A content-based video recommendation method, apparatus, device, and medium

Abstract

This invention discloses a video recommendation method, apparatus, device, and medium based on content recommendation. The method includes: obtaining a set of videos of interest to a first user; extracting a target feature vector from the set of videos of interest; inputting the target feature vector into a content recommendation model based on the Rocchio algorithm to obtain a first set of recommended videos for the first user; obtaining a second set of recommended videos for a second user; and obtaining a target recommended video for the first user based on the first and second recommended video sets; the second user is a user who has an association or potential association with the first user. This invention can recommend videos that match user preferences based on content recommendation methods and video characteristics and user preferences, and can be widely applied in the field of intelligent recommendation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent recommendation, and in particular to a video recommendation method and device based on content recommendation, equipment and a medium. BACKGROUND

[0002] The content-based (CB) recommendation method is basically proposed at the same time as the collaborative filtering recommendation method. At present, the content recommendation method can make recommendations directly based on the content description information of an item. Essentially, it is based on direct analysis and calculation of the characteristics or attributes of the item and the user. The content recommendation method establishes user profiles based on the past browsing records of the user, extracts feature vectors of the item, establishes item profiles, and compares the similarity between the item profiles and the user profiles (including the information of the items purchased by the user) to recommend other items similar to the browsing records of the user. However, the existing video recommendation method based on content recommendation still has the problem of low accuracy of recommended videos. SUMMARY

[0003] Therefore, the embodiments of the present application provide a video recommendation method and device based on content recommendation to improve the accuracy of recommended videos.

[0004] An aspect of the embodiments of the present application provides a video recommendation method based on content recommendation, comprising:

[0005] obtaining a set of videos of interest of the first user;

[0006] extracting a target feature vector from the set of videos of interest;

[0007] inputting the target feature vector into a content recommendation model based on the Rocchio algorithm to obtain a first set of recommended videos of the first user;

[0008] obtaining a second set of recommended videos of a second user, and obtaining a target recommended video of the first user according to the first set of recommended videos and the second set of recommended videos; the second user is a user having an association relationship or a potential association relationship with the first user.

[0009] Optionally, the obtaining of the set of videos of interest of the first user comprises:

[0010] obtaining a plurality of historical videos watched by the first user for a time length reaching a set threshold and scored within a set time range;

[0011] extracting content features of the plurality of historical videos, and obtaining a plurality of videos as the set of videos of interest according to the content features.

[0012] Optionally, the extracting the content features of the plurality of historical videos comprises:

[0013] extracting keywords or phrases in the historical videos, classification of the historical videos, scores and comments of the historical videos by the first user, keywords in the comments, and whether the historical videos are scored within a set time range after watching, as the content features.

[0014] Optionally, the extracting the target feature vector from the set of videos of interest comprises:

[0015] obtaining a user feature vector from the set of videos of interest;

[0016] inputting the user feature vector into a pre-constructed regression model to obtain a target feature vector;

[0017] the expression of the regression model is:

[0018] ||Xθ-y|| 2 +||Γθ|| 2

[0019] wherein X represents the user feature vector; y represents the target feature vector; || represents norm; Γ=aI, Γ represents the fitting value;

[0020] the regression model performs an anti-overfitting operation according to the following expression:

[0021] θ(a)=(X T X+aI) -1 X T y

[0022] wherein I is a unit matrix, a is the weight of the unit matrix; θ is a fitting hyperparameter; T represents matrix transposition; θ(a) represents θ determined in the case of a.

[0023] Optionally, the obtaining the user feature vector from the set of videos of interest comprises:

[0024] obtaining a first vector corresponding to videos of interest and a second vector corresponding to videos of no interest in the historical videos watched by the first user;

[0025] obtaining a current feature vector of the first user from the set of videos of interest;

[0026] determining the user feature vector according to the first vector, the second vector, and the current feature vector.

[0027] Optionally, determining the user feature vector based on the first vector, the second vector, and the current feature vector includes:

[0028] The user feature vector is determined based on a first expression, which is:

[0029] U opt =arg max[Sim(U c ,V1)-Sim(U c V h )]

[0030] Among them, U opt U represents the user feature vector. C V represents the current feature vector. l V represents the first vector. h This represents the second vector; max is the function for finding the maximum value.

[0031] Sim(U c V l Sim(U) represents calculating the similarity between the current feature vector and the first vector. c V h The expression () represents calculating the similarity between the current feature vector and the second vector; cosine similarity is used to distinguish between relevant and irrelevant feature vectors in the first expression, resulting in the second expression, which is:

[0032]

[0033] Where || represents the magnitude of the vector. Let ω1 and W2 be the j-th feature vector of the video, and let ω1 and W2 represent two feature vectors in the same video.

[0034] The current feature vector is updated according to the third expression, which is:

[0035]

[0036] Where U0 represents the historical feature vector of the first user, α, β, and γ represent the weights of the historical feature vector, the weight of the first vector, and the weight of the second vector, respectively, and j, v, and ω represent the feature vectors of different videos, respectively. j express eigenvectors;

[0037] The current feature vector obtained from the third expression is updated according to the fourth expression, which is:

[0038]

[0039] wherein, is the i-th feature vector of the video, ω l ∈V l denotes ω l is a feature vector in the video of interest, ω j ∈V h denotes ω j is a feature vector in the video of non-interest.

[0040] Optionally, the target recommended video of the first user is obtained according to the first recommended video set and the second recommended video set, comprising:

[0041] performing weighted summation on the feature vectors corresponding to the first recommended video set and the feature vectors corresponding to the second recommended video set to obtain a comprehensive feature vector;

[0042] obtaining a corresponding recommended video according to the comprehensive feature vector as the target recommended video.

[0043] Another aspect of the embodiment of the present application further provides a video recommendation device based on content recommendation, comprising:

[0044] a first unit configured to obtain a video set of interest of the first user;

[0045] a second unit configured to extract a target feature vector from the video set of interest;

[0046] a third unit configured to input the target feature vector into a content recommendation model based on a Rocchio algorithm to obtain a first recommended video set of the first user;

[0047] a fourth unit configured to obtain a second recommended video set of a second user and obtain a target recommended video of the first user according to the first recommended video set and the second recommended video set; the second user is a user having an associated relationship or a potential associated relationship with the first user.

[0048] Another aspect of the embodiment of the present application further provides an electronic device comprising a processor and a memory;

[0049] the memory is configured to store a program;

[0050] the processor executes the program to realize the method.

[0051] Another aspect of the embodiment of the present application further provides a computer readable storage medium, the storage medium stores a program, and the program is executed by a processor to realize the method.

[0052] The embodiment of the present application further discloses a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device can read the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the foregoing method.

[0053] Compared with the prior art, the present application can at least include the following beneficial effects:

[0054] The present application can obtain a first user's interested video set, which covers various types of videos in the first user's historical browsing process, so that a target feature vector close to the user's real preference can be obtained based on the interested video set; then, a content recommendation model based on the Rocchio algorithm is used to obtain a recommended video corresponding to the target feature vector as a first recommended video set according to the target feature vector close to the user's real preference, so that real-time and accurate recommendation is obtained; and the second recommended video set corresponding to a second user having an association relationship or a potential association relationship with the first user is combined with the first recommended video set, so that the video recommendation result based on content and the video recommendation result of the associated user or the potential user are fused to obtain more accurate video recommendation content. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0056] Figure 1 A flowchart of a video recommendation method based on content recommendation provided by the embodiment of the present application is shown in the figure.

[0057] Figure 2 A flowchart of obtaining a first user's interested video set provided by the embodiment of the present application is shown in the figure.

[0058] Figure 3 A flowchart of obtaining four index association data provided by the embodiment of the present application is shown in the figure.

[0059] Figure 4 A flowchart of extracting a target feature vector provided by the embodiment of the present application is shown in the figure.

[0060] Figure 5 A flowchart of obtaining a user feature vector provided by the embodiment of the present application is shown in the figure.

[0061] Figure 6 This is a schematic diagram of a process for obtaining target recommended videos according to an embodiment of the present invention;

[0062] Figure 7 A specific example flowchart of a content-based video recommendation method provided in this embodiment of the invention;

[0063] Figure 8 This is a schematic diagram illustrating a video recommendation effect provided in an embodiment of the present invention;

[0064] Figure 9 A structural block diagram of a content-based video recommendation device provided in an embodiment of the present invention;

[0065] Figure 10 A structural block diagram of the first unit provided in an embodiment of the present invention;

[0066] Figure 11 This is a structural block diagram of the second unit provided in an embodiment of the present invention;

[0067] Figure 12 This is a structural block diagram of the user feature vector acquisition unit provided in an embodiment of the present invention;

[0068] Figure 13 A structural block diagram of the fourth unit provided in an embodiment of the present invention;

[0069] Figure 14 This is a hardware structure block diagram of a video recommendation electronic device provided in an embodiment of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0071] It should be understood that in the description of the embodiments of the present invention, "multiple" (or "amounts") means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. If "first," "second," etc., are used in the description, they are only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0072] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0073] It should be noted that although functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart.

[0074] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0075] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0076] Before providing a further detailed description of the embodiments of the present invention, the nouns and terms involved in the embodiments of the present invention are explained, and the nouns and terms involved in the embodiments of the present invention are subject to the following interpretations:

[0077] Rocchio Algorithm: The Rocchio algorithm is an efficient classification algorithm widely used in text or item classification, as well as query expansion. It obtains the optimal solution by constructing prototype vectors.

[0078] The basic idea of ​​the Rocchio algorithm is to construct a prototype vector for each class using the training set, as follows:

[0079] Given a class, the vector components of all documents belonging to this class in the training set are represented by positive numbers, and the vector components of all documents not belonging to this class are represented by negative numbers. The sum of all these vectors gives the prototype vector of this class. The similarity between two vectors is defined as the cosine of the angle between them. The similarity between each document in the training set and the prototype vector is calculated one by one, and a certain similarity is selected as a bound using a specific algorithm. Given a document, if the document has a high similarity to the prototype vector, then the document belongs to this class; otherwise, the document does not belong to this class. The Rocchio algorithm's outstanding advantages are its ease of implementation and its particularly simple computation (training and classification). It is often used to implement benchmark systems for measuring the performance of classification systems, but practical classification systems rarely use this algorithm to solve specific classification problems.

[0080] Content-based recommendation methods: Content-based recommendation is a continuation and development of information filtering technology. It makes recommendations based on the content information of items, rather than relying on user evaluations of items. It relies more on machine learning methods to obtain user interest information from examples of content feature descriptions.

[0081] In content-based recommendation systems, items or objects are defined by attributes of relevant features. The system assesses the degree of match between user profiles and the items to be predicted based on the features of user-rated objects and the user's interests. The user profile model depends on the learning method used, commonly including decision trees, neural networks, and vector-based representations. Content-based user profiles require historical user data, and the user profile model may change as user preferences evolve.

[0082] Content-based recommendation is similar to demographic-based recommendation, except that the focus of system evaluation shifts to the items themselves, using the similarity of items rather than the similarity of users for recommendation.

[0083] The advantages of content-based recommendation algorithms are: they can effectively model user interests and achieve better recommendation accuracy by increasing the dimensions of item attributes. However, their disadvantages are: ① the limited number of item attributes makes it difficult to obtain more data effectively; ② the criteria for measuring item similarity only consider the items themselves, which is somewhat one-sided; ③ they require historical data of the user's items, leading to the cold start problem.

[0084] Collaborative filtering recommendation is one of the earliest and most successful techniques used in recommendation systems. It typically employs nearest neighbor technology, calculating the distance between users based on their historical preferences. Then, it uses the weighted ratings of the target user's nearest neighbors to predict the target user's preference for specific products, and makes recommendations based on this preference level.

[0085] The biggest advantage of collaborative filtering-based recommendation algorithms is that they do not have special requirements for the recommended objects and can handle unstructured complex objects such as music and videos. Collaborative filtering recommendation methods do not require any descriptive information about the recommended items; the recommendation process is entirely based on the collaborative principle. Its disadvantage is that it cannot directly make recommendations based on item characteristics and user preferences.

[0086] Collaborative filtering-based recommendation algorithms are based on the assumption that the best way to find content a user is truly interested in is to first find other users with similar interests, and then recommend content that interests them. The basic idea is very easy to understand; in daily life, people often use recommendations from friends to make choices. Collaborative filtering-based recommendation algorithms apply this idea to e-commerce recommendation systems, recommending content to the target user based on other users' evaluations of that content.

[0087] Recommendation systems based on collaborative filtering can be said to make recommendations from the user's perspective, and this is automatic. That is, the recommendations that users receive are implicitly obtained by the system from purchasing patterns or browsing behavior, without requiring users to make efforts to find recommendations that suit their interests, such as by filling out survey forms.

[0088] It is understood that the content-based video recommendation method provided in this embodiment of the invention can be applied to any computer device with data processing and computing capabilities, and this computer device can be various terminals or servers. When the computer device in this embodiment of the invention is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal can be a smartphone, tablet computer, laptop computer, or desktop computer, but it is not limited to these.

[0089] It should be further noted that the terminals involved in the embodiments of the present invention include, but are not limited to, smartphones, computers, smart voice interaction devices, smart home appliances, and vehicle terminals.

[0090] In some possible implementations, a computer program capable of implementing a content-based video recommendation method provided in the embodiments of the present invention can be deployed and executed on a computer device, or on multiple computer devices located in one location, or on multiple computer devices distributed in multiple locations and interconnected through a communication network.

[0091] Furthermore, it should be noted that in various specific embodiments of this application, when processing data related to the identity or characteristics of an object, such as user information, object behavior data, object historical data, and object location information, the object's permission or consent is obtained first. Moreover, the collection, use, and processing of this data comply with the relevant laws, regulations, and standards of the relevant countries and regions. Additionally, when embodiments of this application require access to sensitive object information, separate permission or consent from the object is obtained through pop-ups or redirection to a confirmation page. Only after obtaining the object's separate permission or consent is the necessary object-related data required for the proper functioning of the embodiments of this application obtained.

[0092] This invention belongs to the field of intelligent recommendation. Specifically, it addresses the Rocchio algorithm, a commonly used algorithm for constructing user profiles by extracting video features favored by users from their viewing history. Rocchio is a well-known algorithm in information retrieval for handling relevance feedback, providing a method for calculating attribute values ​​in a user's feature vector based on user viewing feedback. In video recommendation systems, the Rocchio algorithm continuously modifies the user's original feature vector based on historical data, achieving real-time updates to the user profile. The Rocchio algorithm is based on the assumption that if this invention requires calculating the most accurate user feature vector U, then this user feature vector should be most similar to the features of videos the user likes and least similar to the features of videos the user dislikes.

[0093] Reference Figure 1 This invention provides a content-based video recommendation method, which specifically includes the following steps:

[0094] S100: Obtain the set of videos that the first user is interested in.

[0095] Specifically, the set of videos of interest can include multiple videos that match the first user's preferences.

[0096] Reference Figure 2This invention provides a flowchart illustrating the process of obtaining a first set of videos that a user is interested in.

[0097] Specifically, the process of obtaining the first set of videos that the user is interested in may include:

[0098] S101: Obtain multiple historical videos that the first user has rated within a set time range and whose viewing time has reached a set threshold.

[0099] Specifically, if a user watches a video for a set duration, the user is considered interested in that video, and this embodiment of the invention can then include that video as a candidate for historical videos of interest. The threshold for viewing duration can be freely set, and this embodiment of the invention does not impose strict limitations on it.

[0100] In one optional implementation, the viewing duration threshold of this invention can be set to one minute. That is, if a user watches the video for one minute, the video can be selected as a candidate video of historical interest.

[0101] Furthermore, for the aforementioned candidate videos of historical interest, this embodiment of the invention exemplarily provides two optional implementation methods for obtaining these videos: one is to rate them immediately after viewing (i.e., the set time range is 0); the other is to rate them within a certain period of time after viewing (e.g., 3, 5, or 7 days). This invention does not impose specific limitations on the set time range, and the method can be selected according to actual needs.

[0102] Therefore, if the first user rates the video immediately after watching it, or rates it within a set time range, then it can be assumed that the first user is genuinely interested in the candidate video of historical interest, and thus the candidate video of historical interest that the first user has rated is identified as the historical video.

[0103] S102: Extract content features from multiple historical videos, and obtain multiple videos based on the content features as the set of videos of interest.

[0104] Specifically, extracting content features from multiple historical videos may include: extracting keywords or phrases from the historical videos, the classification of the historical videos, the ratings and comments of the first user on the historical videos, keywords in the comments, and whether the historical videos were rated within a set time range after viewing, as the content features.

[0105] For example, the extraction of content features from multiple historical videos is further explained as follows:

[0106] In this embodiment of the invention, the above-mentioned content features may include four indicators of historical videos and corresponding associated data.

[0107] Specifically, when using Rocchio's content-based recommendation algorithm to determine the four metrics for historical videos, these four metrics may include: the overall mean of all ratings, the video's rating tendency, the user's rating tendency, and the time elapsed since the user's first rating. It should be noted that the videos described in this embodiment of the invention may include short videos, movies, documentaries, or live streams, etc.

[0108] Reference Figure 3 Steps T1 to T4 are the process of acquiring the associated data of the four indicators.

[0109] Indicator 1: Correlation data of the overall mean of all scores.

[0110] Key Metric: Videos that users are interested in.

[0111] The process of obtaining associated data features is as follows: First, the historical database is accessed, and videos with longer user dwell times are sorted. Videos with dwell times exceeding one minute are identified as content of interest. Second, keywords or phrases are extracted from this portion of video content, and videos containing these keywords or phrases are queried in the database to form a data set of associated videos of interest.

[0112] Indicator 2: Correlation data on the video's rating tendency.

[0113] Key metrics: Video category and features.

[0114] The process of obtaining related data features is as follows: from the historical database, obtain users' classification of videos, video ratings, extract key fields from the comments, and improve the multi-dimensional feature data of the corresponding videos.

[0115] Indicator 3: Correlation data of user rating preferences.

[0116] Key metric: User rating preferences.

[0117] The process of obtaining related data features is as follows: obtain user ratings for videos from historical databases, obtain multi-dimensional feature data that users are interested in, and recommend content data that matches the user's dimensions based on the feature data that the user likes.

[0118] Indicator 4: Relational data on the time elapsed since the user's first rating.

[0119] Key Metric: Rate immediately after viewing.

[0120] The process of obtaining related data features is as follows: rating after watching for a period of time.

[0121] S110: Extract the target feature vector from the set of videos of interest.

[0122] Reference Figure 4 The step of extracting the target feature vector from the set of videos of interest may include:

[0123] S111. Obtain the user feature vector from the set of videos of interest;

[0124] S112. Input the user feature vector into a pre-built regression model to obtain the target feature vector;

[0125] The expression for the regression model is:

[0126] ||Xθ-y|| 2 +||Γθ|| 2

[0127] Where X represents the input, i.e., the user feature vector; y represents the output, i.e., the target feature vector; || represents regularization; Γ = aI, where Γ represents the fitted value;

[0128] The regression model performs overfitting calculations based on the following expression:

[0129] θ(a)=(X T X+aI) -1 X T y

[0130] Where I is the identity matrix, a is the weight of the identity matrix; θ is the fitting hyperparameter; T denotes matrix transpose; θ(a) means finding θ given a.

[0131] Furthermore, the process of obtaining the user feature vector from the set of videos of interest in S111 described above will be explained in more detail, such as... Figure 5 As shown, the process may include:

[0132] S1110. Obtain the first vector corresponding to the videos that the first user is interested in from the videos he / she has watched in his / her history, and the second vector corresponding to the videos that he / she is not interested in.

[0133] S1111. Obtain the current feature vector of the first user from the set of videos of interest;

[0134] S1112. Determine the user feature vector based on the first vector, the second vector, and the current feature vector.

[0135] S120: Input the target feature vector into the content recommendation model based on the Rocchio algorithm to obtain the first recommended video set for the first user.

[0136] First, it should be noted that the first vector corresponding to the videos that the first user is interested in from his / her historical video viewing history can represent the vector corresponding to the videos that match the user's preferences; the second vector corresponding to the videos that the first user is not interested in from his / her historical video viewing history can represent the vector corresponding to the videos that the first user is least interested in.

[0137] To further explain, the determination of the user feature vector based on the first vector, the second vector, and the current feature vector in S1112 is described in detail below:

[0138] The user feature vector is determined based on a first expression, which is:

[0139] U opt =arg max[Sim(U c ,V1)-Sim(U c V h )]

[0140] Among them, U opt U represents the user feature vector. C V represents the current feature vector. l V represents the first vector. h This represents the second vector; max is the function for finding the maximum value.

[0141] Sim(U c V l Sim(U) represents calculating the similarity between the current feature vector and the first vector. c V h The expression () represents calculating the similarity between the current feature vector and the second vector; cosine similarity is used to distinguish between relevant and irrelevant feature vectors in the first expression, resulting in the second expression, which is:

[0142]

[0143] Where || represents the magnitude of the vector. Let ω1 and W2 be the j-th feature vector of the video, and let ω1 and W2 represent two feature vectors in the same video.

[0144] After distinguishing the feature vectors according to the second expression, the user feature vectors are updated according to the third expression, which is:

[0145]

[0146] Where U0 represents the historical feature vector of the first user, α, β, and γ represent the weights of the historical feature vector, the weight of the first vector, and the weight of the second vector, respectively, and j, v, and ω represent the feature vectors of different videos, respectively. j express One of the feature vectors;

[0147] The current feature vector obtained from the third expression is updated according to the fourth expression, which is:

[0148]

[0149] in, Let ω be the i-th feature vector of the video. l ∈V l Represents ω l The feature vector is a feature vector belonging to the video of interest, ω j ∈V h Represents ω j The feature vectors belong to the feature vectors of the videos that are not of interest.

[0150] S130: Obtain the second recommended video set of the second user, and obtain the target recommended video of the first user based on the first recommended video set and the second recommended video set; the second user is a user who has an association or potential association with the first user.

[0151] It should be noted that the first user is the target user of the videos to be recommended in this embodiment of the invention, while the second user is other users who have a relationship or potential relationship with the first user. Other users who have a relationship can include other users who are friends on social chat software, other users who are following each other, or other users who are being followed. Other users who have a potential relationship can include other users who are interested in the same category of videos, other users in the same region, or other users who are watching videos at the same time.

[0152] Reference Figure 6 This invention provides a flowchart illustrating the process of obtaining target recommended videos.

[0153] Specifically, the process of obtaining target recommended videos may include:

[0154] S131. The feature vector corresponding to the first recommended video set and the feature vector corresponding to the second recommended video set are weighted and summed to obtain a comprehensive feature vector.

[0155] S132. Obtain the corresponding recommended video based on the comprehensive feature vector, and use it as the target recommended video.

[0156] To more clearly illustrate the implementation process of the present invention, the application process of the present invention will be described below with specific examples.

[0157] Reference Figure 7 This invention provides a specific example flowchart of a content-based video recommendation method, which may include the following:

[0158] S1. By using four indicators from historical data, specifically including the overall mean of all ratings, the video rating tendency, the user rating tendency, and the time elapsed since the user's first rating, and inputting the correlation data of the four indicators into the regression model, effective training data is obtained, making the regression model calculation more accurate.

[0159] The expression for the regression model is:

[0160] ||Xθ-y|| 2 +||Γθ|| 2

[0161] Where X represents the input, i.e., the user feature vector; y represents the output, i.e., the target feature vector; || represents regularization; Γ=aI represents the fitted value;

[0162] The regression model performs overfitting calculations based on the following expression:

[0163] θ(a)=(X T X+aI) -1 X T y

[0164] Where I is the identity matrix, a is the weight of the identity matrix; θ is the fitting hyperparameter; T denotes matrix transpose; θ(a) means finding θ given a.

[0165] Compared to existing technologies, the embodiments of the present invention can use four indicators of historical data to extract related data features, thus making up for the lack of analytical indicators in the Rocchio algorithm.

[0166] First, a content-based recommendation algorithm based on the Rocchio algorithm is used to extract associated user feature vectors and common features to generate a dataset using four indicators from historical data: the overall mean of all ratings, the rating tendency of videos (such as movies, short videos, etc.), the user's rating tendency, and the time since the user's first rating.

[0167] Secondly, the data related to the four indicators were obtained.

[0168] Indicator 1: Correlation data of the overall mean of all scores.

[0169] Key Metric: Videos that users are interested in.

[0170] Related data feature description: First, the historical access database sorts video data with longer user dwell time, identifying videos viewed for more than 1 minute as content of interest. Second, keywords or phrases are extracted from this portion of video content, and videos containing these keywords or phrases are queried in the database to form a data set of related videos of interest.

[0171] Indicator 2: Correlation data on the video's rating tendency.

[0172] Key metrics: Video category and features.

[0173] Related data feature description: From historical databases, obtain user classifications and ratings of videos, extract key fields from comments, and improve the multi-dimensional feature data of the corresponding videos.

[0174] Indicator 3: Correlation data of user rating preferences.

[0175] Key metric: User rating preferences.

[0176] Related data feature description: Obtain user ratings for videos from historical databases, acquire multi-dimensional feature data that users are interested in, and recommend content data that matches user dimensions based on the feature data that users like.

[0177] Indicator 4: Relational data on the time elapsed since the user's first rating.

[0178] Key Metric: Rate immediately after viewing.

[0179] Related data feature description: Rating after watching for a period of time.

[0180] First, the time users watched and rated videos is retrieved from the historical database to determine the time difference between viewing and rating. Immediately rating after watching indicates that the user likes this type of video. Second, for videos with a small time difference between viewing and rating, their content features are extracted. The database is then searched for videos containing those keywords and phrases to create a dataset of videos of interest.

[0181] Finally, common features of the data associated with the four indicators are extracted and a dataset is generated. This dataset is then input into a content recommendation model based on the Rocchio algorithm to obtain new real-time viewing feedback updates, thereby generating more accurate recommendation results.

[0182] S2. Construct a regression model to improve the effectiveness of training data for the Rocchio-based content recommendation model. Simultaneously, using the Rocchio-based content recommendation model, user feature vectors are extracted from the four historical and related indicators from the previous step. The Rocchio algorithm is then used to update the user profile based on the existing user profile, incorporating real-time feedback from new user viewing to generate more accurate recommendations. Furthermore, the update cost is relatively low.

[0183] Finally, this embodiment of the invention can obtain a more accurate recommendation by weighting the predicted results of new real-time viewing feedback from associated or potential users with the predicted results of new real-time viewing feedback from the previous step. This compensates for the lack of indicators for associated or potential users in the Rocchio algorithm's analysis metrics, while also having a low update cost.

[0184] The specific description is as follows:

[0185] In the video recommendation process, the Rocchio algorithm continuously modifies the user's original feature vector based on the user's historical data, achieving real-time updates to the user profile. The Rocchio algorithm is based on the assumption that if this embodiment of the invention needs to calculate the most accurate user feature vector U, then this user feature vector should be most similar to the features of videos the user likes and least similar to the features of videos the user dislikes. l V corresponds to the videos that users like. h For videos that users dislike, the optimal user feature vector can be defined based on the Rocchio algorithm and the first expression, which is:

[0186] U opt =arg max[Sim(U c ,V1)-Sim(U c V h )]

[0187] Among them, U opt U represents the user feature vector. C V represents the current feature vector. l V represents the first vector. h This represents the second vector; max is the function for finding the maximum value.

[0188] Sim(U c V l Sim(U) represents calculating the similarity between the current feature vector and the first vector. c V hThe expression () represents calculating the similarity between the current feature vector and the second vector; cosine similarity is used to distinguish between relevant and irrelevant feature vectors in the first expression, resulting in the second expression, which is:

[0189]

[0190] Where || represents the magnitude of the vector. Let ω1 and W2 be the j-th feature vector of the video, and let ω1 and W2 represent two feature vectors in the same video.

[0191] In the actual calculation of user feature vectors, the given target user feature vector may already exist. In this case, it is necessary to update the user's feature vector by updating the current feature vector according to the third expression, which is:

[0192]

[0193] In the formula, U0 is the original user feature vector, and α, β, and γ are the weights, respectively. If there is a lot of new historical data for the user, the values ​​of β and γ can be increased; conversely, if the user updates less data, the values ​​of β and γ can be appropriately decreased. In the Rocchio algorithm, the weights corresponding to the weight components of the vector are automatically modified to 0. In practice, for example, weights can be set as β = 0.8 and γ = 0.2, because in actual recommendations, the influence of positive feedback is often greater than that of negative feedback. The video recommendation effect of applying the embodiments of this invention is as follows: Figure 8 As shown in the figure. Here, 1 represents the recommended video from the initial query, and 2 represents the recommended video from the modified query.

[0194] Based on video content and historical behavioral data, this invention can use the Rocchio algorithm to continuously adjust the user's feature vector U to build user profiles. c .

[0195]

[0196] In the formula, U0 is the original user feature vector (initial U = 0), ω i This represents the attribute of video i. V l and V h These represent the sets of videos that the user liked and disliked.

[0197] For example, in a video collection with ratings from 1 to 5, assuming a user rating below 3 indicates dislike, then V l For videos that users have watched and rated 3 stars or higher, V hThe set of videos is rated 1-2 by the user. β and γ are the weights of U0 and positive / negative feedback mentioned above. The user profile is updated based on the original user profile using the Rocchio algorithm, incorporating new real-time viewing feedback to generate more accurate recommendations, while maintaining a low update cost.

[0198] Regression model formula: ||Xθ-y|| 2 +||Γθ|| 2 ;

[0199] X = Input; y = Output (prediction result); || = Regular expression;

[0200] In this embodiment of the invention, Γ = aI (objective training result = fitted value);

[0201] The process to prevent overfitting is as follows: θ(a) = (X T X+aI) -1 X T y and I are identity matrices.

[0202] θ represents the fitting hyperparameters; T represents the matrix transpose;

[0203] a is the weight of the identity matrix;

[0204] θ(a) is used to find θ when a is determined.

[0205] Finally, the model is trained on the four indicators and related indicator data to obtain the loss function, thereby improving the effective training data of Rocchio's CB recommendation algorithm and filtering out invalid data.

[0206] Reference Figure 9 This invention provides a content-based video recommendation device, comprising:

[0207] The first unit is used to obtain the set of videos that the first user is interested in;

[0208] The second unit is used to extract target feature vectors from the set of videos of interest;

[0209] The third unit is used to input the target feature vector into a content recommendation model based on the Rocchio algorithm to obtain the first recommended video set for the first user.

[0210] The fourth unit is used to obtain the second recommended video set of the second user, and to obtain the target recommended video of the first user based on the first recommended video set and the second recommended video set; the second user is a user who has an association or potential association with the first user.

[0211] Reference Figure 10Optionally, the first unit includes:

[0212] The historical video acquisition unit is used to acquire multiple historical videos that the first user has rated within a set time range and whose viewing time has reached a set threshold.

[0213] The content feature acquisition unit is used to extract content features from multiple historical videos and acquire multiple videos based on the content features as the set of videos of interest.

[0214] Optionally, the content feature acquisition unit includes:

[0215] The keywords or phrases in the historical videos, the classification of the historical videos, the ratings and comments of the first user on the historical videos, the keywords in the comments, and whether the historical videos were rated within a set time range after viewing are extracted as the content features.

[0216] Reference Figure 11 Optionally, the second unit includes:

[0217] The user feature vector acquisition unit is used to acquire user feature vectors from the set of videos of interest;

[0218] The target feature vector acquisition unit is used to input the user feature vector into a pre-built regression model to obtain the target feature vector;

[0219] The expression for the regression model is:

[0220] ||Xθ-y|| 2 +||Γθ|| 2

[0221] Where X represents the user feature vector; y represents the target feature vector; || represents regularization; Γ = aI, Γ represents the fitted value;

[0222] The regression model performs overfitting calculations based on the following expression:

[0223] θ(a)=(X T X+aI) -1 X T y

[0224] Where I is the identity matrix, a is the weight of the identity matrix; θ is the fitting hyperparameter; T is the weight constant; θ(a) means finding θ when a is determined.

[0225] Reference Figure 12 Optionally, the user feature vector acquisition unit includes:

[0226] The historical interest vector acquisition unit is used to acquire the first vector corresponding to the videos that the first user is interested in and the second vector corresponding to the videos that are not interested in in the first user's historical video viewing history.

[0227] The current feature vector acquisition unit is used to acquire the current feature vector of the first user from the set of videos of interest;

[0228] The user feature vector determination unit is used to determine the user feature vector based on the first vector, the second vector, and the current feature vector.

[0229] Optionally, the user feature vector determination unit includes:

[0230] The feature vector calculation unit is used to determine the user feature vector according to a first expression, wherein the first expression is:

[0231] U opt =arg max[Sim(U c ,V1)-Sim(U c V h )]

[0232] Among them, U opt U represents the user feature vector. C V represents the current feature vector. l V represents the first vector. h This represents the second vector; max is the function for finding the maximum value.

[0233] Sim(U c V l Sim(U) represents calculating the similarity between the current feature vector and the first vector. c V h The expression () represents calculating the similarity between the current feature vector and the second vector; cosine similarity is used to distinguish between relevant and irrelevant feature vectors in the first expression, resulting in the second expression, which is:

[0234]

[0235] Where || represents the magnitude of the vector. Let ω1 and W2 be the j-th feature vector of the video, and let ω1 and W2 represent two feature vectors in the same video.

[0236] The current feature vector is updated according to the third expression, which is:

[0237]

[0238] Where U0 represents the historical feature vector of the first user, α, β, and γ represent the weights of the historical feature vector, the weight of the first vector, and the weight of the second vector, respectively, and j, v, and ω represent the feature vectors of different videos, respectively. j express One of the feature vectors;

[0239] The current feature vector obtained from the third expression is updated according to the fourth expression, which is:

[0240]

[0241] in, Let ω be the i-th feature vector of the video. l ∈V l Represents ω l The feature vector is a feature vector belonging to the video of interest, ω j ∈V h Represents ω j The feature vectors belong to the feature vectors of the videos that are not of interest.

[0242] Reference Figure 13 Optionally, the fourth unit includes:

[0243] The comprehensive feature vector acquisition unit is used to perform a weighted summation of the feature vectors corresponding to the first recommended video set and the feature vectors corresponding to the second recommended video set to obtain a comprehensive feature vector.

[0244] The target recommended video acquisition unit is used to acquire the corresponding recommended video based on the comprehensive feature vector, and use it as the target recommended video.

[0245] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned content-based video recommendation method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0246] Please see Figure 14 , Figure 14 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0247] The processor 1401 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0248] The memory 1402 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1402 and is called and executed by the processor 1401 to execute a content-based video recommendation method according to an embodiment of this application.

[0249] The input / output interface 1403 is used to implement information input and output;

[0250] The communication interface 1404 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0251] Bus 1405 transmits information between various components of the device (e.g., processor 1401, memory 1402, input / output interface 1403, and communication interface 1404);

[0252] The processor 1401, memory 1402, input / output interface 1403 and communication interface 1404 are connected to each other within the device via bus 1405.

[0253] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned content-based video recommendation method.

[0254] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0255] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform... Figure 1 The method shown.

[0256] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0257] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0258] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0259] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0260] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0261] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0262] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0263] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0264] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A video recommendation method based on content recommendation, characterized in that, The method comprises: obtaining a set of videos of interest of a first user; extracting a target feature vector from the set of videos of interest; inputting the target feature vector into a content recommendation model based on a Rocchio algorithm to obtain a first set of recommended videos of the first user; obtaining a second set of recommended videos of a second user, and obtaining a target recommended video of the first user according to the first set of recommended videos and the second set of recommended videos; the second user is a user having an association relationship or a potential association relationship with the first user; the obtaining of the set of videos of interest of the first user comprises: obtaining a plurality of historical videos of the first user which have a viewing time length reaching a set threshold and are scored within a set time range; extracting content features of the plurality of historical videos, and obtaining a plurality of videos as the set of videos of interest according to the content features; the extracting of the content features of the plurality of historical videos comprises: extracting keywords or phrases in the historical videos, classifications of the historical videos, scores and comments of the first user on the historical videos, keywords in the comments, and whether the historical videos are scored within a set time range after viewing, as the content features; the extracting of the target feature vector from the set of videos of interest comprises: obtaining a user feature vector from the set of videos of interest; inputting the user feature vector into a pre-constructed regression model to obtain a target characteristic vector; the expression of the regression model is: wherein X represents the user feature vector; y represents the target characteristic vector; || represents a norm; Γ=aI, Γ represents a fitting value; θ is a fitting hyperparameter; I is a unit matrix, and a is a weight of the unit matrix; the regression model performs an anti-overfitting operation according to the following expression: where T denotes matrix transposition; represents a determination of a in the case of θ.

2. The video recommendation method based on content recommendation according to claim 1, wherein, the obtaining of the user feature vector from the set of videos of interest comprises: obtaining a first vector corresponding to a video of interest and a second vector corresponding to a video of no interest in historical viewing videos of the first user; obtaining a current feature vector of the first user from the set of videos of interest; determining the user feature vector according to the first vector, the second vector, and the current feature vector. 3.The video recommendation method based on content recommendation of claim 2, wherein, the determining of the user feature vector according to the first vector, the second vector, and the current feature vector comprises: determining the user feature vector according to a first expression, the first expression being: wherein U opt represents the user feature vector, U C represents the current feature vector, V l represents the first vector, V h represents the second vector; max is a maximum function; Sim(U c ,V l ) represents the similarity of the current feature vector and the first vector, Sim(U c ,V h ) represents the similarity of the current feature vector and the second vector; the relevant feature vectors and the irrelevant feature vectors in the first expression are distinguished by cosine similarity, and a second expression is obtained, and the second expression is: where || denotes the norm of a vector, is the first feature vector of the video, j ω1, ω2 represent two feature vectors in the same video. updating the current feature vector according to a third expression, the third expression being: wherein U0 represents the historical feature vector of the first user, a, β, γ respectively represent the weight of the historical feature vector, the weight of the first vector and the weight of the second vector, j, v , ω respectively represent the feature vectors of different videos, ω j represents the feature vector of the first user. updating the current feature vector obtained by the third expression according to a fourth expression, the fourth expression being: wherein, is a first feature vector of the video, i is a second feature vector of the video, l denotes that the feature vector ω l is a feature vector of the video of interest, l denotes that the feature vector ω j is a feature vector of the video of interest, h denotes that the feature vector ω j is a feature vector of the video of non-interest.

4. The video recommendation method based on content recommendation according to claim 1, wherein, the obtaining of the target recommended video of the first user according to the first set of recommended videos and the second set of recommended videos comprises: performing weighted summation on feature vectors corresponding to the first set of recommended videos and feature vectors corresponding to the second set of recommended videos to obtain a comprehensive feature vector; obtaining a corresponding recommended video as the target recommended video according to the comprehensive feature vector.

5. A video recommendation apparatus based on content recommendation, characterized by, The device is applied to the video recommendation method based on content recommendation in claim 1. The first unit is configured to acquire a set of interested videos of a first user; The second unit is configured to extract a target feature vector from the set of interested videos; The third unit is configured to input the target feature vector into a content recommendation model based on a Rocchio algorithm to obtain a first set of recommended videos of the first user; The fourth unit is configured to acquire a second set of recommended videos of a second user, and to acquire a target recommended video of the first user according to the first set of recommended videos and the second set of recommended videos; the second user is a user having an associated relationship or a potential associated relationship with the first user.

6. An electronic device, comprising: The device comprises a processor and a memory; The memory is configured to store a program; The processor executes the program to implement the video recommendation method based on content recommendation according to any one of claims 1 to 4.

7. A computer readable storage medium characterized in that, The storage medium stores a program, and the program is executed by the processor to implement the video recommendation method based on content recommendation according to any one of claims 1 to 4.

Images