Video recommendation method and apparatus

Abstract

The application discloses a video recommendation method. The method comprises the following steps: determining whether a user likes a target video according to behavior data of the user in the process of watching the target video; if the user likes the target video, searching for a video group having a logical correlation relationship with the target video from a video library; when a video recommendation instruction is received, acquiring a preference feature vector of the user, a description feature vector of each to-be-recommended video in the video library, and a logical weighting value of all first to-be-recommended videos except the target video in the video group; calculating a preference degree of the user to each to-be-recommended video according to the preference feature vector, the description feature vector of each to-be-recommended video, the logical weighting value, and a preset matching degree algorithm; and selecting to-be-recommended videos meeting a preset condition according to all the calculated preference degrees to form a recommendation list and recommend the recommendation list to the user. The application can improve user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of video technology, and in particular to a video recommendation method and device. BACKGROUND

[0002] Currently, the algorithms used for recommending videos mainly include content-based recommendation algorithm, collaborative-based recommendation algorithm, or association rule-based recommendation algorithm. These recommendation algorithms determine whether a video should be recommended by calculating the similarity between a user and a video or the similarity between videos.

[0003] However, the inventors have found that the existing recommendation algorithms have the following problems in video recommendation:

[0004] In addition to the similarity relationship, there is also a logical relationship between some videos. For example, a movie has two parts, the first part and the second part, and in addition to the similarity relationship, there is also a sequence relationship. The current recommendation algorithm can recommend both movies at the same time, but the recommended sequence is often random. SUMMARY

[0005] An embodiment of the present application aims to provide a video recommendation method and system, which can solve the problem that the recommended sequence of videos with a logical association relationship is often random when recommending videos in the prior art.

[0006] One aspect of an embodiment of the present application provides a video recommendation method, comprising:

[0007] determining whether the user likes the target video according to the behavior data of the user in the process of watching the target video;

[0008] if the user likes the target video, finding a video group having a logical association relationship with the target video from a video library;

[0009] when receiving a video recommendation instruction, obtaining a preference feature vector of the user, a description feature vector of each first recommended video in the video group except the target video, and a logical weighting value of each first recommended video in the video group except the target video;

[0010] calculating a preference degree of each second recommended video in the video library except the video group for the user according to the preference feature vector, the description feature vector of each second recommended video, and a preset matching degree algorithm, and calculating a preference degree of each first recommended video in the video group for the user according to the preference feature vector, the description feature vector of each first recommended video, the corresponding logical weighting value of each first recommended video, and the matching degree algorithm;

[0011] According to the calculated preference degree of all, the video group to be recommended meeting the preset condition is selected to form a recommendation list and is recommended to the user.

[0012] Optionally, the behavior data includes watching completeness data, interaction times data and other behavior data of the user to the target video, and the other behavior data includes at least one of comment data, like data and collection data, and whether the user likes the target video is determined according to the behavior data of the user in the process of watching the target video, including:

[0013] The watching completeness, the interaction times and the other behavior data of the user to the target video are extracted from the behavior data.

[0014] The love value of the user to the target video is calculated according to the watching completeness, the interaction times, the other behavior data and a preset love value algorithm, and whether the user likes the target video is determined according to the calculated love value and a preset threshold.

[0015] Optionally, the method further includes:

[0016] The content text of each video to be recommended in the video library is extracted.

[0017] The content text of each video to be recommended is subjected to semantic induction by using a preset semantic recognition algorithm to obtain content description information of each video to be recommended.

[0018] The videos having a logical association relationship are determined according to the content description information of each video to be recommended, and all the videos having a logical association relationship are combined to form a video group.

[0019] For each video group, all the third videos to be recommended contained in each video group are sorted according to the event development sequence, and each third video to be recommended is configured with a corresponding initial weighting value.

[0020] The corresponding logical weighting value of each third video to be recommended is calculated according to the initial weighting value of each third video to be recommended, the behavior data of the user to each third video to be recommended and a preset logical weighting value algorithm, and the calculated logical weighting value is saved in association with the corresponding third video to be recommended.

[0021] Optionally, the extraction of the content text of each video to be recommended in the video library includes:

[0022] The subtitle text of each video to be recommended in the video library is extracted by using a preset character recognition algorithm to obtain the first content text of each video to be recommended.

[0023] extracting the voice of each video to be recommended in the video library by using a preset voice recognition algorithm, to obtain a second content text of each video to be recommended;

[0024] merging and processing the first content text and the second content text of each video to be recommended in a time sequence to obtain a content text of each video to be recommended.

[0025] Optionally, the merging and processing the first content text and the second content text of each video to be recommended in a time sequence to obtain a content text of each video to be recommended comprises:

[0026] For the first content text and the second content text of each video to be recommended, it is determined whether the language texts corresponding to the first content text and the second content text of each video to be recommended at the same video progress are similar.

[0027] If similar, one of the similar language texts is discarded, and the first content text and the second content text after discarding the language text are cross-merged and processed in a time sequence to obtain a content text of each video to be recommended.

[0028] Optionally, the searching for a video group having a logical association relationship with the target video from the video library comprises:

[0029] It is sequentially determined whether there is a video group containing the target video in a plurality of video groups contained in the video library.

[0030] If there is a video group containing the target video, the video group containing the target video is taken as a video group having a logical association relationship with the target video.

[0031] Optionally, the method further comprises:

[0032] obtaining historical behavior data of the user, the historical behavior data comprising viewing completeness data, interaction frequency data and other historical behavior data of each historical video watched by the user in a historical time period, the other historical behavior data comprising at least one of comment data, like data and collection data, wherein each historical video has a corresponding description feature vector;

[0033] determining a preference feature vector of the user according to the description feature vector of each historical video, a preset weight vector, a viewing completeness extracted from the viewing completeness data corresponding to each historical video, an interaction frequency extracted from the interaction frequency data corresponding to each historical video, and the other historical behavior data corresponding to each historical video.

[0034] Optionally, the description feature vector is composed of a plurality of description feature components, the weight vector is composed of a plurality of weight components, and the determining of the preference feature vector of the user according to the description feature vector of each historical video, the preset weight vector, the viewing completeness extracted from the viewing completeness data corresponding to each historical video, the interaction times extracted from the interaction times data corresponding to each historical video, and the other historical behavior data corresponding to each historical video comprises:

[0035] obtaining the weight component corresponding to each description feature component from the weight vector;

[0036] calculating the preference feature component corresponding to each description feature component according to the viewing completeness, the interaction times, the other historical behavior data of each historical video corresponding to each description feature component, the weight component corresponding to each description feature component, and a preset preference feature algorithm, and composing the preference feature vector by using all the obtained preference feature components.

[0037] Optionally, the selecting of the to-be-recommended videos that meet the preset condition from all the calculated preference degrees to form a recommendation list for recommending to the user comprises:

[0038] sorting all the calculated preference degrees in descending order;

[0039] selecting the to-be-recommended videos whose preference degrees are within a preset position to form a recommendation list for recommending to the user.

[0040] An aspect of an embodiment of the present application further provides a video recommendation device, which comprises:

[0041] a determining module configured to determine whether the user likes the target video according to the behavior data of the user in the process of watching the target video;

[0042] a searching module configured to search for a video group having a logical association relationship with the target video from a video library if the user likes the target video;

[0043] an obtaining module configured to obtain a preference feature vector of the user, a description feature vector of each to-be-recommended video in the video library, and a logical weight of all first to-be-recommended videos in the video group except the target video when a video recommendation instruction is received;

[0044] a calculating module, configured to calculate a preference degree of each second video to be recommended in the video library other than the video group to the user according to the preference feature vector, a description feature vector of each second video to be recommended in the video library other than the video group, and a preset matching degree algorithm, and calculate a preference degree of each first video to be recommended in the video group to the user according to the preference feature vector, a description feature vector of each first video to be recommended in the video group, a corresponding logical weighting value of each first video to be recommended, and the matching degree algorithm;

[0045] a recommending module, configured to select a video to be recommended meeting a preset condition to form a recommendation list to recommend to the user according to all the calculated preference degrees.

[0046] The application further provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the method when executing the computer program.

[0047] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the steps of the method.

[0048] The video recommendation method provided by the application can determine whether the user likes the target video according to behavior data of the user in the process of watching the target video, and if the user likes the target video, find a video group having a logical association relationship with the target video from a video library, and when a video recommendation instruction is received, obtain a preference feature vector of the user, a description feature vector of each video to be recommended in the video library, and a logical weighting value of each first video to be recommended in the video group other than the target video, calculate a preference degree of each second video to be recommended in the video library other than the video group to the user according to the preference feature vector, a description feature vector of each second video to be recommended in the video library other than the video group, and a preset matching degree algorithm, and calculate a preference degree of each first video to be recommended in the video group to the user according to the preference feature vector, a description feature vector of each first video to be recommended in the video group, a corresponding logical weighting value of each first video to be recommended, and the matching degree algorithm, and select a video to be recommended meeting a preset condition to form a recommendation list to recommend to the user according to all the calculated preference degrees, so that when a video is recommended, not only whether the characteristics of the video meet the preference habit of the user is considered, but also the logical association relationship order of the videos is considered, so that the videos are recommended according to the logical association relationship order when the videos are recommended, the order of the recommended videos is avoided to be disorderly, and the user experience is improved. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 This is a schematic diagram of the environment for the video recommendation method according to an embodiment of this application;

[0050] Figure 2 A flowchart of one embodiment of the video recommendation method described in this application;

[0051] Figure 3 A detailed flowchart illustrating the steps for determining whether a user likes a target video based on behavioral data during the viewing process of the target video, as described in one embodiment of this application.

[0052] Figure 4 This is a detailed flowchart illustrating the steps of searching a video group in a video library that has a logical relationship with the target video in one embodiment of this application.

[0053] Figure 5 A flowchart illustrating another embodiment of the video recommendation method described in this application;

[0054] Figure 6 This is a detailed flowchart illustrating the steps of extracting the content text of each video to be recommended from the video library in one embodiment of this application.

[0055] Figure 7 This is a detailed flowchart illustrating the steps in one embodiment of the present application to cross-merge the first and second content texts of each video to be recommended in chronological order to obtain the content text of each video to be recommended.

[0056] Figure 8 A flowchart illustrating another embodiment of the video recommendation method described in this application;

[0057] Figure 9 This is a detailed flowchart illustrating the steps for determining the user based on the descriptive feature vectors of each historical video, a preset weight vector, the viewing completeness extracted from the viewing completeness data corresponding to each historical video, the number of interactions extracted from the interaction count data corresponding to each historical video, and other historical behavior data corresponding to each historical video in one embodiment of this application.

[0058] Figure 10 This is a detailed flowchart illustrating the steps of selecting videos that meet preset conditions based on all calculated preference levels and recommending them to the user.

[0059] Figure 11 This is a program block diagram of one embodiment of the video recommendation device described in this application;

[0060] Figure 12A schematic diagram of the hardware structure of a computer device for executing a video recommendation method provided in an embodiment of this application. Detailed Implementation

[0061] The advantages of this application are further illustrated below with reference to the accompanying drawings and specific embodiments.

[0062] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0063] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the listed items.

[0064] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0065] In the description of this application, it should be understood that the numerical labels before the steps do not indicate the order of the steps, but are only used to facilitate the description of this application and to distinguish each step, and therefore should not be construed as a limitation of this application.

[0066] Figure 1 This illustration schematically depicts an application environment diagram of the video recommendation method according to an embodiment of this application. In an exemplary embodiment, the system of this application environment may include a user terminal 10 and a server 20. The user terminal 10 and server 20 are connected wirelessly or via a wired connection, and the user terminal 10 has a corresponding application client or web client. The user terminal 10 can be a PC, mobile phone, iPad, tablet computer, laptop computer, personal digital assistant, etc. The server 20 can be a rack server, blade server, tower server, or cabinet server (including a standalone server or a server cluster composed of multiple servers), etc.

[0067] See Figure 2 This is a flowchart illustrating a video recommendation method according to an embodiment of this application. It should be understood that the flowchart in this method embodiment is not intended to limit the order of execution steps. This video recommendation method is applied to the server side. As can be seen from the figure, the video recommendation method provided in this embodiment includes:

[0068] Step S20: Determine whether the user likes the target video based on the user's behavioral data during the viewing process.

[0069] Specifically, the behavioral data includes the user's viewing completion rate data, interaction frequency data, and other behavioral data, including at least one of comment data, like data, and collection data.

[0070] The viewing completeness data refers to the data that records the viewing completeness of the target video watched by the user.

[0071] The interaction count data is used to record the number of interactions a user makes while watching the target video.

[0072] In this embodiment, after a user finishes watching the target video, the user's behavioral data during the viewing process can be uploaded to the server so that the server can obtain the behavioral data and determine whether the user likes the target video based on the obtained behavioral data.

[0073] It should be noted that the target video refers to the video currently being watched by the user. The target video can be a long video or a short video, and this embodiment does not impose any limitation.

[0074] In one exemplary embodiment, the behavioral data includes the user's viewing completion rate data, interaction frequency data, and other behavioral data. The other behavioral data includes at least one of comment data, like data, and collection data. See [reference needed]. Figure 3 Determining whether a user likes a target video based on the user's behavioral data during the viewing process includes:

[0075] Step S30: Extract the user's viewing completion rate, number of interactions, and other behavioral data of the target video from the behavioral data.

[0076] Specifically, the viewing completeness is the proportion of the time a user spends watching the target video to the total length of the target video. For example, if a user watches a 10-minute target video for 8 minutes, the viewing completeness is 8 / 10 = 80%.

[0077] The number of interactions refers to the number of times a user sends bullet comments while watching the target video.

[0078] Step S31: Calculate the user's liking value for the target video based on the viewing completion rate, the number of interactions, the other behavioral data, and a preset liking value algorithm, and determine whether the user likes the target video based on the calculated liking value and a preset threshold.

[0079] Specifically, the "like value" algorithm is a pre-defined formula used to calculate the user's "like" value for the target video. The "like value" is a numerical value used to measure the user's degree of liking for the target video; a higher "like value" indicates a greater liking for the target video, and a lower "like value" indicates a less liking for the target video. The "like value" algorithm is as follows:

[0080] f(x) = Viewing Completeness * Ψ0 + 1 / (1 + exp(-log(2, Interaction Count + 1))) - 0.5) * Ψ1 + (Comments? Ψ2:0) + (Likes? Ψ3:0) + (Favorites? Ψ4:0), where f(x) is the user's liking value for the target video, exp is an exponential function with the natural constant e as its base, and Ψ0, Ψ1, Ψ2, Ψ3, and Ψ4 are preset values. "Comments? 2:0" means that if the other behavioral data includes comment data, then "Ψ2" is added to the liking value algorithm; if the other behavioral data does not include comment data, then "0" is added to the liking value algorithm. Similarly, "Like? Ψ3:0" means that if the other behavioral data includes likes, then "Ψ3" is added to the like value algorithm; if the other behavioral data does not include likes, then "0" is added to the like value algorithm. Likewise, "Favorite? Ψ4:0" means that if the other behavioral data includes favorites, then "Ψ4" is added to the like value algorithm; if the other behavioral data does not include favorites, then "0" is added to the like value algorithm.

[0081] In this embodiment, after obtaining the viewing completeness, the number of interactions, and other behavioral data, these data are substituted into the like value algorithm to calculate the user's like value for the target video.

[0082] The preset threshold is a pre-set threshold used to determine whether a user likes the target video. The preset threshold can be set and adjusted according to the actual situation. For example, if the preset threshold is 2, it means that when the like value is greater than or equal to 2, it can be determined that the user likes the target video, and when the like value is less than 2, it can be determined that the user does not like the target video.

[0083] Step S21: If the user likes the target video, then search the video library for video groups that have a logical relationship with the target video.

[0084] Specifically, all videos stored in the video library are pre-divided into multiple video groups according to logical relationships. In this way, when it is detected that a user likes the target video, the video group that has a logical relationship with the target video can be found from the video library.

[0085] If two or more videos involve similar event names, the same main characters, and similar event content, then these videos can be considered to have a logical connection. All videos with a logical connection are grouped together as a video group.

[0086] The video library is a database used to store all videos.

[0087] In one exemplary implementation, see [reference] Figure 4 The step of searching for video groups in the video library that have a logical relationship with the target video includes: step S40, sequentially determining whether there is a video group containing the target video among the multiple video groups contained in the video library; step S41, if there is a video group containing the target video, then the video group containing the target video is regarded as the video group that has a logical relationship with the target video.

[0088] In one embodiment, the existence of a video group containing the target video can be determined by comparing the identifier information of the target video with the identifier information of the videos in each video group one by one. If a video group contains a video with the same identifier information as the target video, then that video group can be determined to be a video group with a logical relationship to the target video. The identifier information is used to distinguish different videos; for example, the identifier information is a video ID.

[0089] Step S22: Upon receiving a video recommendation instruction, obtain the user's preference feature vector, the description feature vector of each video to be recommended in the video library, and the logical weighted value of all first videos to be recommended in the video group except for the target video.

[0090] Specifically, the video recommendation instruction is an instruction used to recommend videos to the user. In this embodiment, when a user wants to obtain recommended videos, the user can trigger a recommended video retrieval request through a terminal device. After detecting the recommended video retrieval request triggered by the user, the terminal device will generate a video recommendation instruction based on the retrieved recommended video and send the video recommendation instruction to the server. Upon receiving the video recommendation instruction, the server can obtain the user's preference feature vector, the descriptive feature vector of each video to be recommended in the video library, and the logical weighted value of all first videos to be recommended in the video group except for the target video. Then, based on this information, the server determines the videos to be recommended and recommends them to the user.

[0091] The preference feature vector is a first-type vector describing the user's preferred video type (video characteristics), which can be determined by obtaining the user's preferred video type. For example, the user's preferences are as follows: [Comedy: 0.8, Thriller: 0.2, War: 1.0, Crime: 0.4, Historical Drama: 0.4], where the numbers are the preference type values ​​obtained by quantifying the user's preferences; the larger the number, the stronger the user's preference for that type. Mapped to the video type vector, the expression in the same vector space as the descriptive feature vector of the video type can be obtained: [0.8,0,0.2,0,1.0,0.4,0,0,0,0,0.4], where the user's preference corresponds to the 0th, 2nd, 4th, 5th, and 10th positions of the vector. The expression obtained above in the same vector space as the descriptive feature vector of the video type can be called the preference feature vector.

[0092] The descriptive feature vector is a second type vector used to describe the type of the video watched by the user. The descriptive feature vector can be determined by obtaining the type tags that the user has pre-assigned to the video. Assuming there are 11 video types in total, and the video watched by the user has the following type tags: comedy, suspense, thriller, then the corresponding descriptive feature vector for that video is: [1,0,0,1,0,1,0,0,0,0,0], where the type tag of the video corresponds to the 0th, 3rd, and 5th bits of the vector.

[0093] It is understandable that if there are a total of 11 video types, then the feature vector will have 11 dimensions. The total number of video types that all videos have can be set according to actual needs, and is not limited in this embodiment. Each video may have one type label or multiple type labels. For example, a video may simultaneously satisfy the characteristics of "thriller", "crime" and "suspense" type videos, then these three characteristics are all video labels for that video.

[0094] In this embodiment, when each video is created or generated, the video's author or content creator can add descriptive tags Ti of different dimensions to the video based on its various characteristics. Combining all tags Ti generates a descriptive feature vector: T = {T0, T1, T2, ..., Tn}, which comprehensively describes the characteristics of the video. Specifically, when a video possesses a corresponding characteristic, the descriptive tag Ti for that characteristic is represented by "1" in the descriptive feature vector; when a video does not possess a corresponding characteristic, the descriptive tag for that characteristic is represented by "0" in the descriptive feature vector.

[0095] The logical weighted value is a numerical value calculated by weighting the user's preference for each first recommended video in the video group. Different first recommended videos have different logical weighted values. The logical weighted value can be pre-calculated and associated with the corresponding first recommended video for storage, or it can be calculated according to preset parameters when needed. This embodiment does not impose any limitations on this.

[0096] In one exemplary implementation, see [reference] Figure 5 The method further includes:

[0097] Step S50: Extract the content text of each video to be recommended from the video library.

[0098] Specifically, the content text can be extracted from each video to be recommended in the video library using speech recognition algorithms and / or text recognition algorithms.

[0099] The content text is a detailed description of the video to be recommended.

[0100] In one exemplary implementation, see [reference] Figure 6 Extracting the content text of each video to be recommended from the video library includes:

[0101] Step S60: Use a preset text recognition algorithm to extract the subtitle text of each video to be recommended in the video library to obtain the first content text of each video to be recommended.

[0102] Specifically, the text recognition algorithm is an existing OCR recognition algorithm, which will not be described in detail in this embodiment.

[0103] The subtitle text refers to all the subtitles in the video to be recommended.

[0104] Step S61: Use a preset speech recognition algorithm to extract the speech of each video to be recommended in the video library to obtain the second content text of each video to be recommended.

[0105] Specifically, the speech recognition algorithm is an existing algorithm, such as an algorithm based on dynamic time warping, a method based on a parametric hidden Markov model (HMM), etc.

[0106] Step S62: Merge the first and second content texts of each video to be recommended in chronological order to obtain the content text of each video to be recommended.

[0107] Specifically, the time sequence refers to the video progress order corresponding to the sentences contained in the first content text and the second content text.

[0108] As an example, the first content text consists of 5 sentences, each corresponding to a video progress of 1 minute, 2 minutes, 3 minutes, 4 minutes, and 5 minutes. The second content text also consists of 5 sentences, each corresponding to a video progress of 1 minute 10 seconds, 2 minutes 10 seconds, 3 minutes 10 seconds, 4 minutes 10 seconds, and 5 minutes 10 seconds. After merging the content, the resulting content text contains sentences in the following order: 1 minute, 1 minute 10 seconds, 2 minutes, 2 minutes 10 seconds, 3 minutes, 3 minutes 10 seconds, 4 minutes, 4 minutes 10 seconds, 5 minutes, and 5 minutes 10 seconds.

[0109] In one exemplary implementation, see [reference] Figure 7 The step of cross-merging the first and second content texts of each video to be recommended in chronological order to obtain the content text of each video to be recommended includes: Step S70, for each video to be recommended, determining whether the language texts corresponding to the first and second content texts of each video to be recommended are similar under the same video progress; S71, if similar, discarding one of the similar language texts, and cross-merging the first and second content texts after discarding the language text in chronological order to obtain the content text of each video to be recommended.

[0110] Specifically, the language text refers to the sentences contained in the first content text or the second content text.

[0111] In this embodiment, the corresponding language texts of the first and second content texts at the same video progress can be determined using statistical or deep learning methods. Statistical methods can include the Levenstein distance algorithm, BM25 algorithm, TF-IDF algorithm, etc. Deep learning methods can include the cosine similarity algorithm based on Word2vec. When two language texts are detected to be similar, one of them can be discarded to avoid duplicate text in the final content text. After discarding all the calculated similar language texts, the remaining language texts can be cross-merged in the same way as the content text obtained in the example above to obtain the final content text K.

[0112] As an example, K = {K0, K1, K2, ..., Kn}, where K0, K1, K2, ..., Kn refer to the spoken text at a certain video progress.

[0113] Step S51: Use a preset semantic recognition algorithm to perform semantic summarization on the content text of each video to be recommended, and obtain the content description information of each video to be recommended.

[0114] Specifically, the semantic recognition algorithm can also be called a semantic understanding algorithm. The speech recognition algorithm in this embodiment adopts any mature semantic recognition algorithm in the prior art, and the specific algorithm will not be described in detail in this embodiment.

[0115] The content description information is used to describe the main content of the video to be recommended, and may include the event name, main characters, event content, and event progress of the video to be recommended.

[0116] Step S52: Determine the videos with logical relationships based on the content description information of each video to be recommended, and group all the videos with logical relationships into a video group.

[0117] Specifically, after obtaining the content description information of each video to be recommended, the logical relationship between the videos can be determined based on the event name, event content, and main characters contained in the content description information. In this embodiment, when multiple videos involve similar event names, the same main characters, and similar event content, these videos can be determined to be logically related, and these videos can be grouped into a video group.

[0118] It should be noted that videos with different logical relationships can be grouped into different video groups.

[0119] Step S53: For each video group, sort all the third videos to be recommended in each video group according to the order of event development, and configure a corresponding initial weighting value for each third video to be recommended.

[0120] Specifically, the sequence of events refers to the order in which the content recorded in the video occurs. For example, if a video group consists of 10 episodes of a variety show, then the videos in the video group need to be ordered according to the episode numbers.

[0121] As an example, the initial weighting value corresponding to each third video to be recommended is as follows:

[0122] Ψ = {Ψ0, Ψ1, Ψ2, ..., Ψn}, where Ψ0, Ψ1, Ψ2, ..., Ψn are the initial weighted values ​​of each video in the video group after being sorted according to the order of events. That is, the initial weighted value of the first video is Ψ0, the initial weighted value of the second video is Ψ2, and so on, with the initial weighted value of the nth video being Ψn, where Ψi < Ψ(i+1).

[0123] Step S54: Calculate the logical weighting value corresponding to each third video to be recommended based on the initial weighting value of each third video to be recommended, the user's behavior data for each third video to be recommended, and the preset logical weighting value algorithm, and associate and save the calculated logical weighting value with the corresponding third video to be recommended.

[0124] Specifically, the logical weighting algorithm is a pre-defined formula used to calculate the logical weighting value for each third video to be recommended. The logical weighting algorithm is as follows:

[0125] The current Ψi = Ψi * f(x), where the current Ψi is the logical weighted value corresponding to the third video to be recommended, Ψi is the initial weighted value corresponding to the third video to be recommended, and f(x) is the user's liking value for the third video to be recommended. The specific calculation method of f(x) has been described in detail above and will not be repeated in this embodiment.

[0126] In this embodiment, after calculating the logical weighted value for each third video to be recommended using the above formula, the calculated logical weighted value can be associated with and saved as the corresponding third video to be recommended. This way, the logical weighted value can be easily obtained when calculating the preference for the third video to be recommended later.

[0127] Step S23: Calculate the user's preference degree for each second recommended video based on the preference feature vector, the description feature vector of each second recommended video in the video library (excluding the video group), and a preset matching degree algorithm; and calculate the user's preference degree for each first recommended video based on the preference feature vector, the description feature vector of each first recommended video in the video group, the logical weighted value corresponding to each first recommended video, and the matching degree algorithm.

[0128] Specifically, the matching degree algorithm is a pre-defined formula used to calculate the user's preference for each video to be recommended. The matching degree algorithm is as follows:

[0129] Y = α * 1 / exp(d), where Y is the user's matching degree for each video to be recommended, α is a preset value, and d is the similarity value between the preference feature vector and the descriptive feature vector of each video to be recommended. It should be noted that the similarity value can be the cosine similarity between the preference feature vector and the descriptive feature vector of each video to be recommended, or it can be the Euclidean distance between the preference feature vector and the descriptive feature vector of each video to be recommended.

[0130] In this embodiment, the smaller d is, the greater the user preference, indicating that the user likes the video more; the larger d is, the smaller the user preference, indicating that the user dislikes the video less.

[0131] In this embodiment, after calculating the matching degree of the user for each of the two videos to be recommended using the above formula, the matching degree can be used as the user's preference degree for each of the two videos to be recommended. After calculating the matching degree of the user for each of the first videos to be recommended using the above formula, the calculated matching degree value needs to be multiplied by the corresponding logical weighting value, and the resulting value is used as the user's preference degree for each of the first videos to be recommended. That is, the user's preference degree for each of the two videos to be recommended is α*1 / exp(d), and the user's preference degree for each of the first videos to be recommended is α*1 / exp(d)*Ψi, where Ψi is the logical weighting value corresponding to the first video to be recommended.

[0132] In one exemplary implementation, see [reference] Figure 8 The method further includes:

[0133] Step S80: Obtain the user's historical behavior data. The historical behavior data includes the user's viewing completion data, interaction count data, and other historical behavior data for each historical video watched within a historical time period. The other historical behavior data includes at least one of comment data, like data, and collection data. Each historical video has a corresponding descriptive feature vector.

[0134] Specifically, the historical time period can be set and adjusted according to the actual situation. For example, the historical time period may be the past six months, the past three months, etc.

[0135] The viewing completion data, interaction frequency data, and other historical behavior data have been described in detail above and will not be repeated in this embodiment.

[0136] The historical videos are those watched by the user within a historical time period.

[0137] Step S81: Determine the user's preference feature vector based on the description feature vector of each historical video, the preset weight vector, the viewing completeness extracted from the viewing completeness data corresponding to each historical video, the number of interactions extracted from the interaction number data corresponding to each historical video, and other historical behavior data corresponding to each historical video.

[0138] Specifically, the weight vector is a pre-defined vector used to identify the weight component corresponding to each descriptive feature vector.

[0139] As an example, the weight vector V = {V0, V1, V2, ..., Vn}, where V0, V1, V2, ..., Vn are weight components, and each weight component corresponds to a descriptive feature component. For example, if the descriptive feature vector T = {T0, T1, T2, ..., Tn}, then weight component V0 corresponds to T0, weight component V1 corresponds to T1, weight component V2 corresponds to T2, and weight component Vn corresponds to Tn.

[0140] In this embodiment, machine learning algorithms are used to learn user behavior data such as viewing completion rate of various types of historical videos, number of bullet screen interactions, comments, likes, and favorites, thereby obtaining the user's preference feature vector.

[0141] In one exemplary embodiment, the descriptive feature vector is composed of multiple descriptive feature components. For example, the descriptive feature vector T = {T0, T1, T2, ..., Tn}, where T0, T1, T2, ..., Tn are descriptive feature components. The weight vector is composed of multiple weight components. For example, the weight vector V = {V0, V1, V2, ..., Vn}, where V0, V1, V2, ..., Vn are weight components. See also... Figure 9 The step of determining the user's preference feature vector based on the descriptive feature vector of each historical video, a preset weight vector, the viewing completeness extracted from the viewing completeness data corresponding to each historical video, the number of interactions extracted from the interaction number data corresponding to each historical video, and other historical behavior data corresponding to each historical video includes:

[0142] Step S90: Obtain the weight component corresponding to each descriptive feature component from the weight vector.

[0143] Specifically, assuming there are a total of 11 descriptive feature components, during the calculation process, it is necessary to sequentially obtain the weight components corresponding to the 1st, 2nd, 3rd, 4th, 5th, 6th, 7th, 8th, 9th, 10th, and 11th descriptive feature components.

[0144] Step S91: Calculate the preference feature component corresponding to each description feature component based on the viewing completeness, interaction count, other historical behavior data, weight component corresponding to each description feature component, and preset preference feature algorithm for each historical video corresponding to each description feature component, and form the preference feature vector by combining all the obtained preference feature components.

[0145] Specifically, the preference feature algorithm is a pre-defined formula used to calculate the preference feature components.

[0146] The preference feature algorithm is as follows:

[0147] Pi = Pi 历史 +Vi*f(x);

[0148] f(x) = Viewing completeness * Ψ0 + 1 / (1 + exp(-log(2, number of interactions + 1))) - 0.5) * Ψ1 + (Comments? Ψ2:0) + (Likes? Ψ3:0) + (Favorites? Ψ4:0), where Pi is the preference feature component corresponding to the feature component Ti. 历史 Let Pi be the value before the current update, Ti be any descriptive feature component in the descriptive feature vector of the historical video, and Vi be the weight component corresponding to the descriptive feature component Ti. f(x) has been described in detail in the above embodiments and will not be repeated in this embodiment.

[0149] In this embodiment, the above formula is used to train and learn each historical video corresponding to each descriptive feature component in turn, so as to finally obtain the preference feature component corresponding to each descriptive feature component.

[0150] As an example, the training process for learning and training each historical video corresponding to the first descriptive feature component T0 to finally obtain the preference feature component P0 corresponding to T0 is as follows:

[0151] First, substitute the viewing completeness, interaction count, and other historical behavior data of the first historical video corresponding to the first descriptive feature component T0 into the formula for calculating f(x) to obtain the f(x) value after the first training. After obtaining the f(x) value after the first training, based on this f(x) value, the weight component corresponding to the first descriptive feature component T0, and Pi... 历史 The value of P0 after the first training iteration is calculated, where Pi is the initial value during the first training iteration. 历史 The default initial value.

[0152] After completing the first training, the viewing completeness, interaction count, and other historical behavior data of the second historical video corresponding to the first descriptive feature component T0 are substituted into the formula for calculating f(x) to obtain the value of f(x) after the second training. After obtaining the value of f(x) after the second training, based on this value of f(x), the weight component corresponding to the first descriptive feature component T0, and Pi... 历史 The value of P0 after the second training is calculated, where Pi is the value after the second training. 历史 This is the P0 value obtained after the first training.

[0153] Similarly, during the final training, the viewing completion rate, interaction count, and other historical behavior data of the last historical video corresponding to the first descriptive feature component T0 are substituted into the formula for calculating f(x) to obtain the f(x) value for the final training. After obtaining the f(x) value for the final training, based on this f(x) value, the weight component corresponding to the first descriptive feature component T0, and Pi... 历史 The P0 value after the last training iteration is calculated, and this P0 value is used as the first preference feature component corresponding to the first descriptive feature component T0. Wherein, Pi... 历史 This is the P0 value obtained after the previous training.

[0154] Step S24: Select videos that meet the preset conditions based on all the calculated preference scores to form a recommendation list and recommend them to the user.

[0155] Specifically, the preset condition is a pre-set condition for selecting videos from the video library as videos in the recommendation list. For example, the preset condition is to select videos with a preference ranking within the top 10. That is, when making video recommendations, videos with a preference value within the top 10 are selected as videos in the recommendation list.

[0156] In one exemplary implementation, to facilitate the selection of videos for the recommended list, see [reference needed]. Figure 10The step of selecting videos that meet preset conditions based on all calculated preference scores to form a recommendation list and recommend them to the user includes: step S100, sorting all calculated preference scores in descending order; step S101, selecting videos with preference scores within a preset position to form a recommendation list and recommending them to the user.

[0157] Specifically, after calculating the user's preference for all the videos to be recommended, these preference values ​​can be sorted in descending or ascending order. Then, videos ranked within a preset position can be selected to form a recommendation list and recommended to the user.

[0158] The preset position is pre-set and can be set and adjusted according to the actual situation. For example, the preset position is the top 10.

[0159] This embodiment determines whether a user likes a target video based on their behavioral data during viewing. If the user likes the target video, a video group logically related to the target video is searched from the video library. Upon receiving a video recommendation instruction, the user's preference feature vector, the description feature vector of each video to be recommended in the video library, and the logical weighted values ​​of all first videos to be recommended in the video group except the target video are obtained. Based on the preference feature vector, the description feature vector of each second video to be recommended in the video library except the video group, and a preset matching degree algorithm, the user's preference for each second video to be recommended is calculated. The system calculates the user's preference for each first video to be recommended based on the preference feature vector, the descriptive feature vector of each first video to be recommended in the video group, the logical weighting value corresponding to each first video to be recommended, and the matching degree algorithm. Based on all the calculated preference scores, videos that meet preset conditions are selected to form a recommendation list and recommended to the user. This ensures that when recommending videos, not only are the characteristics of the videos themselves considered to meet the user's preferences, but also the logical relationship between the videos. This ensures that video recommendations are made according to the logical relationship between the videos, avoiding random video recommendations and improving the user experience.

[0160] See Figure 11 The diagram shown is a program block diagram of an embodiment of the video recommendation device 110 of this application.

[0161] In this embodiment, the video recommendation device 110 includes a series of computer program instructions stored in a memory. When these computer program instructions are executed by a processor, the video recommendation functions of the various embodiments of this application can be implemented. In some embodiments, based on the specific operations implemented by each part of the computer program instructions, the video recommendation device 110 can be divided into one or more modules. Specifically, the modules that can be divided are as follows:

[0162] The determination module 111 is used to determine whether the user likes the target video based on the user's behavioral data during the viewing process of the target video;

[0163] The search module 112 is used to search for a video group that has a logical relationship with the target video from the video library if the user likes the target video;

[0164] The acquisition module 113 is used to acquire, upon receiving a video recommendation instruction, the user's preference feature vector, the description feature vector of each video to be recommended in the video library, and the logical weighted value of all first videos to be recommended in the video group except for the target video;

[0165] The calculation module 114 is used to calculate the user's preference degree for each second video to be recommended based on the preference feature vector, the description feature vector of each second video to be recommended in the video library other than the video group, and a preset matching degree algorithm, and to calculate the user's preference degree for each first video to be recommended based on the preference feature vector, the description feature vector of each first video to be recommended in the video group, the logical weighting value corresponding to each first video to be recommended, and the matching degree algorithm.

[0166] The recommendation module 115 is used to select videos that meet preset conditions based on all calculated preference degrees to form a recommendation list and recommend them to the user.

[0167] In one exemplary embodiment, the behavioral data includes the user's viewing completion rate data, interaction frequency data, and other behavioral data for the target video. The other behavioral data includes at least one of comment data, like data, and collection data. The determination module 111 is further configured to extract the user's viewing completion rate, interaction frequency, and other behavioral data for the target video from the behavioral data; calculate the user's liking value for the target video based on the viewing completion rate, the interaction frequency, the other behavioral data, and a preset liking value algorithm; and determine whether the user likes the target video based on the calculated liking value and a preset threshold.

[0168] In one exemplary embodiment, the video recommendation device 110 further includes an extraction module, a summarization module, a composition module, a sorting module, and a storage module.

[0169] The extraction module is used to extract the content text of each video to be recommended in the video library;

[0170] The summarization module is used to perform semantic summarization on the content text of each video to be recommended using a preset semantic recognition algorithm, so as to obtain the content description information of each video to be recommended.

[0171] The component module is used to determine videos with logical relationships based on the content description information of each video to be recommended, and to group all videos with logical relationships into a video group.

[0172] The sorting module is used to sort all the third videos to be recommended in each video group according to the order of event development, and to configure a corresponding initial weighting value for each third video to be recommended.

[0173] The storage module is used to calculate the logical weighting value corresponding to each third video to be recommended based on the initial weighting value of each third video to be recommended, the user's behavior data for each third video to be recommended, and a preset logical weighting value algorithm, and to associate and save the calculated logical weighting value with the corresponding third video to be recommended.

[0174] In one exemplary embodiment, the extraction module is further configured to: extract the subtitle text of each video to be recommended in the video library using a preset text recognition algorithm to obtain the first content text of each video to be recommended; extract the audio of each video to be recommended in the video library using a preset speech recognition algorithm to obtain the second content text of each video to be recommended; and merge the first content text and the second content text of each video to be recommended in chronological order to obtain the content text of each video to be recommended.

[0175] In an exemplary embodiment, the extraction module is further configured to determine, for each video to be recommended, whether the language text corresponding to the first content text and the second content text of each video to be recommended is similar under the same video progress; if similar, one of the similar language texts is discarded, and the first content text and the second content text after discarding the language text are cross-merged in chronological order to obtain the content text of each video to be recommended.

[0176] In an exemplary embodiment, the search module 112 is further configured to sequentially determine whether there is a video group containing the target video among the multiple video groups contained in the video library; if there is a video group containing the target video, then the video group containing the target video is regarded as a video group that has a logical association with the target video.

[0177] In one exemplary embodiment, the video recommendation device 110 further includes a behavior data acquisition module and a determination module.

[0178] The behavior data acquisition module is used to acquire the user's historical behavior data. The historical behavior data includes the user's viewing completion data, interaction count data, and other historical behavior data for each historical video watched within a historical time period. The other historical behavior data includes at least one of comment data, like data, and collection data. Each historical video has a corresponding descriptive feature vector.

[0179] The determining module is used to determine the user's preference feature vector based on the descriptive feature vector of each historical video, a preset weight vector, the viewing completeness extracted from the viewing completeness data corresponding to each historical video, the number of interactions extracted from the interaction number data corresponding to each historical video, and other historical behavior data corresponding to each historical video.

[0180] In an exemplary embodiment, the descriptive feature vector is composed of multiple descriptive feature components, and the weight vector is composed of multiple weight components. The determining module is further configured to obtain the weight component corresponding to each descriptive feature component from the weight vector; calculate the preference feature component corresponding to each descriptive feature component based on the viewing completeness, interaction count, other historical behavior data, weight component corresponding to each descriptive feature component, and a preset preference feature algorithm for each historical video corresponding to each descriptive feature component, and assemble all the obtained preference feature components into the preference feature vector.

[0181] In one exemplary embodiment, the recommendation module 115 is further configured to sort all the calculated preference scores in descending order; select videos with preference scores within a preset position to form a recommendation list and recommend them to the user.

[0182] Figure 12This illustration schematically shows a hardware architecture diagram of a computer device 12 suitable for implementing a video recommendation method according to an embodiment of this application. In this embodiment, the computer device 12 is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. For example, it may be a tablet computer, laptop computer, desktop computer, rack server, blade server, tower server, or cabinet server (including standalone servers or server clusters composed of servers), etc. Figure 12 As shown, the computer device 12 includes, but is not limited to, at least: a memory 120, a processor 121, and a network interface 122 that can communicate with each other via a system bus. Wherein:

[0183] The memory 120 includes at least one type of computer-readable storage medium, which can be volatile or non-volatile. Specifically, the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 120 may be an internal storage module of the computer device 12, such as the hard disk or memory of the computer device 12. In other embodiments, the memory 120 may also be an external storage device of the computer device 12, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on the computer device 12. Of course, the memory 120 may also include both the internal storage module and the external storage device of the computer device 12. In this embodiment, the memory 120 is typically used to store the operating system and various application software installed on the computer device 12, such as the program code for video recommendation methods. In addition, the memory 120 can also be used to temporarily store various types of data that have been output or will be output.

[0184] In some embodiments, processor 121 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other video recommendation chip. Processor 121 is typically used to control the overall operation of computer device 12, such as performing control and processing related to data interaction or communication with computer device 12. In this embodiment, processor 121 is used to run program code stored in memory 120 or process data.

[0185] Network interface 122 may include a wireless network interface or a wired network interface, which is typically used to establish communication links between computer device 12 and other computer devices. For example, network interface 122 is used to connect computer device 12 to external terminals via a network, establishing data transmission channels and communication links between computer device 12 and external terminals. The network may be an intranet, the Internet, Global System for Mobile Communication (GSM), Wideband Code Division Multiple Access (WCDMA), 4G network, 5G network, Bluetooth, Wi-Fi, or other wireless or wired networks.

[0186] It should be pointed out that, Figure 12 Only computer devices with components 120 to 122 are shown; however, it should be understood that it is not required to implement all of the shown components, and more or fewer components may be implemented instead.

[0187] In this embodiment, the video recommendation method stored in memory 120 can be divided into a program module and executed by a processor (processor 121 in this embodiment) to complete this application.

[0188] This application provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the video recommendation method in the embodiment.

[0189] In this embodiment, the computer-readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the computer-readable storage medium may be an internal storage unit of a computer device, such as the hard disk or memory of the computer device. In other embodiments, the computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device. Of course, the computer-readable storage medium may include both the internal storage unit and the external storage device of the computer device. In this embodiment, the computer-readable storage medium is typically used to store the operating system and various application software installed on the computer device, such as the program code of the video recommendation method in the embodiment. In addition, the computer-readable storage medium can also be used to temporarily store various types of data that have been output or will be output.

[0190] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across at least two network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of this application. Those skilled in the art can understand and implement this without any creative effort.

[0191] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0192] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A video recommendation method, characterized in that, The method includes: Determine whether the user likes the target video based on the user's behavioral data during the viewing process; If the user likes the target video, then search the video library for video groups that have a logical relationship with the target video; Upon receiving a video recommendation instruction, the system obtains the user's preference feature vector, the description feature vector of each video to be recommended in the video library, and the logical weighted value of all first videos to be recommended in the video group except for the target video. The user's preference for each second recommended video is calculated based on the preference feature vector, the description feature vector of each second recommended video in the video library other than the video group, and a preset matching degree algorithm. The user's preference for each first recommended video is also calculated based on the preference feature vector, the description feature vector of each first recommended video in the video group, the logical weighted value corresponding to each first recommended video, and the matching degree algorithm. Based on all the calculated preference scores, select videos that meet the preset conditions to form a recommendation list and recommend them to the user; The method further includes: Based on the content description information of each video to be recommended in the video library, videos with logical relationships are determined, and all videos with logical relationships are grouped into a video group. For each video group, all third videos to be recommended within that video group are sorted according to the order of events, and an initial weighting value Ψ is assigned to each third video to be recommended, Ψ = {Ψ0, Ψ1, Ψ2, ..., Ψn}, where Ψ0, Ψ1, Ψ2, ..., Ψn are the initial weighting values ​​for each video in the video group after being sorted according to the order of events; Ψi < Ψ(i+1), where i refers to the i-th video in the video group; The logical weighting value for each third video to be recommended is calculated based on the initial weighting value of each video, the user's behavior data for each video, and the preset logical weighting algorithm. The calculated logical weighting value is then associated with and saved with the corresponding video.

2. The video recommendation method according to claim 1, characterized in that, The behavioral data includes data on the user's viewing completion rate of the target video, interaction frequency data, and other behavioral data. The other behavioral data includes at least one of comment data, like data, and collection data. Determining whether the user likes the target video based on the user's behavioral data during viewing includes: Extract the user's viewing completion rate, number of interactions, and other behavioral data from the behavioral data; The user's liking value for the target video is calculated based on the viewing completion rate, the number of interactions, other behavioral data, and a preset liking value algorithm. The user's liking value is then determined based on the calculated liking value and a preset threshold to determine whether the user likes the target video.

3. The video recommendation method according to claim 1, characterized in that, The method further includes: Extract the content text of each video to be recommended from the video library; A pre-defined semantic recognition algorithm is used to semantically summarize the content text of each video to be recommended, thereby obtaining the content description information of each video to be recommended.

4. The video recommendation method according to claim 3, characterized in that, The extraction of the content text of each video to be recommended from the video library includes: A preset text recognition algorithm is used to extract the subtitle text of each video to be recommended in the video library, so as to obtain the first content text of each video to be recommended; The speech of each video to be recommended in the video library is extracted using a preset speech recognition algorithm to obtain the second content text of each video to be recommended; The first and second content texts of each video to be recommended are merged in chronological order to obtain the content text of each video to be recommended.

5. The video recommendation method according to claim 4, characterized in that, The step of cross-merging the first and second content texts of each video to be recommended in chronological order to obtain the content text of each video to be recommended includes: For each video to be recommended, determine whether the language text corresponding to the first and second content texts of each video to be recommended is similar under the same video progress. If similar, one of the similar language texts is discarded, and the first and second content texts after discarding the language texts are cross-merged in chronological order to obtain the content text of each video to be recommended.

6. The video recommendation method according to claim 3, characterized in that, The step of searching the video library for video groups that have a logical relationship with the target video includes: Sequentially determine whether there is a video group containing the target video among the multiple video groups contained in the video library; If a video group contains the target video, then the video group containing the target video is considered to be a video group with a logical relationship to the target video.

7. The video recommendation method according to any one of claims 1 to 6, characterized in that, The method further includes: The user's historical behavior data is obtained. The historical behavior data includes the user's viewing completion data, interaction count data, and other historical behavior data for each historical video watched within a historical time period. The other historical behavior data includes at least one of comment data, like data, and collection data. Each historical video has a corresponding descriptive feature vector. The user's preference feature vector is determined based on the descriptive feature vector of each historical video, the preset weight vector, the viewing completeness extracted from the viewing completeness data corresponding to each historical video, the number of interactions extracted from the interaction number data corresponding to each historical video, and other historical behavior data corresponding to each historical video.

8. The video recommendation method according to claim 7, characterized in that, The descriptive feature vector is composed of multiple descriptive feature components, and the weight vector is composed of multiple weight components. Determining the user's preference feature vector based on the descriptive feature vectors of each historical video, the preset weight vector, the viewing completeness extracted from the viewing completeness data corresponding to each historical video, the number of interactions extracted from the interaction count data corresponding to each historical video, and other historical behavior data corresponding to each historical video includes: Obtain the weight component corresponding to each descriptive feature component from the weight vector; The preference feature component corresponding to each descriptive feature component is calculated based on the viewing completeness, interaction count, other historical behavior data, weight component corresponding to each descriptive feature component, and preset preference feature algorithm for each historical video corresponding to each descriptive feature component, and all the obtained preference feature components are combined to form the preference feature vector.

9. The video recommendation method according to claim 1, characterized in that, The step of selecting videos that meet preset conditions based on all calculated preference levels to form a recommendation list and recommending them to the user includes: Sort all the calculated preference levels in descending order; The recommended list is created by selecting videos that rank within a preset position based on user preference and then recommending them to the user.

10. A video recommendation device, characterized in that, The video recommendation device includes: The determination module is used to determine whether the user likes the target video based on the user's behavioral data during the viewing process. The search module is used to search for video groups that have a logical relationship with the target video from the video library if the user likes the target video; The acquisition module is used to acquire, upon receiving a video recommendation instruction, the user's preference feature vector, the description feature vector of each video to be recommended in the video library, and the logical weighted value of all the first videos to be recommended in the video group except for the target video; The calculation module is used to calculate the user's preference degree for each second recommended video based on the preference feature vector, the description feature vector of each second recommended video in the video library other than the video group, and a preset matching degree algorithm, and to calculate the user's preference degree for each first recommended video based on the preference feature vector, the description feature vector of each first recommended video in the video group, the logical weighted value corresponding to each first recommended video, and the matching degree algorithm. The recommendation module is used to select videos that meet preset conditions based on all calculated preference scores to form a recommendation list and recommend them to the user. The composition module is used to determine videos with logical relationships based on the content description information of each video to be recommended in the video library, and to group all videos with logical relationships into a video group; The sorting module is used to sort all the third videos to be recommended in each video group according to the order of events, and to configure an initial weighting value Ψ for each third video to be recommended, Ψ = {Ψ0, Ψ1, Ψ2, ..., Ψn}, where Ψ0, Ψ1, Ψ2, ..., Ψn are the initial weighting values ​​of each video in the video group after being sorted according to the order of events; Ψi < Ψ(i+1), where i refers to the i-th video in the video group; The component module is used to calculate the logical weighting value corresponding to each third video to be recommended based on the initial weighting value of each third video to be recommended, the user's behavior data for each third video to be recommended, and a preset logical weighting value algorithm, and to associate and save the calculated logical weighting value with the corresponding third video to be recommended.

11. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

13. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

Images