dialogue texts in the script text, determining a repetition rate between the plurality of dialogue texts according to similarities between the plurality of dialogue texts; and The propagation characteristics of the lines in the play are determined according to the repetition rate between the multiple line texts.

6. The method according to claim 1, wherein The multimodal feature includes a picture rhythm sub-feature; and determining the multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource includes: For the video in the resource, extract multiple images from the video to obtain an image sequence; If it is determined that the similarity between adjacent images in the image sequence is less than or equal to an image similarity threshold, determining that a shot cut occurs; and The picture rhythm sub-feature is determined according to the number of shot switching times.

7. The method according to claim 6, wherein: The step of extracting a plurality of images from the video in the resource to obtain an image sequence includes: For the video in the resource, the multiple images are extracted from a video segment between a start time and a predetermined time of the video to obtain the image sequence.

8. The method according to claim 1, wherein The multimodal feature includes an associated sub-feature; and determining the multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource includes: Obtaining a graph, the graph including a plurality of nodes, the plurality of nodes including a subject matter node and a cast and crew node; Determining a target theme node and a target cast and crew node from the plurality of nodes according to the theme of the resource and the first object; Determining associated nodes from the graph according to the target theme node and the target cast and crew node; and The associated sub-feature is determined according to the associated node.

9. The method according to claim 1, wherein The multimodal feature includes an influence sub-feature; and determining the multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource includes: The influence sub-feature representing the influence of the first object is determined according to at least one of the number of works and the number of fans of the first object.

10. The method according to claim 1, wherein The multimodal feature includes a lifecycle sub-feature; and determining the multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource includes: The life cycle sub-feature is determined based on the relationship between the playback volume and time of the reference resource.

11. The method according to any one of claims 1 to 10, wherein: Determining the popularity evaluation value for the resource according to the basic feature of the resource, the multimodal feature, and the interaction feature sequence includes: Determining feature weights of the basic features, the multimodal features, and the interaction feature sequence based on historical interaction data within a predetermined time period; and The popularity evaluation value is determined according to the basic features of the resource, the multimodal features, the interactive feature sequence and the feature weight.

12. The method according to claim 1, wherein The basic feature is determined based on at least one of a basic attribute and a posteriori attribute of the resource; The basic attributes include at least one of the following: resource name, resource category, resource identifier, and identifier of a video in the resource; The a posteriori attributes include at least one of the following: fast scrolling rate, satisfaction rate, and continued playback rate; wherein the fast scrolling rate is determined based on the number of times the duration of a single playback of the resource is less than a predetermined duration and the total number of playbacks of the resource; the satisfaction rate is determined based on the number of satisfactory interactive behaviors generated between the second object and the resource and the total number of playbacks; the continued playback rate is determined based on the number of continued playbacks and the total number of playbacks, and the continued playback characterizes that the second object continues to play other videos in the resource after playing one video in the resource.

13. The method according to claim 1, wherein The historical interaction data includes at least one of the following: distribution volume, comment volume, like volume, attention volume, forwarding volume, and average playback time per person.

14. The method according to claim 1, wherein The recommending of resources according to the popularity evaluation value includes: Adjusting the initial evaluation value of the resource according to the heat evaluation value to obtain a target evaluation value; and Resource recommendations are made based on the target evaluation values ​​of each of the multiple resources.

15. A resource recommendation device, comprising: a multimodal feature determination module, configured to determine a multimodal feature for the resource based on at least one of an image in the resource, a script text for the resource, a first object associated with the resource, and a resource attribute of a reference resource of the same category as the resource; a sequence determination module, configured to determine an interaction feature sequence based on historical interaction data generated by the second object with respect to the resource within at least one time period; a popularity evaluation value determination module, configured to determine a popularity evaluation value for the resource based on the basic features of the resource, the multimodal features, and the interaction feature sequence; as well as The recommendation module is used to recommend resources based on the heat evaluation value.

16. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 14.

17. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 14.

18. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 14.

Images