Recommendation method and device, electronic equipment and computer readable storage medium

Abstract

The embodiment of the invention discloses a recommendation method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of recommendation. The method comprises the following steps: acquiring a behavior sequence of a user, and determining a media asset preference vector and an operation preference vector of the user according to the behavior sequence; according to the behavior sequence and the media asset preference vector, predicting recommended media assets, and determining a display page of the recommended media assets; determining a current page; according to the operation preference vector, the current page and the display page, determining a shortest preference path from the current page to the display page; and recommending the shortest preference path. According to the scheme, the user can be helped to quickly obtain the recommended media assets, so that the recommendation effect is improved.

Description

Technical Field

[0001] This application relates to the field of recommendation technology, specifically to a recommendation method, apparatus, electronic device, and computer-readable storage medium. Background Technology

[0002] In internet products such as content platforms and e-commerce platforms, the display page for recommended media assets (such as videos, audio, and news) is a core module for achieving user conversion and retention. The rationality of its access path directly determines the recommendation effect. In the product architecture of related technologies, users often encounter the problem of a lengthy redirection path from the current page to the display page of recommended content. They need to go through multiple intermediate pages such as the homepage, category page, and search results page to reach the target page. Each additional redirection significantly increases the user abandonment rate, resulting in low reach and conversion efficiency of recommended media assets. Summary of the Invention

[0003] This application provides a recommendation method, apparatus, electronic device, and computer-readable storage medium, which can help users quickly obtain recommendation media resources, thereby improving the recommendation effect.

[0004] In a first aspect, embodiments of this application provide a recommended method, including: Obtain the user's behavior sequence, and determine the user's media asset preference vector and operation preference vector based on the behavior sequence; Based on the behavior sequence and the media asset preference vector, predict recommended media assets and determine the display page for the recommended media assets; Confirm the current page; Based on the operation preference vector, the current page, and the displayed page, determine the shortest preference path from the current page to the displayed page; The recommended shortest preferred path is described above.

[0005] Secondly, embodiments of this application provide a recommended apparatus, including: The vector determination module is used to acquire the user's behavior sequence and determine the user's media asset preference vector and operation preference vector based on the behavior sequence. The media asset prediction module is used to predict recommended media assets based on the behavior sequence and the media asset preference vector, and determine the display page of the recommended media assets. The page determination module is used to determine the current page; The path determination module is used to determine the shortest preference path from the current page to the display page based on the operation preference vector, the current page, and the display page; The path recommendation module is used to recommend the shortest preferred path.

[0006] Thirdly, embodiments of this application also provide an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps in the recommended method described above.

[0007] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the recommended method described above.

[0008] Fifthly, embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described in embodiments of this application.

[0009] The embodiments of this application have the following beneficial effects: It can acquire user behavior sequences, thereby determining the user's media asset preference vector and operation preference vector based on the behavior sequences. It can predict recommended media assets based on these behavior sequences and media asset preference vectors. The behavior sequences include real-time behavior sequences, which can predict the user's next intention. The media asset preference vectors reflect the user's long-term interest in media assets. Therefore, the recommended media assets predicted based on behavior sequences and media asset preference vectors have dynamic adaptability and can accurately match the user's real needs. The shortest preference path from the current page to the recommended media asset display page is determined based on the operation preference vector. This ensures that the shortest preference path is the user's preferred operation path, increasing the probability that the user will operate according to the shortest preference path. This helps users quickly reach the recommended media asset display page based on the shortest preference path, thus improving the recommendation effect. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of the steps of a recommended method provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the shortest preferred path provided in one embodiment of this application; Figure 3This is a schematic diagram of a television page provided in an embodiment of this application; Figure 4 This is an interactive schematic diagram of a recommended method provided in an embodiment of this application; Figure 5 This is a flowchart illustrating a recommended method provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a recommended device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0012] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0013] In one embodiment, such as Figure 1 As shown, a recommended method is provided. Although the logical order is illustrated in the step diagram, in some cases, the steps shown or described can be performed in a different order than that shown in the diagram. Specifically, this exception handling method can be applied to electronic devices, which may include terminals or servers. The terminal may include, but is not limited to, one or more of smart TVs, smartphones, tablets, laptops, and desktop computers. The server may be a physical server or a cloud server providing various cloud services. It is worth noting that this application does not limit the number of terminals or servers. Depending on the implementation needs, there can be any number of terminals or servers. For example, the server may be a single server or a server cluster composed of multiple servers, etc.

[0014] In one embodiment, the recommendation method is executed by a terminal. The terminal can directly collect user behavior sequences, determine recommendation materials and the shortest preference path, and display the recommendation materials on the display page. In another embodiment, the recommendation method is executed by a server. The server can receive user behavior sequences uploaded by the terminal, determine recommendation materials and the shortest preference path, and then send the recommendation materials and the shortest preference path to the terminal so that the terminal can display the shortest preference path and the recommendation materials on the display page.

[0015] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the priority of the embodiments.

[0016] according to Figure 1 The recommended method shown includes at least steps S110 to S150, which are described in detail below: In step S110, the user's behavior sequence is obtained, and the user's media asset preference vector and operation preference vector are determined based on the behavior sequence.

[0017] A user's behavior sequence is a collection of continuous or discrete operations, interactions, or decision-making actions that reflect the user's behavioral trajectory and operational logic. User behavior sequences are time-series-based datasets of user actions that can be used to analyze user preferences, behavioral patterns, and user intent.

[0018] It is possible to obtain user behavior sequences, which can include real-time and non-real-time behavior sequences. Real-time behavior sequences refer to sequences where the time difference between the execution time and the current time is no greater than a threshold, while non-real-time behavior sequences refer to sequences where the time difference is greater than the threshold. For example, if the threshold is 5 minutes, then a real-time behavior sequence could be a sequence of actions performed by the user within the past 5 minutes, and a non-real-time behavior sequence could be a sequence of actions performed more than 5 minutes ago. Real-time behavior sequences can be used to determine a user's real-time intent; for example, if a user is currently searching for the name of a comedic movie star, their real-time intent might be that they want to watch a movie starring that star. Non-real-time behavior sequences can be used to determine a user's long-term preferences; for example, if a user has watched comedy movies multiple times in the past, their long-term preference might be that they enjoy watching comedy movies.

[0019] A user's behavior sequence can consist of multiple user actions, which may include, but are not limited to: a timestamp, page identifier, element identifier, action type, duration, and / or next page identifier. The timestamp represents the execution time of the action; the page identifier is the identifier of the page where the action is performed (e.g., if a user clicks a tab on the homepage, the page identifier is the homepage identifier); the element identifier is the identifier of the element corresponding to the action (e.g., if a user clicks the movie tab element on the homepage, the element identifier is the movie tab element identifier); the action type may include, but is not limited to: clicking, browsing, saving, playing, swiping, and / or going back; the duration may include the duration of the action; and the next page identifier is the identifier of the page to which the action leads (e.g., after clicking a movie name, the user is redirected to the movie's playback page, and the next page identifier is the identifier of the movie's playback page).

[0020] In one embodiment, the timestamp, page identifier, element identifier, action type, duration, and / or next page identifier corresponding to each operation can be obtained; and a user behavior sequence can be generated based on each of the operation behaviors.

[0021] First, obtain the timestamp, page ID, element ID, action type, duration, and / or next page ID for each operation. Then, convert each operation into a unified format easily understood by computers: (timestamp, page_id, element_id, action_type, duration, next_page_id), where timestamp represents the timestamp, page_id represents the page ID, element_id represents the element ID, action_type represents the action type, duration represents the duration, and next_page_id represents the next page ID. For example, an operation could be (1620000000, "home", "search_btn", "click", 0.5, "search_page"), representing that at the time corresponding to "1620000000" (Beijing time: May 3, 2021, 21:20:00), the search button was clicked on the homepage, the click lasted for 0.5 seconds, and the user was redirected to the search page after the click.

[0022] Based on the timestamps corresponding to each operation, the operations can be integrated in chronological order to generate multiple user behavior sequences. This can be achieved by using timestamps as the core index dimension, and then systematically associating information such as page identifiers, element identifiers, action types, durations, and / or next page identifiers from a single log entry to fully reconstruct the user's operation trajectory. After being sorted by timestamps, the behavior data of different users can form unique, personalized behavior sequences. The operation records of the same user at different times can also be divided into multiple independent behavior sequences based on timestamps, facilitating subsequent analysis of operation behavior patterns and optimization of interaction paths.

[0023] Sequence embedding can be used to transform action sequences into vectors. The Word2Vec model can be used to process user action sequences. Specifically, the core class `Word2Vec`, which implements the Word2Vec algorithm, can be imported from the `models` module of the `gensim` library using the import statement `from gensim.models import Word2Vec`. This class can then be used to train word vectors (or action vectors). Each action is treated as a "word," such as `home->tab>click` representing "the user clicked on an object from the homepage." A list-based action sequence dataset can be used to store action sequences from multiple users. Each element in the dataset corresponds to one or more complete action trajectories for a user. Each complete action trajectory includes multiple strings, each representing a specific action. For example, the string `home->tab_id1->location>block>click` represents clicking on the tab bar with ID 1 located within the page's block area on the homepage.

[0024] Based on the behavior sequence dataset, a behavior sequence embedding model (essentially a Word2Vec model) can be trained: the Word2Vec algorithm is called, the constructed behavior sequence dataset is used as the input of the behavior sequence embedding model, the final vector dimension of each behavior sequence is set to 128 dimensions (vector_size=128), when analyzing the context of each operation behavior, the five consecutive operation behaviors before and after it are referenced (window=5), all the operations that have appeared are retained, and four threads are enabled to accelerate the model training, and finally the trained behavior sequence embedding model is obtained.

[0025] During training, the behavior sequence embedding model treats each action string as a "word" in the text and the user's behavior sequence as a "sentence". The behavior sequence embedding model learns the contextual relationship between different "words" in the "sentence" to generate a unique vector for each action. The more similar the context of the actions, the closer the vector values ​​are.

[0026] After the behavior sequence embedding model is trained, vector representations of specified operations can be extracted from it. The `wv` attribute of the behavior sequence embedding model (which encapsulates vector information for all operations) can be used to query the vectors corresponding to each operation. The queried 128-dimensional dense real vector is used as the vector representation of that operation. This vector representation is a feature abstraction of the operation and can reflect the correlation between the user and other operations. For example, it can show which subsequent operations are often associated with the operation "clicking the homepage tab_id1 block," thus enabling subsequent analysis and recommendations of user behavior.

[0027] In this way, user behavior sequences can be generated based on user actions, and then user actions can be encoded into standardized behavior "words" and user behavior sequences can be encoded into standardized behavior "sentences," thereby accurately capturing the correlation between user actions so as to identify user intent based on user behavior sequences in the future.

[0028] Media asset preference vectors reflect the types and characteristics of media assets that users are interested in; for example, users may prefer comedy movies. Operation preference vectors reflect user preferences, habitual operational behaviors, and operation paths; for example, users may prefer clicking rather than swiping, or they may prefer returning to the homepage before entering the expected page. Based on user behavior sequences, user media asset preference vectors and operation preference vectors can be analyzed. Specifically, the vectors of the behavior sequences can be input into a media asset preference vector prediction model. This model can learn information related to media assets within the behavior sequence to determine the user's media asset preference vector, which directly reflects the user's interest in certain types of goods, content, news, etc. Alternatively, the vectors of the behavior sequences can be input into an operation preference vector prediction model. This model can learn features related to actions and paths within the behavior sequence to determine the user's operation preference vector, which characterizes the user's interaction habits, such as whether they prefer clicking or swiping, and what interaction path they prefer to reach the expected page. Among them, the media asset preference vector and the operation preference vector can both be generated based on standardized behavior coding, with unified dimensions and clear features. They can be directly used for downstream tasks such as user profile construction and personalized recommendation, effectively improving the accuracy of analysis and decision-making.

[0029] In step S120, based on the behavior sequence and the media asset preference vector, recommended media assets are predicted, and the display page for the recommended media assets is determined.

[0030] User behavior sequences include real-time behavior sequences, which can be used to predict the user's next intention. Media asset preference vectors reflect the user's long-term interest in media assets. Based on the user's real-time behavior sequences and media asset preference vectors, recommended media assets can be predicted. Recommended media assets are those that the user is currently interested in.

[0031] The display page for recommended media assets can be determined. For example, if the recommended media asset is a movie, it needs to be played on the movie playback page; if the recommended media asset is a game live stream, it needs to be played on the live stream page.

[0032] In step S130, the current page is determined.

[0033] The current page is the page currently displayed on the terminal. The page identifier of the current page can be obtained.

[0034] In step S140, the shortest preference path from the current page to the display page is determined based on the operation preference vector, the current page, and the display page.

[0035] There are usually multiple interaction paths from the current page to the recommended media asset display page. For example, if the current page is a movie page and the recommended media asset display page is a live streaming page, one interaction path could be to first return to the homepage from the movie page, then go to the movie and live streaming selection page from the homepage, and then enter the live streaming page from that selection page; another interaction path could be to go back to the movie and live streaming selection page from the movie page, and then enter the live streaming page from that selection page.

[0036] The shortest preference path can be selected from multiple interaction paths based on the user's operation preference vector. The shortest preference path is the shortest interaction path composed of user-preferred operation behaviors and interaction paths. The difference between the shortest preference path and the shortest path is that the shortest path simply minimizes the cost of the operation (e.g., number of steps, duration), while the shortest preference path considers the user's operation preference vector, including operation behaviors and interaction paths that satisfy the user's operation preferences. Using the previous example, the first interaction path from the movie page → homepage → selection page → live stream page requires three steps, while the second interaction path only requires two steps. However, if the user prefers to return to the movie page each time and is unaware that they can directly return to the selection page from the movie page, the shortest preference path might be the first interaction path instead of the second.

[0037] Specifically, different weights can be set for each operation based on the operation preference vector. Based on these weights, the operation cost from each page to the next page can be determined, thereby determining the interaction path with the minimum weighted operation cost from the current page to the display page, and identifying this interaction path as the shortest preference path.

[0038] In step S150, the shortest preferred path is recommended.

[0039] Optionally, the terminal can directly highlight the recommended shortest preferred path on the current page. Using the previous example, if the first interaction path is the shortest preferred path, the first interaction path of "movie page → homepage → selection page → live page" can be directly displayed on the current movie page.

[0040] Optionally, Figure 2 This is a schematic diagram illustrating the shortest preferred path provided in one embodiment of this application; as shown... Figure 2 As shown, the terminal can also highlight the corresponding operation behavior on the current page within the recommended shortest preferred path on each page. Continuing with the previous example, if the first interaction path is the shortest preferred path, then "Return to Homepage" can be displayed on the movie page, "Click XX to enter the selection page" can be displayed on the homepage, and "Select Live Stream Page" can be displayed on the selection page.

[0041] By displaying the recommended shortest preference path, users can be guided from the current page to the recommended media asset display page quickly, so that the recommended media assets can be displayed on the display page, thereby achieving effective recommendation of media assets and improving recommendation results.

[0042] The technical solution of this application embodiment can obtain the user's behavior sequence, thereby determining the user's media asset preference vector and operation preference vector based on the behavior sequence. Recommended media assets can be predicted based on the behavior sequence and media asset preference vector. The behavior sequence includes real-time behavior sequences, which can predict the user's next intention. The media asset preference vector reflects the user's long-term interest in media assets. Therefore, the recommended media assets predicted based on the behavior sequence and media asset preference vector have dynamic adaptability and can accurately match the user's real needs. The shortest preference path from the current page to the recommended media asset display page is determined based on the operation preference vector. Therefore, it can be guaranteed that the shortest preference path is the user's preferred operation path, thereby increasing the probability that the user will operate according to the shortest preference path. This helps the user quickly reach the recommended media asset display page based on the shortest preference path, so that the recommended media assets can be displayed on the display page, improving the recommendation effect.

[0043] Based on the above technical solution, as an embodiment, determining the shortest preference path from the current page to the display page according to the operation preference vector, the current page, and the display page may include: obtaining graph structure data of multiple pages; the nodes of the graph structure data are the page states of each page, the page state represents the page, the tabs within the page, and the row and column coordinates within the tabs, and the edges between the nodes represent the operation behaviors and operation costs between the page states; determining the weight of each edge according to the operation preference vector; determining the weighted cost of multiple paths from the current page to the display page according to the nodes corresponding to the current page and the nodes corresponding to the display page, as well as the weight of each edge; determining the shortest preference path according to the weighted cost of the multiple paths, and determining the operation behavior sequence corresponding to the shortest preference path.

[0044] In one embodiment, the terminal can be a smart TV. Figure 3 This is a schematic diagram of a television page provided in an embodiment of this application; see reference. Figure 3 The diagram shows a schematic of the TV homepage, which includes four tabs: Movies, Variety Shows, Live Broadcasts, and Settings. The Variety Shows tab includes eight page states. The row and column coordinates corresponding to the page state "Variety Show 2" are (1,2), the row and column coordinates corresponding to the page state "Variety Show 3" are (2,3), and the row and column coordinates corresponding to the other page states can be deduced in the same way.

[0045] Based on the page state of each page, we can model the nodes corresponding to each page state, and construct edges between nodes based on the operation behavior and cost of jumping from one page state to another, thus modeling the terminal's pages as graph structure data. Edges also exist between different nodes on the same page; using the previous example, there is also an edge between the nodes "Variety Show 2" and "Variety Show 3". The focus of the edges is on the operation behavior and cost, rather than how to jump from one page to another. The edge corresponds to the operation behavior between two interface states, as well as the specific method and cost of that operation.

[0046] A node can be represented using (page_id, tab_id, row, column), where page_id represents the page the node belongs to, tab_id represents the tab ID of the tab containing the node, and row and column represent the row and column coordinates of the node within that tab, respectively. For example... Figure 3 The page state "Variety 2" can be represented as (00,02,1,2). (00,02,1,2) represents the page state of Variety 2 in the first row and second column of the tab page (Variety tab page) corresponding to number 02 in the page numbered 00 (i.e., the homepage).

[0047] A node is the absolute coordinate of a page's state within the page, describing the current location, not the path from the current page to the next. Using the previous example, the node "Variety Show 2" describes its specific location on the homepage, not how to navigate to the page corresponding to Variety Show 2 from the homepage. Because a page can include multiple page states, it can also include multiple nodes; for example, the homepage can include nodes like "Variety Show 1," "Variety Show 2," "Variety Show 2," and so on. By defining multiple nodes, the machine can accurately locate the operation position, thereby planning the shortest preferred path from the current page to the displayed page.

[0048] An edge can be represented using (from_node, to_node, action, cost), where from_node represents the starting node, to_node represents the next node, action represents the operation, and cost represents the operation cost. The operation cost includes the basic operation time and user operation habit penalty. For example, the basic operation time for the "click" action is 1, and the basic operation time for "focusing" (the element gains focus) is 2. If users rarely use the back button, the operation cost can be increased by 1 when performing the "back" operation. Continuing with the previous example, the edge corresponding to node "Variety 2" to node "Variety 3" can be (Variety 2, Variety 3, click Variety 3, 1), representing that the operation of clicking Variety 3 is required to get from node "Variety 2" to node "Variety 3", and the corresponding operation cost is 1. The edge corresponding to node "Variety" to node "Movie" can be (Variety, Movie, switch tabs, 2), representing that the operation of switching tabs is required to get from node "Variety" to node "Movie", and the corresponding operation cost is 2.

[0049] The weights of each edge can be determined based on the user's operation preference vector. The more preferred, familiar, and dependent the user is on the operation behavior and interaction path, the lower the weight of the corresponding edge.

[0050] In one embodiment, the nodes corresponding to the current page and the display page can be determined in the graph structure data. A path search is then performed on the graph structure data to find all paths from the current page node to the display page node. The multiple edges contained in each path are identified, along with their weights and operation costs, to determine the weighted cost of each edge. The weighted costs of the multiple edges in each path are summed to obtain the weighted cost for each path. The path with the lowest weighted cost is then selected as the shortest preferred path. The operation behaviors corresponding to each edge of this shortest preferred path are then grouped into an operation behavior sequence. Executing this operation behavior sequence achieves the shortest preferred path, thereby reaching the display page.

[0051] In another embodiment, to reduce computational load, a heuristic shortest path search algorithm (A*Star algorithm) can be used. This algorithm combines consumed cost and estimated cost to prioritize searching for the path with the lowest total cost, thus obtaining the shortest preferred path. Here, the total estimated cost = consumed cost + estimated cost. The total estimated cost is the estimated cost from the starting node (the node corresponding to the current page) to the target node (the node corresponding to the displayed page). The consumed cost is the actual cost incurred from the starting node to the current node, and the estimated cost is the estimated cost required to reach the target node from the current node. Determining the shortest preferred path based on the heuristic shortest path search algorithm may include: calculating the estimated cost; combining the user's operation preference vector, applying a penalty weight when the current node is a node the user dislikes, and not applying a penalty weight when the current node is a node the user likes. This penalty logic increases the estimated cost of nodes the user dislikes, thus causing the algorithm to prioritize avoiding nodes the user dislikes and prioritizing nodes the user likes.

[0052] By incorporating user operation preference vectors into the estimated cost, the algorithm adapts to user operating habits while retaining the core of the shortest path. For example, users rarely use the "Settings page," so even though "Home → Settings → Target Page" is the shortest path in terms of steps, it's not as good as "Home → My → Target Page" (one more step, but more familiar to the user). By penalizing the estimated cost of "unfamiliar pages," the algorithm prioritizes the shortest preferred path that users are familiar with and find easier to operate, rather than simply the shortest path with the fewest steps.

[0053] For example, the possible paths include: Path 1 is Homepage → Settings Page → Order Page (the operation cost is 2, but the Settings Page is unfamiliar, so the weight is 2, and the estimated cost = 2 × 2 = 4); Path 2 is Homepage → My Page → Order Page (the operation cost is 3, because My Page is familiar, no penalty weight needs to be applied, so the estimated cost = 3); since the total estimated cost of Path 1 = consumed cost 2 + estimated cost 4 = 6, and the total cost of Path 2 = consumed cost 2 + cost 3 = 5, Path 2 can be chosen, and Path 2 is more in line with user habits.

[0054] A heuristic shortest path search algorithm may specifically include: initializing a queue of nodes to be explored, with the first node as the starting node, whose consumed cost is 0 (no operations have been performed yet); determining the estimated remaining cost from the starting node to the target node (e.g., "the estimated distance from the homepage to the order page is 2 steps; if the homepage is a familiar page, h=2; if not, h=3"); automatically placing the node with the smallest estimated remaining cost at the head of the priority queue, thus ensuring that the node with the "lowest overall cost" is processed first each time; and setting an empty dictionary for later path backtracking, based on which, for example, finding the "order page" is done by clicking "order" from the "my page". If the value obtained from the "button" is [order page] = (My page, clicked order), then [order page] = (My page, clicked order) can be stored in an empty dictionary. The node with the lowest estimated cost is popped from the priority queue as the current node for exploration. All neighboring nodes (page states that can be operated on in the next step) of this current node are explored. The estimated cost of each neighboring node is calculated, and the neighboring node with the lowest estimated cost is added to the head of the priority queue. This process continues until the current node is the target node. If so, the shortest preferred path has been found. The empty dictionary is then queried back from the target node to piece together the complete shortest preferred path.

[0055] Optionally, the user's operation preference vector may include interaction paths where the user's historical operation failure rate is greater than the failure rate threshold. When determining the shortest preference path, interaction paths where the user's historical operation failure rate is greater than the failure rate threshold can be avoided.

[0056] The technical solution adopted in this application does not require blindly traversing all nodes. Instead, it prioritizes processing the node with the lowest total estimated cost. The node with the lowest total estimated cost satisfies the requirement of a short path and conforms to user operation preferences, thereby efficiently determining the shortest preferred path. Based on the shortest preferred path, it helps users quickly reach the display page of recommended media assets so that the recommended media assets can be displayed on the display page, thus improving the recommendation effect.

[0057] Based on the above technical solution, as an embodiment, the operation preference vector may include: familiarity and search dependency; familiarity and preference are directly proportional, and search dependency can characterize whether the user prefers to perform a search. Determining the user's operation preference vector based on the behavior sequence may include determining the operation efficiency of the behavior sequence based on the total time and number of operations corresponding to the user's behavior sequence; determining the user's familiarity with the behavior sequence based on the operation efficiency of the behavior sequence; and determining the search dependency based on the number of search operations and the total number of operations corresponding to the behavior sequence.

[0058] Specifically, the average operation time of each operation within a user's behavior sequence can be determined based on the total time and number of operations. Average operation time = total time ÷ number of operations. The operation efficiency of the behavior sequence is then determined based on the average operation time; however, average operation time and operation efficiency are inversely proportional—the longer the average operation time, the lower the operation efficiency, and vice versa. Finally, the user's familiarity with the behavior sequence can be determined based on the operation efficiency; operation efficiency and familiarity are directly proportional—the higher the operation efficiency, the greater the familiarity.

[0059] The total time taken from entering the initial page to achieving the goal can be defined as the total time corresponding to the user's behavior sequence, and the number of effective operations such as clicks, jumps, and inputs included in the behavior sequence can be determined. Operational efficiency is determined by the average operation time. For example, the behavior sequence of "watching a movie" takes 60 seconds with 5 operations, averaging 12 seconds; while the behavior sequence of "watching a live stream" takes 90 seconds with 5 operations, averaging 18 seconds. It can be directly concluded that the behavior sequence of "watching a movie" is more efficient. Based on the user's familiarity with the behavior sequence, the user's preference for the corresponding interaction path can be determined. For example, if a user arrives at the second page from the first page based on two behavior sequences, and the user is more familiar with the first behavior sequence than the second, it can be determined that the user prefers the interaction path corresponding to the first behavior sequence to the second behavior sequence.

[0060] Search dependency calculation focuses on the proportion of search operations within a behavioral sequence. It counts the number of times a user actively triggers the search function in each behavioral sequence, with search dependency calculated as: Search Dependency = Number of Search Operations ÷ Total Number of Operations. For example, if a user completes the "Find Products" sequence 8 times, including 3 search operations, the dependency is 37.5%; while completing the "Browse Recommended Products" sequence involves no search operations, resulting in a 0% dependency. This clearly reflects the difference in user dependence on the search function across different task scenarios, allowing for the selection of which paths containing search behavior should be prioritized when determining the shortest preferred path.

[0061] By employing the technical solutions of this application embodiment, it is possible to determine the user's familiarity with different interaction paths, thereby reflecting the user's preference for different interaction paths; it is possible to determine the user's search dependence, thereby determining the user's preference for search behavior; based on the user's preference for different interaction paths and search behavior, when determining the shortest preferred path, the user's preferred interaction path can be determined to help the user quickly reach the display page of recommended media assets, so that the recommended media assets can be displayed on the display page, thereby improving the recommendation effect.

[0062] Based on the above technical solution, as an embodiment, determining the user's media asset preference vector according to the behavior sequence may include: determining the number of times the user operates on various types of media assets according to the behavior sequence; determining the user's attention to various types of media assets according to the proportion of the number of times the user operates on various types of media assets to the total number of operations; determining the scarcity of various types of media assets; determining the weight of various types of media assets according to the attention and scarcity of various types of media assets; and determining the user's media asset preference vector according to the weight of various types of media assets.

[0063] The number of times a user interacts with a type of media asset can be considered the number of positive actions the user takes with that type of media asset, such as the number of times they watch it or like it. Positive actions towards media assets indicate user interest in them. For each type of media asset, the user's attention level for that type of media asset is determined by the percentage of their interactions with it relative to the total number of interactions across all media asset types. For example, if a user watched a total of 45 media assets across all types, including 20 suspense dramas and 8 comedies, their attention level for suspense dramas would be approximately 44.4% (20 ÷ 45) and for comedies approximately 17.8% (8 ÷ 45), clearly showing that their attention to suspense dramas was significantly higher than their attention to comedies.

[0064] Scarcity assessment requires combining overall media asset data to measure the market scarcity of various media assets. Scarcity is calculated as follows: Scarcity = 1 ÷ (Number of this type of media asset across the entire platform ÷ Total number of media assets across the entire platform). Alternatively, scarcity can be calculated based on user reach; scarcity equals the reciprocal of the percentage of users who access this type of media asset out of the total user base. The lower the percentage, the higher the scarcity. For example, if only 10% of users on the entire platform are interested in suspense dramas, while 60% are interested in comedies, then suspense dramas are far scarcer than comedies.

[0065] The weights of various media assets can be calculated by multiplying their attention and scarcity; higher attention and scarcity result in greater weights. Using various media asset types as dimensions and their corresponding weights as numerical values, a user's media asset preference vector is constructed. For example, a suspense drama has a weight of 0.444 × a high scarcity coefficient of 0.8 = 0.355, and a comedy film has a weight of 0.178 × a low scarcity coefficient of 0.2 = 0.036, forming a user's media asset preference vector of [suspense drama: 0.355, comedy: 0.036, …], thus accurately reflecting the user's preference for various media assets.

[0066] The technical solution adopted in this application calculates attention based on the actual number of operations, which truly reflects the user's core interests and avoids subjective judgment bias. After incorporating scarcity assessment, it can identify niche but high-value preferences, improve the uniqueness of preference representation, and thus ensure the accuracy and practicality of the user's media asset preference vector, providing accurate data support for the determination of subsequent recommended media assets.

[0067] Based on the above technical solution, as an embodiment, predicting recommended media assets according to the behavior sequence and the media asset preference vector may include: determining reference users whose media asset preference vectors have a first similarity greater than a similarity threshold with the user's media asset preference vector; predicting first media assets and a first score for each of the first media assets based on the first similarity, the reference users' preferred media assets, and the reference users' preference for the preferred media assets; predicting a user intent vector based on the behavior sequence and the media asset preference vector; obtaining the media asset features of each media asset and calculating a second similarity between the user intent vector and the media asset features of each media asset; determining second media assets and a second score for each of the second media assets based on the second similarity corresponding to each of the media assets; obtaining the popularity of each of the first media assets and the popularity of each of the second media assets; and determining the recommended media assets based on the popularity and first score of each of the first media assets, and the popularity and second score of each of the second media assets.

[0068] Based on user behavior sequences and media asset preference vectors, the system predicts and recommends media assets. The core of the system adopts a hybrid model architecture of "collaborative filtering + deep matching". Through multi-dimensional data fusion and score optimization, it achieves accurate media asset recommendations that meet user needs.

[0069] Reference users can be selected from multiple users based on their media asset preference vectors. These reference users are those whose media asset preference vectors match the target user's. Therefore, precise targeting can be achieved by leveraging the user similarity calculation capabilities of the collaborative filtering layer. The target user's media asset preference vector can be incorporated into a user behavior matrix, where rows represent users and columns represent media asset click counts. The Pearson correlation coefficient algorithm is then used to calculate the first similarity between the target user and all other users on the platform. A similarity threshold is set, and users whose first similarity exceeds the threshold are selected as reference users. Reference users share a highly similar media asset preference base with the target user, and their preferred media assets are of significant reference value to the target user.

[0070] The prediction of the first media asset is determined based on the reference user's preferred media assets, the first similarity, and the reference user's degree of preference for the preferred media assets. For each reference user, the top 10 media assets with the highest preference in their historical interactions (e.g., the top 10 media assets by click count) are extracted. This is combined with the reference user's degree of preference for the media asset (quantified comprehensively through clicks, favorites, viewing time, etc.), and then multiplied by the first similarity between the target user and the reference user as a weight. A weighted calculation is then performed to obtain the first score for each candidate media asset. The preferred media asset with the highest score is the first media asset. This fully leverages the advantages of collaborative filtering to uncover the common choices of similar groups.

[0071] The second media asset prediction can achieve precise matching between users and media assets through a dual-tower model in the deep matching layer. Based on the target user's behavioral sequence and media asset preference vector, a trained intent prediction model predicts the user's intent vector. This vector not only includes long-term interests but also incorporates immediate needs reflected in short-term behavior. Simultaneously, multi-dimensional features of each media asset are extracted, including type, tags, popularity, and content keywords, to construct a media asset feature vector. The dual-tower model performs deep encoding on both the user intent vector and the media asset feature vector, calculating their cosine similarity as a second similarity. The second media assets and their corresponding second scores are then ranked according to this second similarity, achieving personalized deep matching between users and media assets.

[0072] The final selection of recommended media assets can combine scores and popularity. The popularity of both the first and second media assets (e.g., click-through rate and conversion rate in the last 7 days) can be obtained. The overall score is calculated as: Matching Score × 0.7 + Popularity Factor × 0.3, where the Matching Score can be either the first or second score. A real-time adjustment mechanism can be introduced; if a user's last three actions include positive actions on a certain type of media asset, the score for that type of media asset can be increased by a certain amount or percentage. The final recommended media assets are determined based on the overall scores of the first and second media assets.

[0073] The technical solution adopted in this application not only mines collective wisdom through collaborative filtering but also captures individual needs through deep learning models, thereby achieving a dual improvement in recommendation accuracy and practicality; the determination of recommended media assets based on overall popularity takes into account both personalized matching and content popularity; thus, the accuracy of recommended media assets can be improved.

[0074] Optionally, useful behavioral features can be extracted from the raw logs, such as browsing depth, active time periods, periodic hot spots, jump paths, content type preferences, and distribution of consecutive clicks or keywords. Optionally, the user's media asset preference vector can be updated using the collected user action behavior data. Optionally, when new users first use the terminal, a default path can be used, switching to personalized recommendations after collecting three actions.

[0075] Figure 4This is an interactive schematic diagram of a recommendation method provided in one embodiment of this application. In this embodiment, the terminal can be a TV, and the server can include a cloud-based AI recommendation system. The TV can collect user behavior sequences and upload them to the cloud-based AI recommendation system. The cloud-based AI recommendation system performs AI intent analysis and personalized recommendations based on user behavior prediction, thereby determining the user's next intention, media asset preference vector, and operation preference vector. Combining the user's next intention and media asset preference vector, it determines recommended media assets and, based on the operation preference vector, determines the shortest preference path. The recommended media assets and the shortest preference path are then sent to the TV. The TV can render a UI to display the shortest preference path, allowing the user to obtain recommended media assets according to the shortest preference path. The TV can periodically collect new behavioral data and upload it to the cloud-based AI recommendation system, enabling the cloud-based AI recommendation system to iterate on the AI ​​model.

[0076] Figure 5 This is a flowchart illustrating a recommendation method provided in one embodiment of this application. In this embodiment, the terminal can be a TV, and the server can include a cloud and an intelligent agent. Users can access the TV page via voice, remote control, or card, and perform actions such as clicking to open or watch content. The TV can generate a user behavior sequence based on the user's actions and upload this sequence to the cloud. The cloud can collect user behavior data and invoke the intelligent agent to predict recommended media assets and the shortest preference path. The cloud can match recommended media asset resources and send the recommended media assets and the shortest preference path to the TV. The TV can perform UI rendering based on the shortest preference path to display the shortest preference path, allowing users to automatically locate the recommended media assets. The TV can periodically collect new behavior data and upload it to the cloud, enabling the cloud to iterate on the intelligent agent.

[0077] To facilitate better implementation of the recommendation method of this application, this application also provides a recommendation apparatus based on the above recommendation method. The meanings of the terms used are the same as in the recommendation method described above, and specific implementation details can be found in the description of the method embodiments.

[0078] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of the recommended device provided in the embodiments of this application, wherein the recommended device includes: Vector determination module 601 is used to acquire a user's behavior sequence and determine the user's media asset preference vector and operation preference vector based on the behavior sequence; The media asset prediction module 602 is used to predict recommended media assets based on the behavior sequence and the media asset preference vector, and determine the display page of the recommended media assets; Page determination module 603 is used to determine the current page; The path determination module 604 is used to determine the shortest preference path from the current page to the display page based on the operation preference vector, the current page, and the display page; The path recommendation module 605 is used to recommend the shortest preferred path.

[0079] In one embodiment, the path determination module 604 is specifically used to perform: Obtain graph structure data for multiple pages; the nodes of the graph structure data are the page states of each page, the page states represent the page, the tabs within the page, and the row and column coordinates within the tabs, and the edges between the nodes represent the operation behaviors and operation costs between the page states; The weights of each edge are determined based on the operation preference vector. Based on the nodes corresponding to the current page and the display page, as well as the weights of each edge, determine the weighted cost of multiple paths from the current page to the display page. Based on the weighted cost of multiple paths, the shortest preferred path is determined, and the sequence of operational behaviors corresponding to the shortest preferred path is determined.

[0080] In one embodiment, the operation preference vector includes: familiarity and search dependency; the vector determination module 601 is specifically used to perform: The operational efficiency of the user's behavior sequence is determined based on the total time and number of operations corresponding to the user's behavior sequence. The user's familiarity with the behavior sequence is determined based on the operational efficiency of the behavior sequence. The search dependency is determined based on the number of search operations and the total number of operations corresponding to the behavior sequence.

[0081] In one embodiment, the vector determination module 601 is specifically used to perform: Based on the behavioral sequence, determine the number of times the user interacts with various media assets; The user's level of attention to each type of media asset is determined by the proportion of the number of times the user operates on each type of media asset to the total number of operations. Determine the scarcity of each type of media asset; The weight of each type of media asset is determined based on its popularity and scarcity. Based on the weights of each type of media asset, the user's media asset preference vector is determined.

[0082] In one embodiment, the media asset prediction module 602 is specifically used to perform: Based on the user's media asset preference vector, determine reference users whose first similarity to the user's media asset preference vector is greater than a similarity threshold; Based on the first similarity, the reference user's preferred media asset, and the reference user's degree of preference for the preferred media asset, predict the first media asset and the first score of each of the first media assets; Based on the behavior sequence and the media asset preference vector, predict the user intent vector; Obtain the media asset features of each media asset, and calculate the second similarity between the user intent vector and the media asset features of each media asset; Based on the second similarity corresponding to each of the media assets, the second media asset and the second score of each of the second media assets are determined; Obtain the popularity of each of the first media assets and the popularity of each of the second media assets; The recommended media assets are determined based on the popularity and first score of each of the first media assets, and the popularity and second score of each of the second media assets.

[0083] In one embodiment, the vector determination module 601 is specifically used to perform: Get the timestamp, page identifier, element identifier, action type, duration, and / or next page identifier for each operation. Based on each of the described operations, a sequence of user behaviors is generated.

[0084] In one embodiment, the path recommendation module 605 is specifically used to perform: The shortest preferred path is displayed on the current page.

[0085] The technical solution of this application embodiment can obtain the user's behavior sequence, thereby determining the user's media asset preference vector and operation preference vector based on the behavior sequence. Recommended media assets can be predicted based on the behavior sequence and media asset preference vector. The behavior sequence includes real-time behavior sequences, which can predict the user's next intention. The media asset preference vector reflects the user's long-term interest in media assets. Therefore, the recommended media assets predicted based on the behavior sequence and media asset preference vector have dynamic adaptability and can accurately match the user's real needs. The shortest preference path from the current page to the recommended media asset display page is determined based on the operation preference vector. Therefore, it can be guaranteed that the shortest preference path is the user's preferred operation path, thereby increasing the probability that the user will operate according to the shortest preference path. This helps the user quickly reach the recommended media asset display page based on the shortest preference path, so that the recommended media assets can be displayed on the display page, improving the recommendation effect.

[0086] Specific limitations regarding the recommended device can be found in the limitations of the recommended method described above, and will not be repeated here. Each module in the aforementioned recommended device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0087] In addition, this application also provides an electronic device, such as Figure 7 As shown, it illustrates the structural diagram of the electronic device involved in this application, specifically: The electronic device may include components such as a processor 701 with one or more processing cores and a memory 702 with one or more computer-readable storage media. Those skilled in the art will understand that... Figure 7 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 701 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 702, and by calling data stored in the memory 702, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. Optionally, the processor 701 may include one or more processing cores; preferably, the processor 701 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 701.

[0088] The memory 702 can be used to store software programs and modules. The processor 701 executes various functional applications and data processing by running the software programs and modules stored in the memory 702. The memory 702 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 702 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 702 may also include a memory controller to provide the processor 701 with access to the memory 702.

[0089] In one embodiment, the electronic device further includes a power supply 703 that supplies power to the various components. Preferably, the power supply 703 can be logically connected to the processor 701 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 703 may also include one or more DC or AC power supplies, recharging systems, power equipment debugging circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0090] In one embodiment, the electronic device may further include an input unit 704, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0091] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 701 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 702 according to the following instructions, and the processor 701 runs the applications stored in the memory 702, thereby implementing the steps in any of the recommended methods provided in the embodiments of this application.

[0092] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0093] In one embodiment, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the methods described in any embodiment of this application.

[0094] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described in any embodiment of this application.

[0095] In some embodiments, a computer program product is also provided, including a computer program or instructions that, when executed by a processor, implement the methods described in any embodiment of this application.

[0096] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0097] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0098] Therefore, this application provides a computer-readable storage medium storing a computer program that can be loaded by a processor to perform the steps of any of the recommended methods provided in this application.

[0099] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0100] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0101] Since the instructions stored in the computer-readable storage medium can execute the steps of any of the recommended methods provided in this application, the beneficial effects that any of the recommended methods provided in this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0102] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device 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 terminal device. 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 terminal device that includes said element.

[0103] The foregoing has provided a detailed description of a recommended method, apparatus, electronic device, and computer-readable storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A recommendation method, characterized in that, include: Obtain the user's behavior sequence, and determine the user's media asset preference vector and operation preference vector based on the behavior sequence; Based on the behavior sequence and the media asset preference vector, predict recommended media assets and determine the display page for the recommended media assets; Confirm the current page; Based on the operation preference vector, the current page, and the displayed page, determine the shortest preference path from the current page to the displayed page; The recommended shortest preferred path is described above.

2. The method according to claim 1, characterized in that, Determining the shortest preference path from the current page to the display page based on the operation preference vector, the current page, and the display page includes: Obtain graph structure data for multiple pages; the nodes of the graph structure data are the page states of each page, the page states represent the page, the tabs within the page, and the row and column coordinates within the tabs, and the edges between the nodes represent the operation behaviors and operation costs between the page states; The weights of each edge are determined based on the operation preference vector. Based on the nodes corresponding to the current page and the display page, as well as the weights of each edge, determine the weighted cost of multiple paths from the current page to the display page. Based on the weighted cost of multiple paths, the shortest preferred path is determined, and the sequence of operational behaviors corresponding to the shortest preferred path is determined.

3. The method according to claim 1, characterized in that, The operational preference vector includes: familiarity and search dependence; Determining the user's operation preference vector based on the behavior sequence includes: The operational efficiency of the user's behavior sequence is determined based on the total time and number of operations corresponding to the user's behavior sequence. The user's familiarity with the behavior sequence is determined based on the operational efficiency of the behavior sequence. The search dependency is determined based on the number of search operations and the total number of operations corresponding to the behavior sequence.

4. The method according to claim 1, characterized in that, Determining the user's media asset preference vector based on the behavior sequence includes: Based on the behavioral sequence, determine the number of times the user interacts with various media assets; The user's level of attention to each type of media asset is determined by the proportion of the number of times the user operates on each type of media asset to the total number of operations. Determine the scarcity of each type of media asset; The weight of each type of media asset is determined based on its popularity and scarcity. Based on the weights of each type of media asset, the user's media asset preference vector is determined.

5. The method according to claim 1, characterized in that, The step of predicting and recommending media assets based on the behavior sequence and the media asset preference vector includes: Based on the user's media asset preference vector, determine reference users whose first similarity to the user's media asset preference vector is greater than a similarity threshold; Based on the first similarity, the reference user's preferred media asset, and the reference user's degree of preference for the preferred media asset, predict the first media asset and the first score of each of the first media assets; Based on the behavior sequence and the media asset preference vector, predict the user intent vector; Obtain the media asset features of each media asset, and calculate the second similarity between the user intent vector and the media asset features of each media asset; Based on the second similarity corresponding to each of the media assets, the second media asset and the second score of each of the second media assets are determined; Obtain the popularity of each of the first media assets and the popularity of each of the second media assets; The recommended media assets are determined based on the popularity and first score of each of the first media assets, and the popularity and second score of each of the second media assets.

6. The method according to claim 1, characterized in that, The acquisition of the user's behavior sequence includes: Get the timestamp, page identifier, element identifier, action type, duration, and / or next page identifier for each operation. Based on each of the described operations, a sequence of user behaviors is generated.

7. The method according to claim 1, characterized in that, The recommended shortest preferred path includes: The shortest preferred path is displayed on the current page.

8. A recommended device, characterized in that, include: The vector determination module is used to acquire the user's behavior sequence and determine the user's media asset preference vector and operation preference vector based on the behavior sequence. The media asset prediction module is used to predict recommended media assets based on the behavior sequence and the media asset preference vector, and determine the display page of the recommended media assets. The page determination module is used to determine the current page; The path determination module is used to determine the shortest preference path from the current page to the display page based on the operation preference vector, the current page, and the display page; The path recommendation module is used to recommend the shortest preferred path.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the recommended method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the recommended method as described in any one of claims 1 to 7.

Images