Segment recommendation method, segment recommendation device, electronic device and readable storage medium

Abstract

The embodiment of the application provides a segment recommendation method, a segment recommendation device, an electronic device and a readable storage medium, and belongs to the technical field of electronic devices. The segment recommendation method comprises the following steps: performing conversion processing on user information corresponding to an application program and a target multimedia file of the application program, and determining target features; inputting the target features into a recommendation model corresponding to the application program, wherein the recommendation model comprises multiple expert networks, a gating network and a fusion network; performing analysis processing on the target features based on the multiple expert networks to obtain multiple sample time information; determining weight information corresponding to the multiple expert networks based on the gating network; performing weighted operation on the multiple sample time information and the weight information based on the fusion network to obtain recommendation time information of the target multimedia file; and outputting a recommended segment of the target multimedia file according to the recommendation time information.

Description

Technical Field

[0001] The present application relates to the technical field of electronic devices, and in particular to a segment recommendation method, a segment recommendation device, an electronic device, and a readable storage medium. Background Art

[0002] In addition to providing a library of songs or videos for users to search, current media players typically recommend videos or audio to users to increase user engagement. However, when users listen to audio or watch videos, these players typically only recommend clips of the audio or video, which they then need to listen to or watch for a while. Because the parts that resonate with users vary from song to song or video to video, users are prone to skipping recommended clips if they initially don't like them. This hinders users' rapid exploration and judgment within the information flow, making it difficult for users to quickly find their favorite songs or videos. Summary of the Invention

[0003] The embodiments of the present application provide a segment recommendation method, a segment recommendation device, an electronic device, and a readable storage medium, which can accurately recommend segments for users to listen to or watch, so that users can quickly find their favorite multimedia files.

[0004] In a first aspect, an embodiment of the present application provides a method for recommending a segment, comprising:

[0005] Converting and processing the user information corresponding to the application and the target multimedia file of the application to determine the target features;

[0006] Input the target features into the recommendation model corresponding to the application, which includes multiple expert networks, gating networks, and fusion networks;

[0007] Based on multiple expert networks, target features are analyzed and processed to obtain multiple sample time information;

[0008] Based on the gating network, determine the weight information corresponding to multiple expert networks;

[0009] Based on the fusion network, a weighted operation is performed on multiple sample time information and weight information to obtain the recommended time information of the target multimedia file;

[0010] The recommended segment of the target multimedia file is output according to the recommended time information.

[0011] In a second aspect, an embodiment of the present application provides a segment recommendation device, comprising:

[0012] A first processing module is used to convert user information corresponding to the application and a target multimedia file of the application to determine target features;

[0013] The recommendation module is used to input the target features into the recommendation model corresponding to the application. The recommendation model includes multiple expert networks, gating networks, and fusion networks;

[0014] Based on multiple expert networks, target features are analyzed and processed to obtain multiple sample time information;

[0015] Based on the gating network, determine the weight information corresponding to multiple expert networks;

[0016] Based on the fusion network, a weighted operation is performed on multiple sample time information and weight information to obtain the recommended time information of the target multimedia file;

[0017] The output module is used to output the recommended segment of the target multimedia file according to the recommended time information.

[0018] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and running on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the fragment recommendation method provided in the first aspect.

[0019] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the fragment recommendation method provided in the first aspect are implemented.

[0020] In a fifth aspect, an embodiment of the present application provides a chip, which includes a processor and a communication interface, the communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the steps of the fragment recommendation method provided in the first aspect.

[0021] In an embodiment of the present application, user information corresponding to an application and the target multimedia file of the application are converted to determine target features. The target features are then input into a recommendation model corresponding to the application, which comprises multiple expert networks, a gating network, and a fusion network. Based on the multiple expert networks, the target features are analyzed and processed to obtain multiple sample time information. Based on the gating network, weight information corresponding to the multiple expert networks is determined. Based on the fusion network, a weighted calculation is performed on the multiple sample time information and the weight information to obtain recommended time information for the target multimedia file. Recommended segments of the target multimedia file are then output based on the recommended time information. This allows for the analysis of personalized recommendation time information related to the user information, namely, the start and end times of the recommended segments in the target multimedia file. Based on the recommended time information, segments of the target multimedia file are recommended to the user. Generating personalized recommended segments based on a deep learning algorithm reduces link logic and improves recommendation efficiency. Furthermore, the recommendation model is not limited by historical user behavior and labeled segment data. Even if the user's historical behavior and labeled segments are sparse, personalized recommended segments can be generated with higher accuracy and more consistent with user preferences, thus reducing the time required for users to explore multimedia files. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 One of the flowcharts of the segment recommendation method according to one embodiment of the present application is shown;

[0023] Figure 2 FIG2 shows a second flowchart of a method for recommending a segment according to an embodiment of the present application;

[0024] Figure 3 FIG3 shows a third flowchart of a method for recommending a segment according to an embodiment of the present application;

[0025] Figure 4 FIG4 shows a fourth flowchart of a method for recommending a segment according to an embodiment of the present application;

[0026] Figure 5 FIG5 shows a fifth flowchart of a method for recommending a segment according to an embodiment of the present application;

[0027] Figure 6 FIG6 shows a sixth flowchart of a method for recommending a segment according to an embodiment of the present application;

[0028] Figure 7 FIG7 shows a seventh flowchart of a method for recommending a segment according to an embodiment of the present application;

[0029] Figure 8 FIG8 shows an eighth flowchart of a method for recommending a segment according to an embodiment of the present application;

[0030] Figure 9 A schematic diagram of the structure of a recommendation model according to an embodiment of the present application is shown;

[0031] Figure 10 shows one of the timeline diagrams of multimedia files according to one embodiment of the present application;

[0032] Figure 11 FIG2 shows a second schematic diagram of a timeline of a multimedia file according to an embodiment of the present application;

[0033] Figure 12 FIG3 shows a third schematic diagram of a timeline of a multimedia file according to an embodiment of the present application;

[0034] Figure 13 shows a structural block diagram of a segment recommendation device according to an embodiment of the present application;

[0035] Figure 14 shows a structural block diagram of an electronic device according to an embodiment of the present application;

[0036] Figure 15 A hardware structure block diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0037] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are 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 are within the scope of protection of this application.

[0038] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0039] Refer to the following Figures 1 to 15 The present invention describes a segment recommendation method, a segment recommendation device, an electronic device, and a readable storage medium according to some embodiments of the present application.

[0040] In one embodiment of the present application, Figure 1A flowchart of a method for recommending a segment according to an embodiment of the present application is shown, including:

[0041] Step 102: converting the user information corresponding to the application and the target multimedia file of the application to determine the target characteristics;

[0042] The target multimedia file may be a multimedia file selected by the user in the application, or a multimedia file actively recommended to the user by the application. The target multimedia file includes audio and / or video, and of course, may also be other types of files. For example, music, song MVs, movies, or short videos. The user information corresponding to the application may be relevant information of the account used to log in to the application, such as the user name, user gender, user preference tags, etc. It may also be information representing the user of the electronic device, for example, using the hardware information of the electronic device as user information, such as a MAC address or an International Mobile Equipment Identity (IEMI) code.

[0043] In this embodiment, the user information corresponding to the application and the target multimedia file to be recommended to the user are first obtained and converted to embed the user information and the target multimedia file, forming vector content that can be recognized by the recommendation model, namely the target feature. This facilitates the recommendation model to output personalized recommendation segments for the user information.

[0044] Specifically, the target features include but are not limited to at least one of the following: user portrait features, user sequence features, user statistical features and user category features, as well as portrait features, embedding features, sequence features and statistical features of multimedia files.

[0045] Step 104: input the target features into the recommendation model corresponding to the application;

[0046] The recommendation model has a multi-task learning (Multi-gate Mixture-of-Experts, MMOE) model structure, which includes multiple expert networks, gate networks, and tower networks.

[0047] Step 106: Analyze and process the target features based on multiple expert networks to obtain multiple sample time information;

[0048] Step 108: determining weight information corresponding to the plurality of expert networks based on the gating network;

[0049] Step 110: performing a weighted operation on the multiple sample time information and weight information based on the fusion network to obtain the recommended time information of the target multimedia file;

[0050] The recommended time information includes the start time and the end time of the segment.

[0051] In this embodiment, each expert network is used to extract features of the target features input to the multi-task recommendation model from one dimension and analyze and calculate a sample moment information, where the sample moment information includes the sample start moment and the sample end moment. The gating network is used to output the weight information of multiple expert networks in different dimensions. The weight information of each expert network may be the same or different. The fusion network is used to use the weight of each expert network to fuse the sample moment information output by multiple expert networks in multiple different dimensions, that is, to perform weighted operations on multiple sample moment information and weight information to calculate the segment start moment and segment end moment.

[0052] For example, Figure 9 As shown, the recommendation model includes two gating networks, Gate-01 and Gate-02, three expert networks, Expert-01, Expert-02, and Expert-03, and two fusion networks, Tower-A and Tower-B. Gate-01 is responsible for assigning weight information to the start time, while Gate-02 is responsible for assigning weight information to the end time. Expert-01, Expert-02, and Expert-03 each output a sample time, meaning Tower-A and Tower-B each receive three sample time information. The gated network Gate-01 performs an inner product with the outputs of the three experts to obtain the input value for Tower-A. Similarly, the inner product of Gate-02 with the outputs of the three experts provides the input value for Tower-B. Tower-A predicts the start time of the recommended segment, while Tower-B predicts the end time of the recommended segment. Finally, the network outputs the start time (start-T) and end time (end-T) of the target multimedia file, which the user requested. Specifically, the gating network, expert network and fusion network are all deep neural network (DNN) structures.

[0053] Basic user features (input) are generated from user information, along with simple statistical categorical features. Complex statistical features and sequence features are generated from the user's historical listening behavior. Multimedia features and statistical features are obtained from the target multimedia file, and embedding features are generated using graph networks or the EGES (Enhanced Graph Embedding with Side Information) algorithm. Finally, these features related to the user information and the target multimedia file are concatenated to obtain the target features. The target features are fed into two gated networks and three expert networks. The two gated networks each output three weighted values, which are then inner-producted with the outputs of the three expert networks (sample time information) to obtain the input values ​​for Tower-A and Tower-B. Finally, forward inference is performed in Tower-A and Tower-B, respectively. Tower-A outputs 51s and Tower-B outputs 86s, indicating that user P is predicted to like the segment {51s, 86s} of the predicted song, i.e., the segment between the 51st and 86th seconds of the predicted song.

[0054] Step 112: output the recommended segment of the target multimedia file according to the recommended time information.

[0055] The recommended segment is the content between the segment start time and the segment end time in the target multimedia file.

[0056] In this embodiment, when the application needs to recommend a segment to the user, the user information and the target features of the target multimedia file that serves as the parent file of the recommended segment are input into a pre-trained recommendation model. The recommendation model is based on deep learning technology to analyze personalized recommendation moment information related to the user information, that is, the segment start moment and segment end moment of the recommended segment in the target multimedia file. The segment of the target multimedia file is recommended to the user according to the recommendation moment information. On the one hand, generating personalized recommended segments based on deep learning algorithms reduces link logic and helps improve recommendation efficiency. On the other hand, the recommendation model is not limited to the user's historical behavior and marked segment data. Even if the user's historical behavior and marked segments are sparse, personalized recommended segments can be generated with higher accuracy and make the recommended segments more in line with user preferences, which helps reduce the time cost required for users to explore multimedia files.

[0057] In one embodiment of the present application, Figure 2 The second flowchart of the method for recommending a segment according to an embodiment of the present application is shown, including:

[0058] Step 202: Obtain first subtitle information of the target multimedia file;

[0059] The first subtitle information includes multiple first subtitles, and the subtitle start time and subtitle end time of each first subtitle.

[0060] Step 204: if the segment start time is between the subtitle start time and the subtitle end time of at least one first subtitle, update the segment start time based on the subtitle start time or the subtitle end time of the at least one first subtitle;

[0061] Among them, updating the segment start time includes updating the segment start time to the subtitle start time of the first subtitle to supplement the incomplete first subtitle located at the segment start time, or updating the segment start time to the subtitle end time of the first subtitle to remove the incomplete first subtitle located at the segment start time.

[0062] Step 206: if the segment end time is between the subtitle start time and the subtitle end time of at least one first subtitle, update the segment end time based on the subtitle start time or the subtitle end time of the at least one first subtitle;

[0063] Among them, updating the segment end time includes updating the segment end time to the subtitle start time of the first subtitle to remove the incomplete first subtitle located at the segment end time, or updating the segment end time to the subtitle end time of the first subtitle to supplement the incomplete first subtitle located at the segment end time.

[0064] Step 208: Output the recommended segment of the target multimedia file according to the updated segment start time and segment end time.

[0065] In this embodiment, considering that for different target multimedia files, the segment start and end times output by the recommendation model may not represent complete nodes of a sentence in the target multimedia file, meaning that the recommended segment is not a complete sentence. This can make the recommended segment sound abrupt and detrimental to the user experience. Therefore, before outputting the recommended segment, the first subtitle information of the target multimedia file is obtained. If the segment start time falls between the subtitle start and end times of at least one first subtitle, indicating that the current segment start time is in the middle of a sentence, resulting in incomplete subtitles for the recommended segment, the subtitle start or end time of the first subtitle is used to supplement or delete the segment start time, so that the segment start time can be adaptively aligned with the subtitles. This adaptively smoothes the recommended segment generated by the model, making the recommended segment sound less abrupt and more natural, thereby enhancing the fluency of the segment output. Similarly, if the segment end time falls between the subtitle start and end times of at least one first subtitle, the subtitle start or end time of the first subtitle is used to supplement or delete the segment end time.

[0066] For example, Figure 10 As shown, taking the target multimedia file as a song as an example, the recommended segment {T1, T2} obtained by the segment start time T1 and the segment end time T2 is not a complete segment, that is, the start or end stage of the recommended segment is not the start or end of a line of lyrics. At this time, the segment start time T1 and the segment end time T2 of the recommended segment are aligned with the lyrics. Specifically, first record the start and end time of each line of lyrics of a song (time in seconds), and obtain a list of the start time of the lyrics of a song C = {t1,…,tm,…} and the end time list of the lyrics

[0067] D = {t1, ..., tn, ...}. The song segment starting time T1 is searched for in the lyrics starting time list C.

[0068] The maximum value smaller than T1, the segment end time T2 is found in the lyrics end time list D that is greater than T2

[0069] and updates the start and end time {tm, tn} of the song segment.

[0070] It is worth mentioning that, considering that for multimedia files with multiple voice tracks, such as chorus songs, there may be multiple first subtitles corresponding to the same time, in order to ensure that the start and end times of the segment can be aligned with the first subtitles, in the scenario where subtitles need to be supplemented, the minimum value among the subtitle start times of multiple first subtitles is determined as the updated segment start and end times; in the scenario where subtitles need to be removed, the maximum value among the subtitle end times of multiple first subtitles is determined as the updated segment start time. Similarly, in the scenario where subtitles need to be supplemented, the maximum value among the subtitle end times of multiple first subtitles is determined as the updated segment end time; in the scenario where subtitles need to be removed, the minimum value among the subtitle end times of multiple first subtitles is determined as the updated segment end time. This ensures that the updated segment start and end times can simultaneously meet the subtitle alignment requirements of multiple first subtitles, so that the segment start and end times of the recommended segment can be in the part without voice, thereby improving the user's audio-visual experience.

[0071] It's understandable that when subtitles exist at the corresponding position of the generated recommended segment, adaptive subtitle alignment technology is used to output the complete recommended segment. Otherwise, no subtitle alignment is required. For example, if the start and end time of a pure music segment is exactly between two subtitles, no subtitle alignment is required.

[0072] In one embodiment of the present application, Figure 3 The third flowchart of the method for recommending a segment according to an embodiment of the present application is shown, including:

[0073] Step 302: If the segment start time is within a preset time range of the target multimedia file, adjust the segment start time according to a first preset offset to form an output start time;

[0074] Step 304: If the segment end time is within the preset time range of the target multimedia file, adjust the segment end time according to a second preset offset to form an output end time;

[0075] The preset time range is used to indicate a portion of the target multimedia file that may contain a climax, loud playback volume, or subtitles. The preset time range can be pre-set by the user or automatically identified by a database. The first and second preset offsets can be appropriately set based on user needs or the threshold of the preset time range. For example, the first preset offset can be the difference between the start time of the segment and the threshold of the preset time range, or the user can set the second preset offset to 2 seconds, 4 seconds, 7 seconds, or 10 seconds.

[0076] Step 306: intercepting the target multimedia file according to the output start time and the output end time to form a recommended segment;

[0077] Step 308: Output the recommended segment.

[0078] In this embodiment, when the user instructs to output a recommended segment, if the recommended segment is a climax or lyrics portion of the target multimedia file, it may sound abrupt to the user, or even startle them. To prevent this, before outputting the recommended segment, if it is detected that the segment start time is within the preset time range of the target multimedia file, indicating that playing the recommended segment may cause an uncomfortable audiovisual experience, the segment start time is adjusted according to a first preset offset to form the output start time. Similarly, if it is detected that the segment end time is within the preset time range of the target multimedia file, the segment end time is adjusted according to a second preset offset to form the output end time. The content between the output start time and the output end time in the target multimedia file is then intercepted to obtain the recommended segment, and the recommended segment is recommended to the user. Thus, by extending the recommended segment by a certain offset according to the original corresponding positions at both ends and supplementing or shortening the accompaniment, the output recommended segment will sound less abrupt and more natural, thereby improving the user experience.

[0079] It is understandable that when the segment start time or segment end time is not within the preset time range of the target multimedia file, the segment start time or segment end time output by the recommendation model can be directly used as the basis for intercepting the recommended segment.

[0080] Specifically, outputting includes playing or displaying a play control, that is, outputting the recommended segment of the target multimedia file according to the recommended time information includes the following methods:

[0081] Method 1: Playing recommended clips. If the target multimedia file is a video, the recommended video clips are played along with the audio. For example, if the user first selects the target multimedia file to audition, the recommended clips are played directly.

[0082] Method 2: Display a play control for playing the recommended segment. In this case, the user can control whether the electronic device plays the recommended segment by inputting an operation on the play control.

[0083] Furthermore, when the segment start time is reduced according to the first preset offset to form the output start time or the segment end time is increased according to the second preset offset, it means that the system has instructed to extend the output start and end times by the preset offset. In order to further reduce the abruptness of listening to the recommended segment, after extending the output start and end times, the first segment in the target multimedia file located between the output start time and the segment start time is determined, that is, the portion that is additionally extended based on the segment start time. In the case where there is a vocal signal in the first segment, it means that the extended portion may also cause a sense of abruptness. Then, the vocals of the first segment are eliminated, and the first segment is converted into a pure accompaniment melody, thereby ensuring that the user will not be startled when listening to the recommended segment, thereby improving the user experience. Similarly, the vocals of the second segment in the target multimedia file located between the segment end time and the output end time can also be eliminated.

[0084] For example, Figure 12 As shown, taking songs as an example, assuming that the original song recommendation segment {T1, T2}, if there is a non-vocal accompaniment within 5 seconds before the T1 moment (the first preset offset), then the T1 moment is directly advanced to the T1' moment. If there is a vocal part (lyrics part), the vocal part is filtered, and then the T1 moment is advanced to the T1' moment. Similarly, for the T2 moment, if there is a non-vocal accompaniment within the last 5 seconds (the second preset offset), the T2 moment is extended to the T2' moment. If there is a vocal part, it is filtered and then extended. After the song is played, the song may be used as a historical multimedia file required by the recommendation model. In order to ensure the accuracy of the model, the extended part of the accompaniment is not included in the sample.

[0085] It should be noted that the human voice elimination may adopt conventional signal filtering technology or other existing channel elimination technology, and this application does not make any specific limitation.

[0086] In one embodiment of the present application, Figure 4 The second flowchart of the method for recommending a segment according to an embodiment of the present application is shown, including:

[0087] At step 402, a history output list corresponding to the user information is obtained.

[0088] The history output list includes at least one historical multimedia file corresponding to the user information and a user behavior log of each historical multimedia file. The user behavior log records the operation behavior of the user on the historical multimedia file. For example, for a song, the user played the song at 12:23.02, paused the playing at 12:24.00, continued to play the song at 15:45.12, downloaded the song at 15:46.15, and the playing of the song ended at 15:47.36.

[0089] At step 404, when it is detected that the first target behavior exists in the user behavior log, a sample segment of the historical multimedia file corresponding to the user behavior log is determined.

[0090] The first target behavior includes, but is not limited to, at least one of the following: a like behavior, a collection behavior, a sharing behavior, a comment behavior, and a download behavior, that is, a positive behavior that can represent that the user likes the multimedia file. The sample segment includes a first trigger time of the first target behavior. The like behavior means that the user has triggered a preset like button to like the multimedia file. In a music application, the like button is usually displayed as a red heart button. The collection behavior means that the user has collected the multimedia file to a collection folder or a self-built list. The sharing behavior means that the user has shared the multimedia file to a social media or directly sent to, for example, an instant messaging platform friend, an email contact, and the like. The download behavior means that the user has downloaded the multimedia file from a network terminal to a local behavior.

[0091] At step 406, the preset model is trained according to the sample segment of the at least one historical multimedia file and the user information, to form a recommendation model.

[0092] In this embodiment, if the first target behavior exists in the user behavior log, it means that the user shows the like for the historical multimedia file through the behavior when viewing or playing the historical multimedia file. At this time, the historical multimedia file can be used as a sample file, and a sample segment in which the first target behavior exists in the historical multimedia file is intercepted to represent the user's favorite segment through the sample segment. Then, the sample segment and the user information are used as model training data to train the preset model, to obtain a personalized segment recommendation model for the user information. Thus, the segment features of the sample file and the user sample features are used to train the model in an automatic learning feature manner, the network parameters in the model are fitted in a manner of reducing the prediction error, the big data analysis problem is solved, the preferred target multimedia file segment of different levels of users is predicted, the accuracy of the recommended segment prediction is improved, and the user experience is enhanced.

[0093] In some possible designs, the preset model adopts a multi-gate mixture-of-experts (MMOE) model, which processes massive user data with high scalability and high processing efficiency.

[0094] In one embodiment of the present application, Figure 5 The fifth flowchart of the method for recommending a segment according to an embodiment of the present application is shown, including:

[0095] Step 502: Determine a second triggering time of a second target behavior and a third triggering time of a third target behavior in the user behavior log;

[0096] The second target action is used to trigger the output of the historical multimedia file corresponding to the user behavior log, and the third target action is used to trigger the stop of the output of the historical multimedia file corresponding to the user behavior log. Stopping the output of the historical multimedia file can be pausing the output, ending the output, or switching to other multimedia files.

[0097] Step 504 : intercept and process the historical multimedia files corresponding to the user behavior log according to the second triggering moment and the third triggering moment to form sample segments.

[0098] In this embodiment, after selecting sample files containing the first target behavior, sample segments are extracted based on the adjacent second triggering time for outputting historical multimedia files and the third triggering time for stopping outputting historical multimedia files. This allows the sample segments to represent segments that the user may prefer, allowing the sample segments and user information to be used as model training data to train a pre-set model, resulting in a personalized segment recommendation model tailored to the user's information.

[0099] For example, taking the historical multimedia file as music, the timeline of the user behavior log of the music is as follows: Figure 11 As shown, it is determined through the user behavior log that the music has active positive user behavior (collection, comment, like, download or share). Figure 11 In the example, T3 represents the first time the user instructs to play the song, T4 represents the first time the user instructs to stop playing the song, T5 represents the second time the user instructs to play the song, T6 represents the first time the user takes a positive action (the first triggering moment), and T7 represents the second time the user instructs to stop playing the song. Therefore, between T5 and T7, the user adds the song to their favorites at T6. Therefore, the segment {T5, T7} is set as the sample segment of the song that the user likes.

[0100] In some possible designs, the sample segments extracted based on the second and third triggering moments of user behavior may not be complete nodes of a sentence in the multimedia file. When using a recommendation model trained on incomplete sample segments for segment recommendation, the recommended segments may not be complete sentences. This can make the recommended segments sound abrupt and detrimental to the user experience. Therefore, similar to the subtitle alignment method for recommended segments described above, before extracting and processing the historical multimedia files corresponding to the user behavior log based on the second and third triggering moments, the second subtitle information for the historical multimedia files corresponding to the user behavior log is first obtained. The second subtitle information includes multiple second subtitles and the subtitle start and end times of each second subtitle. If the second triggering moment falls between the subtitle start and end times of any second subtitle, indicating that the current second triggering moment is in the middle of a sentence, resulting in an incomplete sample segment, the second triggering moment is updated to the subtitle start time of any second subtitle. Similarly, if the third triggering moment falls between the subtitle start and end times of any second subtitle, the third triggering moment is updated to the subtitle end time of any second subtitle. This allows the start and end times of the sample segments to be aligned with the subtitles.

[0101] In one embodiment of the present application, after determining the sample segments of the historical multimedia files, the sample segments are screened according to preset conditions and a preset number of samples to eliminate samples of low quality or that cannot reflect the user's preferred segments.

[0102] In this embodiment, when multiple sample segments are identified in a historical multimedia file, a preset number of sample segments are selected from the multiple sample segments according to preset conditions to avoid confusion in model training. This provides high-quality sample segments for training the recommendation model, thereby ensuring the analytical accuracy of the trained recommendation model.

[0103] Among them, the preset conditions include at least one of the following: the number of times the first target behavior exists in the sample segment, the number of times the sample segment is played, and the duration of the sample segment. For example, the segment with the most first target behavior is determined to be the user's favorite segment of the song. For another example, if the duration of the sample segment of any historical multimedia file exceeds the preset duration range, the sample segment of any historical multimedia file is deleted, that is, only the sample segments with a duration within the preset duration range are retained as the data required for model training, thereby preventing the problem that too long sample segments are difficult to reflect the user's preference for part of the file content, and too short sample segments are insufficient in data volume to serve as the basis for model training, providing high-quality sample segments for training the recommendation model, thereby ensuring the analysis accuracy of the trained recommendation model.

[0104] It's worth mentioning that, considering that the second and third trigger moments of a fully output multimedia file are the starting and ending points of the multimedia file, respectively, even if the fully output multimedia file contains the first target behavior, the sample segments obtained through analysis and identification will inevitably be too long. Therefore, after obtaining historical multimedia files, the historical multimedia files that record the complete output can be directly deleted from the user behavior log, thereby reducing the number of sample files used to analyze the sample segments, which helps reduce the processor's computing pressure and improve data analysis efficiency.

[0105] In one embodiment of the present application, Figure 6 Flowchart 6 of the method for recommending a segment according to an embodiment of the present application is shown, including:

[0106] Step 602, determining characteristic information of at least one historical multimedia file;

[0107] The characteristic information includes at least one of the following: style information (such as rock style or folk style, etc.), category information, author information, duration information, click rate information and region information.

[0108] Step 604: update the recommendation model based on the feature information;

[0109] Step 606: Convert each of the multiple target multimedia files into the user information to determine multiple target features.

[0110] Step 608: Input the multiple target features into the updated recommendation model to obtain multiple pieces of recommended time information and recommendation degrees of the multiple pieces of recommended time information.

[0111] The multiple recommended moments output by the model are arranged according to their recommendation level. For example, if a movie is input into the updated recommendation model, the recommended moments for three movie clips are obtained. These three recommended moments have a first order, which is arranged from highest to lowest recommendation level. When the playback controls for the three movie clips are displayed on the program interface of an electronic device, the three playback controls are displayed according to the first order.

[0112] In this embodiment, the recommendation model is retrained using the feature information of historical multimedia files as model training data. This allows the updated recommendation model to further recommend content preferred by users, combining the feature information of target multimedia files. The model also outputs multiple recommended moments and recommendation levels for each moment. A higher recommendation level indicates that the recommended segment corresponding to that moment is more likely to suit the user's preferences. Furthermore, the model ranks the multiple recommended moments according to their recommendation levels, enabling users to quickly find their preferred segments and reducing file exploration costs.

[0113] For example, three fusion networks (Towers) are set up in the MMOE model. The first Tower is responsible for the start time of the segment, the second Tower is responsible for the end time of the segment, and the third Tower is responsible for the ranking score ctr (click-through rate). When analyzed through the model, the ranking results and personalized recommended segments of each song can be directly generated.

[0114] In one embodiment of the present application, Figure 7 Flowchart 7 of the method for recommending a segment according to an embodiment of the present application is shown, including:

[0115] Step 702, receiving a first input to the application;

[0116] The first input includes, but is not limited to, click input, key input, fingerprint input, sliding input, and pressing input. Key input includes, but is not limited to, single-click input, double-click input, long-press input, and combination key input of the power button, volume button, and main menu button of the electronic device. Of course, the first input can also be other operations of the user on the electronic device. The embodiments of this application do not specifically limit the operation method, and any feasible method can be used.

[0117] Step 704: In response to the first input, display a plurality of recommended multimedia files corresponding to the user information;

[0118] Step 706: Use at least one recommended multimedia file as a target multimedia file.

[0119] In this embodiment, the user can trigger the recommendation function of the application through a first input. After the recommendation function of the application is activated, multiple recommended multimedia files that the system has analyzed and that the user may like are directly displayed, and all recommended multimedia files are directly used as target multimedia files, or the user can manually select the recommended multimedia files they like as target multimedia files. This eliminates the need for the user to select the target multimedia files they want to listen to or watch. The electronic device will automatically recommend to the user based on the user's active needs and preferences, allowing the user to listen to more multimedia files that may not have been played or multimedia files that the user may like, thereby improving the user stickiness of the application.

[0120] For example, a music player is installed on a mobile phone. The music player interface is provided with a preset recommendation button. The user can actively click the recommendation button to enter the music recommendation interface. At the same time, the music player generates a recommendation request, and the mobile phone sends the recommendation request to the network terminal (database) via the network. As another example, the music player can also automatically generate a recommendation request when it is started, or automatically generate a recommendation request at a preset time point. The mobile phone can still send the recommendation request to the network terminal via the network.

[0121] It's understandable that because the updated recommendation model incorporates historical multimedia file style, click-through rates, and other information during training, it can also analyze the user's preferred multimedia files. After the application's recommendation function is activated, user information can be input into the updated recommendation model to derive target feature information. Based on this target feature information, the database is used to search for the recommended multimedia files to be recommended to the user.

[0122] In one embodiment of the present application, Figure 8 FIG8 is a flowchart of a method for recommending a segment according to an embodiment of the present application, including:

[0123] Step 802: If there are multiple target multimedia files, multiple recommended segments are spliced ​​together to generate a segment splicing file.

[0124] Step 804: Output the segment splicing file.

[0125] In this embodiment, if there are multiple target multimedia files, to facilitate the user playing all the recommended segments at once, the recommended segments corresponding to the multiple target multimedia files are combined and spliced ​​into a segment splicing file to form a multimedia file mix, and this segment splicing file is output. This allows the user to combine multimedia files that may be of interest, achieving a personalized multimedia file mix and enhancing the user experience.

[0126] In one embodiment of the present application, Figure 13 As shown, the segment recommendation device 1300 includes: a first processing module 1302, the first processing module 1302 is used to convert the user information corresponding to the application and the target multimedia file of the application to determine the target features; a recommendation module 1304, the recommendation module 1304 is used to input the target features into the recommendation model corresponding to the application, the recommendation model includes multiple expert networks, gating networks and fusion networks; based on the multiple expert networks, the target features are analyzed and processed to obtain multiple sample time information; based on the gating network, the weight information corresponding to the multiple expert networks is determined; based on the fusion network, the multiple sample time information and weight information are weighted and calculated to obtain the recommended time information of the target multimedia file; the output module 1306, the output module 1306 is used to output the recommended segment of the target multimedia file according to the recommended time information.

[0127] In this embodiment, when the application needs to recommend a segment to the user, the user information and the target features of the target multimedia file that serves as the parent file of the recommended segment are input into a pre-trained recommendation model. The recommendation model is based on deep learning technology to analyze personalized recommendation moment information related to the user information, that is, the segment start moment and segment end moment of the recommended segment in the target multimedia file. The segment of the target multimedia file is recommended to the user according to the recommendation moment information. On the one hand, generating personalized recommended segments based on deep learning algorithms reduces link logic and helps improve recommendation efficiency. On the other hand, the recommendation model is not limited to the user's historical behavior and marked segment data. Even if the user's historical behavior and marked segments are sparse, personalized recommended segments can be generated with higher accuracy and make the recommended segments more in line with user preferences, which helps reduce the time cost required for users to explore multimedia files.

[0128] Optionally, the segment recommendation apparatus 1300 further includes: a first acquisition module (not shown in the figure), configured to acquire first subtitle information of the target multimedia file, the first subtitle information including multiple first subtitles and a subtitle start time and a subtitle end time of each first subtitle; an update module (not shown in the figure), configured to update the segment start time based on the subtitle start time or subtitle end time of the at least one first subtitle when the segment start time is between the subtitle start time and subtitle end time of the at least one first subtitle; and to update the segment end time based on the subtitle start time or subtitle end time of the at least one first subtitle when the segment end time is between the subtitle start time and subtitle end time of the at least one first subtitle; and an output module 1306, further configured to output the recommended segment of the target multimedia file according to the updated segment start time and segment end time.

[0129] Optionally, the segment recommendation device 1300 further includes: an adjustment module (not shown in the figure), which is used to adjust the segment start time according to a first preset offset to form an output start time when the segment start time is within the preset time range of the target multimedia file; and adjust the segment end time according to a second preset offset to form an output end time when the segment end time is within the preset time range of the target multimedia file; a second processing module (not shown in the figure), which is used to intercept and process the target multimedia file according to the output start time and the output end time to form a recommended segment; and the output module 1306 is further used to output the recommended segment, and the output includes playing or displaying a play control.

[0130] Optionally, the segment recommendation apparatus 1300 further comprises a second acquisition module (not shown in the figure), configured to acquire a historical output list corresponding to the user information, the historical output list comprising at least one historical multimedia file and a user behavior log of each historical multimedia file; a determination module (not shown in the figure), configured to determine a sample segment of the historical multimedia file corresponding to the user behavior log in a case where it is detected that the first target behavior exists in the user behavior log, the sample segment comprising a first trigger time of the first target behavior; and a training module (not shown in the figure), configured to train the preset model according to the sample segment of the at least one historical multimedia file and the user information, to form a recommendation model; wherein the first target behavior comprises at least one of the following: a like behavior, a collection behavior, a sharing behavior, a comment behavior and a download behavior.

[0131] Optionally, the determination module is further configured to determine a second trigger time of a second target behavior and a third trigger time of a third target behavior in the user behavior log, the second target behavior being configured to trigger output of the historical multimedia file corresponding to the user behavior log, and the third target behavior being configured to trigger stop of the output of the historical multimedia file corresponding to the user behavior log; and the segment recommendation apparatus 1300 further comprises a third processing module (not shown in the figure), configured to perform intercept processing on the historical multimedia file corresponding to the user behavior log according to the second trigger time and the third trigger time, to form the sample segment.

[0132] In this embodiment, the modules of the segment recommendation apparatus 1300 implement the steps of the segment recommendation method in any of the embodiments of the first aspect when performing respective functions, and therefore the segment recommendation apparatus 1300 also comprises all the beneficial effects of the segment recommendation method in any of the embodiments of the first aspect, which will not be described here again.

[0133] The segment recommendation apparatus in the embodiments of the present application can be an apparatus, or a component, an integrated circuit or a chip in a terminal. The apparatus can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a notebook computer, a palm computer, a vehicle-mounted electronic device, a smart camera device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook or a personal digital assistant (PDA), etc., and the non-mobile electronic device can be a server, a network attached storage (NAS), a personal computer (PC), a television (TV) or a self-service machine, etc., and the embodiments of the present application are not limited in this regard.

[0134] The fragment recommendation apparatus in the embodiments of the present application can be an apparatus with an operating system. The operating system can be an Android operating system, can be an iOS operating system, and can also be other operating systems, and the embodiments of the present application do not make specific limitations.

[0135] In an embodiment of the present application, as shown in Figure 14 In an embodiment of the present application, as shown in

[0136] In an embodiment of the present application, an electronic device is provided, which includes the fragment recommendation apparatus provided in any of the above embodiments, and thus includes all the beneficial effects of the apparatus provided in any of the above embodiments, which will not be repeated here.

[0137] It should be noted that the electronic device in the embodiments of the present application includes the mobile electronic device and the non-mobile electronic device described above.

[0138] Figure 15 A hardware structure schematic diagram of an electronic device 1500 for implementing the embodiments of the present application. The electronic device 1500 includes but is not limited to: a radio frequency unit 1501, a network module 1502, an audio output unit 1503, an input unit 1504, a sensor 1505, a display unit 1506, a user input unit 1507, an interface unit 1508, a memory 1509, and a processor 1510, etc.

[0139] Those skilled in the art can understand that the electronic device 1500 can also include a power supply (such as a battery) for supplying power to each component, and the power supply can be logically connected to the processor 1510 through a power management system, so as to realize the functions of managing charging, discharging, and power consumption management, etc. through the power management system. Figure 15 The electronic device structure shown in the above embodiment does not constitute a limitation on the electronic device, and the electronic device can include more or fewer components than those shown, or combine certain components, or different component arrangements, which will not be repeated here.

[0140] Among them, the processor 1510 is used to convert and process the user information corresponding to the application and the target multimedia file of the application to determine the target features; input the target features into the recommendation model corresponding to the application, the recommendation model includes multiple expert networks, gating networks and fusion networks; based on multiple expert networks, the target features are analyzed and processed to obtain multiple sample moment information; based on the gating network, the weight information corresponding to the multiple expert networks is determined; based on the fusion network, the multiple sample moment information and weight information are weightedly calculated to obtain the recommended moment information of the target multimedia file; the audio output unit 1503 and / or the display unit 1506 are used to output the recommended segment of the target multimedia file according to the recommended moment information.

[0141] In this embodiment, when the application needs to recommend a segment to the user, the user information and the target features of the target multimedia file that serves as the parent file of the recommended segment are input into a pre-trained recommendation model. The recommendation model is based on deep learning technology to analyze personalized recommendation moment information related to the user information, that is, the segment start moment and segment end moment of the recommended segment in the target multimedia file. The segment of the target multimedia file is recommended to the user according to the recommendation moment information. On the one hand, generating personalized recommended segments based on deep learning algorithms reduces link logic and helps improve recommendation efficiency. On the other hand, the recommendation model is not limited to the user's historical behavior and marked segment data. Even if the user's historical behavior and marked segments are sparse, personalized recommended segments can be generated with higher accuracy and make the recommended segments more in line with user preferences, which helps reduce the time cost required for users to explore multimedia files.

[0142] Furthermore, the recommended time information includes a segment start time and a segment end time. The processor 1510 is also used to obtain first subtitle information of the target multimedia file, the first subtitle information including multiple first subtitles, and the subtitle start time and subtitle end time of each first subtitle; when the segment start time is between the subtitle start time and subtitle end time of at least one first subtitle, the segment start time is updated based on the subtitle start time or subtitle end time of at least one first subtitle; when the segment end time is between the subtitle start time and subtitle end time of at least one first subtitle, the segment end time is updated based on the subtitle start time or subtitle end time of at least one first subtitle; the audio output unit 1503 and / or the display unit 1506 are also used to output the recommended segment of the target multimedia file according to the updated segment start time and segment end time.

[0143] Furthermore, the processor 1510 is also used to adjust the starting time of the segment according to a first preset offset to form an output starting time when the starting time of the segment is within the preset time range of the target multimedia file; adjust the ending time of the segment according to a second preset offset to form an output ending time when the ending time of the segment is within the preset time range of the target multimedia file; intercept and process the target multimedia file according to the output starting time and the output ending time to form a recommended segment; the audio output unit 1503 and / or the display unit 1506 are also used to output the recommended segment, and the output includes playing or displaying a play control.

[0144] Furthermore, the processor 1510 is also used to obtain a historical output list corresponding to the user information, the historical output list includes at least one historical multimedia file and a user behavior log of each historical multimedia file; when a first target behavior is detected in the user behavior log, a sample segment of the historical multimedia file corresponding to the user behavior log is determined, the sample segment includes a first triggering moment of the first target behavior; based on the sample segment of at least one historical multimedia file and the user information, a preset model is trained to form a recommendation model; wherein the first target behavior includes at least one of the following: like behavior, collection behavior, sharing behavior, comment behavior and download behavior.

[0145] Furthermore, the processor 1510 is also used to determine the second trigger moment of the second target behavior in the user behavior log, and the third trigger moment of the third target behavior, the second target behavior is used to trigger the output of the historical multimedia file corresponding to the user behavior log, and the third target behavior is used to trigger the stop of outputting the historical multimedia file corresponding to the user behavior log; the historical multimedia file corresponding to the user behavior log is intercepted and processed according to the second trigger moment and the third trigger moment to form a sample segment.

[0146] It should be understood that in an embodiment of the present application, the input unit 1504 may include a graphics processing unit (GPU) 1541 and a microphone 1542, and the graphics processor 1541 processes the image data of a static picture or video obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The display unit 1506 may include a display panel 1561, and the display panel 1561 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 1507 includes a touch panel 1571 and other input devices 1572. The touch panel 1571 is also called a touch screen. The touch panel 1571 may include two parts: a touch detection device and a touch controller. Other input devices 1572 may include but are not limited to a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, and an operating stick, which will not be repeated here. The memory 1509 can be used to store software programs and various data, including but not limited to applications and operating systems. The processor 1510 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understood that the modem processor may not be integrated into the processor 1510.

[0147] In one embodiment of the present application, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the segment recommendation method provided in any of the above embodiments are implemented.

[0148] In this embodiment, reading the storage medium can implement each process of the segment recommendation method provided in the embodiment of the present application and can achieve the same technical effect. To avoid repetition, it will not be described here.

[0149] The processor is the processor in the communication device in the above embodiment. The read storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0150] An embodiment of the present application also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above-mentioned fragment recommendation method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0151] It should be understood that the chip mentioned in the embodiments of the present application may also be referred to as a system-level chip, a system chip, a chip system, or a system-on-chip chip, etc.

[0152] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0153] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0154] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A fragment recommendation method, characterized in that: include: Converting user information corresponding to an application and a target multimedia file of the application to determine target features; Inputting the target features into a recommendation model corresponding to the application, wherein the recommendation model includes multiple expert networks, a gating network, and a fusion network; Analyzing and processing the target features based on the multiple expert networks to obtain multiple sample time information; Determining weight information corresponding to the plurality of expert networks based on the gating network; Based on the fusion network, performing a weighted operation on the multiple sample time information and the weight information to obtain the recommended time information of the target multimedia file; outputting a recommended segment of the target multimedia file according to the recommended time information; The sample time information includes a sample start time and a sample end time, and the recommended time information includes a segment start time and a segment end time.

2. The fragment recommendation method according to claim 1, characterized in that: Outputting the recommended segment of the target multimedia file according to the recommended time information includes: Acquire first subtitle information of the target multimedia file, where the first subtitle information includes a plurality of first subtitles, and a subtitle start time and a subtitle end time of each first subtitle; When the segment start time is between a subtitle start time and a subtitle end time of at least one first subtitle, updating the segment start time based on the subtitle start time or the subtitle end time of the at least one first subtitle; When the segment end time is between a subtitle start time and a subtitle end time of at least one first subtitle, updating the segment end time based on the subtitle start time or the subtitle end time of the at least one first subtitle; The recommended segment of the target multimedia file is output according to the updated segment start time and the segment end time.

3. The fragment recommendation method according to claim 1, characterized in that: Outputting the recommended segment of the target multimedia file according to the recommended time information includes: If the segment start time is within a preset time range of the target multimedia file, adjusting the segment start time according to a first preset offset to form an output start time; When the segment end time is within a preset time range of the target multimedia file, adjusting the segment end time according to a second preset offset to form an output end time; According to the output start time and the output end time, intercepting the target multimedia file to form the recommended segment; Outputting the recommended segment, wherein the outputting includes playing or displaying a play control.

4. The fragment recommendation method according to claim 1, characterized in that Also includes: Obtaining a history output list corresponding to the user information, the history output list including at least one history multimedia file and a user behavior log for each history multimedia file; In a case where a first target behavior is detected in the user behavior log, determining a sample segment of a historical multimedia file corresponding to the user behavior log, the sample segment including a first triggering moment of the first target behavior; Training a preset model based on the sample segment of the at least one historical multimedia file and the user information to form the recommendation model; The first target behavior includes at least one of the following: like behavior, collection behavior, sharing behavior, comment behavior and download behavior.

5. The fragment recommendation method according to claim 4, characterized in that: The determining of the sample segments of the historical multimedia files corresponding to the user behavior log includes: Determining a second triggering time of a second target behavior in the user behavior log, and a third triggering time of a third target behavior, wherein the second target behavior is used to trigger outputting a historical multimedia file corresponding to the user behavior log, and the third target behavior is used to trigger stopping outputting the historical multimedia file corresponding to the user behavior log; The historical multimedia file corresponding to the user behavior log is intercepted and processed according to the second triggering moment and the third triggering moment to form the sample segment.

6. A fragment recommendation device, characterized in that: include: A first processing module is used to convert user information corresponding to the application and a target multimedia file of the application to determine target features; A recommendation module, configured to input the target features into a recommendation model corresponding to the application, wherein the recommendation model includes a plurality of expert networks, a gated network, and a fusion network; Analyzing and processing the target features based on the multiple expert networks to obtain multiple sample time information; Determining weight information corresponding to the plurality of expert networks based on the gating network; Based on the fusion network, performing a weighted operation on the multiple sample time information and the weight information to obtain the recommended time information of the target multimedia file; An output module, configured to output the recommended segment of the target multimedia file according to the recommended time information; The sample time information includes a sample start time and a sample end time, and the recommended time information includes a segment start time and a segment end time.

7. The segment recommendation device according to claim 6, characterized in that: The segment recommendation device further includes: A first acquisition module is configured to acquire first subtitle information of the target multimedia file, wherein the first subtitle information includes a plurality of first subtitles and a subtitle start time and a subtitle end time of each first subtitle; an updating module, configured to update the segment start time based on the subtitle start time or the subtitle end time of the at least one first subtitle when the segment start time is between the subtitle start time and the subtitle end time of the at least one first subtitle; and When the segment end time is between a subtitle start time and a subtitle end time of at least one first subtitle, updating the segment end time based on the subtitle start time or the subtitle end time of the at least one first subtitle; The output module is further configured to output the recommended segment of the target multimedia file according to the updated segment start time and segment end time.

8. The segment recommendation device according to claim 6, characterized in that: Also includes: an adjusting module, configured to adjust the segment start time according to a first preset offset to form an output start time when the segment start time is within a preset time range of the target multimedia file; as well as When the segment end time is within a preset time range of the target multimedia file, adjusting the segment end time according to a second preset offset to form an output end time; a second processing module, configured to intercept the target multimedia file according to the output start time and the output end time to form the recommended segment; The output module is further configured to output the recommended segment, where the output includes playing or displaying a play control.

9. The segment recommendation device according to claim 6, characterized in that: Also includes: A second acquisition module is configured to acquire a history output list corresponding to the user information, wherein the history output list includes at least one history multimedia file and a user behavior log for each history multimedia file; a determination module, configured to, upon detecting the presence of a first target behavior in the user behavior log, determine a sample segment of a historical multimedia file corresponding to the user behavior log, the sample segment including a first triggering moment of the first target behavior; a training module, configured to train a preset model based on a sample segment of the at least one historical multimedia file and the user information to form the recommendation model; The first target behavior includes at least one of the following: like behavior, collection behavior, sharing behavior, comment behavior and download behavior.

10. The segment recommendation device according to claim 9, characterized in that: The determination module is further configured to determine a second triggering time of a second target behavior in the user behavior log, and a third triggering time of a third target behavior, wherein the second target behavior is configured to trigger output of a historical multimedia file corresponding to the user behavior log, and the third target behavior is configured to trigger stopping output of the historical multimedia file corresponding to the user behavior log; The segment recommendation device further includes: The third processing module is used to intercept and process the historical multimedia file corresponding to the user behavior log according to the second triggering moment and the third triggering moment to form the sample segment.

11. An electronic device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the segment recommendation method according to any one of claims 1 to 5.

12. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the segment recommendation method according to any one of claims 1 to 5 are implemented.

Images