Method for determining target recommendation information, information recommendation method, device and equipment

Abstract

The present application relates to a method for determining target recommendation information, a method, device and equipment for pushing information recommendation. The method relates to artificial intelligence, and includes: obtaining an interactive information set of target multimedia data within a target time window, identifying points of interest within the target time window based on each interactive information, and determining the relevance of each point of interest with the interactive information set. According to the relevance of each point of interest with the interactive information set, target points of interest within the target time window are screened out from each point of interest, and from each target point of interest, directional points of interest that match the portrait data of the target object accessing the target multimedia data are screened out, and the target recommendation information associated with the directional points of interest are pushed to the target object. The use of this method can improve the closeness between the target points of interest and the interactive information set, and when making recommendations based on the target recommendation information of the directional points of interest screened out from the target points of interest, the recommendation accuracy and recommendation hit rate of the target recommendation information can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method for determining target recommendation information, an information recommendation method, an apparatus, and a device. Background Art

[0002] With the development of artificial intelligence technology and the widespread use of various Internet applications and websites, more and more users are obtaining the information or news they need in real time from different applications or websites. Various applications or websites can also recommend information to users during use or access to achieve better interaction and traffic effects.

[0003] Traditionally, when users want to watch online videos or make purchases online, recommendations for similar videos or products are often made based on the user's past visits or purchases. For example, video recommendations can be used to recommend videos with similar titles, tags, or content categories based on a user's viewing history.

[0004] However, traditional information recommendation methods only consider users who have obtained historical records, but fail to fully consider the characteristics of the video or product itself, nor can they follow up in real time on the actual needs and current actions of different users in different scenarios. As a result, the information ultimately predicted and recommended still suffers from low accuracy, large errors, and low hit rate of recommended information. Summary of the Invention

[0005] Based on this, it is necessary to provide a method for determining target recommended information, an information recommendation method, an apparatus and a device that can ensure the accuracy of recommended information and improve the hit rate of recommended information in response to the above technical problems.

[0006] In a first aspect, the present application provides a method for determining target recommendation information. The method comprises:

[0007] Acquire an interaction information set of target multimedia data within a target time window, wherein the interaction information set includes each piece of interaction information;

[0008] Based on each of the interaction information, identifying points of interest within the target time window;

[0009] Determining the relevance of each of the interest points to the interactive information set;

[0010] Filtering target interest points within the target time window from the interest points according to the relevance between the interest points and the interactive information set;

[0011] Target recommendation information is determined based on the target point of interest.

[0012] In one embodiment, determining the relevance of each of the points of interest to the interactive information set includes:

[0013] Determining a first relevance between each of the interest points and each of the interactive information;

[0014] The second relevance between each of the interest points and the set of interactive information is determined based on the first relevance between each of the interest points and each of the interactive information.

[0015] In one embodiment, determining the first relevance between each of the points of interest and each of the interactive information includes:

[0016] Obtaining each first text feature vector corresponding to each point of interest and each second text feature vector corresponding to each interactive information;

[0017] Performing mapping processing on each of the first text feature vectors and each of the second text feature vectors to obtain a corresponding first semantic vector and a second semantic vector;

[0018] Based on each of the first semantic vectors and each of the second semantic vectors, semantic relevance calculations are performed to obtain first relevance between each of the interest points and each of the interactive information.

[0019] In one embodiment, determining the second relevance between each of the interest points and the set of interaction information by respectively determining the first relevance between each of the interest points and each of the interaction information includes:

[0020] performing summation processing on the first correlation between each of the interest points and each of the interactive information to obtain a total correlation between the corresponding interest point and each of the interactive information;

[0021] Obtaining the total number of interactive information in the interactive information set;

[0022] According to the total number of the interactive information, the total correlation is averaged to obtain a second correlation between the corresponding interest point and the interactive information set.

[0023] In one embodiment, identifying points of interest within the target time window based on each piece of interaction information includes:

[0024] Acquire a dictionary tree corresponding to the target multimedia data; the dictionary tree includes each candidate point of interest;

[0025] Performing entity recognition processing on each of the interactive information to obtain a first entity corresponding to each of the interactive information;

[0026] Each of the first entities is matched with each candidate point of interest included in the dictionary tree to obtain each point of interest within the target time window.

[0027] In one embodiment, a method for obtaining the dictionary tree corresponding to the target multimedia data includes:

[0028] Acquire all interactive information and multimedia attribute information corresponding to the target multimedia data, wherein the multimedia attribute information includes title data and classification label data;

[0029] Performing entity recognition on all the interactive information and multimedia attribute information to obtain a corresponding second entity;

[0030] Each of the second entities is determined as a candidate point of interest, and a dictionary tree corresponding to the target multimedia data is constructed based on each of the candidate points of interest.

[0031] In one embodiment, the target recommendation information includes multimedia recommendation data; and determining the target recommendation information based on the target point of interest includes:

[0032] searching for candidate points of interest that match each of the target points of interest in an established point of interest index structure, wherein the point of interest index structure includes an index relationship between the points of interest and multimedia data identifiers;

[0033] Determining multimedia data identifiers corresponding to the candidate points of interest based on index relationships of the points of interest index structure;

[0034] The multimedia recommendation data corresponding to each multimedia data identifier is determined as target recommendation information.

[0035] In one embodiment, the method of establishing the interest point index structure includes:

[0036] Performing entity recognition on multimedia attribute information of each multimedia data in a multimedia database to obtain a third entity of each multimedia data, wherein the multimedia attribute information includes title data and classification label data;

[0037] Matching each of the third entities with each of the candidate interest points in the dictionary tree, determining a target entity in the third entity and a target candidate interest point in the dictionary tree, wherein the target entity matches the target candidate interest point;

[0038] Based on each of the candidate target interest points and the multimedia data identifier corresponding to each of the target entities, a corresponding interest point index structure is established.

[0039] In one embodiment, the target recommendation information includes product recommendation data; and determining the target recommendation information based on the target point of interest includes:

[0040] Performing product recognition processing on each of the target points of interest to generate corresponding product recognition results;

[0041] Based on the product identification result, corresponding product recommendation data is determined, and the product recommendation data is determined as target recommendation information.

[0042] In a second aspect, the present application also provides an information recommendation method. The method comprises:

[0043] Obtaining portrait data of a target object accessing target multimedia data;

[0044] Filtering, from target interest points within a target time window of the target multimedia data, directional interest points that match the portrait data, wherein the target interest points within the target time window are filtered from the interest points based on correlations between the interest points within the target time window of the target multimedia data and the interaction information set; and the interest points within the target time window are obtained by performing interest point recognition on each interactive information in the interaction information set;

[0045] Push target recommendation information associated with the directional interest point to the target object.

[0046] In a third aspect, the present application further provides a device for determining target recommendation information. The device comprises:

[0047] An interactive information set acquisition module, configured to acquire an interactive information set of target multimedia data within a target time window, wherein the interactive information set includes various interactive information;

[0048] an interest point identification module, configured to identify interest points within the target time window based on each of the interaction information;

[0049] A relevance determination module, configured to determine the relevance between each of the interest points and the interactive information set;

[0050] a target interest point screening module, configured to screen target interest points within the target time window from the various interest points according to the relevance between the various interest points and the interactive information set;

[0051] The target recommendation information determination module is used to determine target recommendation information based on the target point of interest.

[0052] In a fourth aspect, the present application further provides an information recommendation device. The device comprises:

[0053] A portrait data acquisition module, used to acquire portrait data of a target object accessing target multimedia data;

[0054] a directional interest point screening module, configured to screen directional interest points that match the portrait data from target interest points within a target time window of the target multimedia data; the target interest points within the target time window are screened from the interest points based on a correlation between each interest point within the target time window of the target multimedia data and the interactive information set; each interest point within the target time window is obtained by identifying an interest point for each interactive information in the interactive information set;

[0055] The target recommendation information pushing module is used to push the target recommendation information associated with the directional interest point to the target object.

[0056] In a fifth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0057] Acquire an interaction information set of target multimedia data within a target time window, wherein the interaction information set includes each piece of interaction information;

[0058] Based on each of the interaction information, identifying points of interest within the target time window;

[0059] Determining the relevance of each of the interest points to the interactive information set;

[0060] Filtering target interest points within the target time window from the interest points according to the relevance between the interest points and the interactive information set;

[0061] Target recommendation information is determined based on the target point of interest.

[0062] In a sixth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0063] Obtaining portrait data of a target object accessing target multimedia data;

[0064] Filtering, from target interest points within a target time window of the target multimedia data, directional interest points that match the portrait data, wherein the target interest points within the target time window are filtered from the interest points based on correlations between the interest points within the target time window of the target multimedia data and the interaction information set; and the interest points within the target time window are obtained by performing interest point recognition on each interactive information in the interaction information set;

[0065] Push target recommendation information associated with the directional interest point to the target object.

[0066] In a seventh aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0067] Acquire an interaction information set of target multimedia data within a target time window, wherein the interaction information set includes each piece of interaction information;

[0068] Based on each of the interaction information, identifying points of interest within the target time window;

[0069] Determining the relevance of each of the interest points to the interactive information set;

[0070] Filtering target interest points within the target time window from the interest points according to the relevance between the interest points and the interactive information set;

[0071] Target recommendation information is determined based on the target point of interest.

[0072] In an eighth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0073] Obtaining portrait data of a target object accessing target multimedia data;

[0074] Filtering, from target interest points within a target time window of the target multimedia data, directional interest points that match the portrait data, wherein the target interest points within the target time window are filtered from the interest points based on correlations between the interest points within the target time window of the target multimedia data and the interaction information set; and the interest points within the target time window are obtained by performing interest point recognition on each interactive information in the interaction information set;

[0075] Push target recommendation information associated with the directional interest point to the target object.

[0076] In a ninth aspect, the present application further provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the following steps:

[0077] Acquire an interaction information set of target multimedia data within a target time window, wherein the interaction information set includes each piece of interaction information;

[0078] Based on each of the interaction information, identifying points of interest within the target time window;

[0079] Determining the relevance of each of the interest points to the interactive information set;

[0080] Filtering target interest points within the target time window from the interest points according to the relevance between the interest points and the interactive information set;

[0081] Target recommendation information is determined based on the target point of interest.

[0082] In a tenth aspect, the present application further provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the following steps:

[0083] Obtaining portrait data of a target object accessing target multimedia data;

[0084] Filtering, from target interest points within a target time window of the target multimedia data, directional interest points that match the portrait data, wherein the target interest points within the target time window are filtered from the interest points based on correlations between the interest points within the target time window of the target multimedia data and the interaction information set; and the interest points within the target time window are obtained by performing interest point recognition on each interactive information in the interaction information set;

[0085] Push target recommendation information associated with the directional interest point to the target object.

[0086] In the aforementioned target recommendation information determination method, information recommendation method, apparatus, computer device, storage medium, and computer program product, by obtaining an interactive information set for target multimedia data within a target time window, the interactive information set comprising each interactive information, real-time identification of points of interest within the target multimedia data within the target time window is achieved, and changes in the points of interest within the target multimedia data within different time windows are tracked. Based on each interactive information, points of interest within the target time window are identified, and the correlation between each point of interest and the interactive information set is determined. Furthermore, based on the correlation between each point of interest and the interactive information set, target points of interest within the target time window are filtered from the points of interest, and target recommendation information is ultimately determined based on the target points of interest. By calculating the correlation between the interactive information set and each point of interest, the characteristics of the target multimedia data itself and the interactive information are comprehensively considered, thereby improving the closeness between the determined target points of interest and the interactive information set, thereby improving the recommendation accuracy and recommendation hit rate of the target recommendation information associated with the target points of interest. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 A diagram illustrating an application environment of a method for determining target recommendation information and an information recommendation method in one embodiment;

[0088] Figure 2 Schematic diagram of a flow chart of a method for determining target recommendation information in one embodiment;

[0089] Figure 3 is a schematic diagram of interactive information of target multimedia data within a target time window in one embodiment;

[0090] Figure 4 FIG. 1 is a flow chart of determining the relevance of each point of interest to an interactive information set in one embodiment;

[0091] Figure 5 is a schematic diagram of a deep semantic matching model for performing relevance calculation in one embodiment;

[0092] Figure 6 A schematic diagram of a process for establishing an interest point index structure in one embodiment;

[0093] Figure 7 is a schematic diagram of an interest point index structure in one embodiment;

[0094] Figure 8 is a flowchart of a method for determining target recommendation information in another embodiment;

[0095] Figure 9 1 is a flow chart of an information recommendation method according to an embodiment;

[0096] Figure 10 is a structural block diagram of a device for determining target recommendation information in one embodiment;

[0097] Figure 11 is a structural block diagram of an information recommendation device in one embodiment;

[0098] Figure 12 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0099] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0100] The methods for determining target recommendation information and recommending information provided in the embodiments of this application involve artificial intelligence technology. Artificial intelligence (AI) refers to theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce new intelligent machines that can react in a manner similar to human intelligence. AI is the study of the design principles and implementation methods of various intelligent machines, enabling them to have the capabilities of perception, reasoning, and decision-making. AI technology is an interdisciplinary discipline covering a wide range of fields, encompassing both hardware and software technologies. Basic AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technologies, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0101] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0102] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, autonomous driving, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0103] The target recommendation information determination method and information recommendation method provided in the embodiments of the present application can be applied to Figure 1 In the application environment shown. The terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The server 104 obtains the interactive information set of the target multimedia data in the target time window, and identifies the points of interest in the target time window based on the interactive information set, so as to further determine the relevance of each point of interest with the interactive information set. The target multimedia data and the interactive information set can be stored in the data storage system, or stored in the local storage of the terminal 102. The server 104 can filter out the target points of interest in the target time window from the various points of interest based on the relevance of each point of interest with the interactive information set, and determine the target recommendation information based on the target points of interest.

[0104] Furthermore, the server 104 obtains the portrait data of the target object accessing the target multimedia data, and filters the directional interest points that match the portrait data from the target interest points within the target time window of the target multimedia data. The target object can access the target multimedia data based on the application or web page loaded on the terminal 102, and the portrait data of the target object can be stored in the data storage system or in the local storage of the terminal 102. The target interest points within the target time window are filtered from the various interest points based on the correlation between the various interest points of the target multimedia data within the target time window and the interactive information set, and the various interest points within the target time window are obtained by identifying the interest points of each interactive information in the interactive information set. Finally, the server 104 can push the target recommendation information associated with the directional interest points to the target object, that is, it can be pushed to the terminal 102 where the target object is located for display. Terminal 102 may include, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, portable wearable devices, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, and aircraft. IoT devices may include smart speakers, smart TVs, smart air conditioners, and smart vehicle-mounted devices. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. Server 104 may be implemented as a standalone server or a server cluster consisting of multiple servers. Embodiments of the present invention are applicable to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0105] In one embodiment, Figure 2 As shown, a method for determining target recommendation information is provided, and the method is applied to Figure 1 The following steps are used as an example to illustrate the server in the example:

[0106] Step S202: Acquire an interaction information set of target multimedia data within a target time window, where the interaction information set includes various pieces of interaction information.

[0107] The target multimedia data can be video data, audio data, or image data, such as video data from video viewing platforms, short video platforms, and live streaming platforms; audio data from audio listening platforms; or image data from image viewing platforms. The interactive information set can include multiple interactive messages in different forms, such as bullet comments, attention messages, and comments. The target time window represents a time window selected for the target multimedia data. That is, the target multimedia data can be divided into multiple time windows. The length of each time window can be adjusted or modified based on actual needs and is not limited to a specific value or values.

[0108] Specifically, for target multimedia data for which target recommendation information is to be determined, it is necessary to obtain multiple pieces of interaction information for the target multimedia data within a target time window to obtain a corresponding interaction information set. The target time window can be any time window selected from multiple time windows divided into the target multimedia data, or it can be the time window currently being played.

[0109] Furthermore, if Figure 3 As shown, a schematic diagram of interactive information of target multimedia data within a target time window is provided. Figure 3 It can be seen that the video title of the target multimedia data is "Subject A sings "XXX Song"". Within the target time window, multiple interactive messages released by each object accessing the target multimedia data can be detected, such as "Subject A's voice is so ethereal", "Subject A's Hanfu is very beautiful", "The combination of Hanfu and headwear when singing each song is very suitable", "Subject B also sang "XXX Song"" and "The scenery of XX city mentioned in "XXX Song" is very nice", etc., and then by uniformly collecting multiple interactive information within the target time window, the interactive information set of the target multimedia data within the target time window is obtained.

[0110] Step S204: identifying points of interest within the target time window based on the interaction information.

[0111] The points of interest represent entities or concepts that the object accessing the target multimedia data is interested in, and may be specific events, people, products, video titles, drama titles, song titles, movie titles, places, etc.

[0112] Specifically, by obtaining a pre-established dictionary tree corresponding to the target multimedia data and obtaining each candidate interest point included in the dictionary tree, interest point matching processing is performed on the first entity corresponding to each interactive information based on each candidate interest point included in the dictionary tree.

[0113] Among them, based on various interactive information, such as the barrage information of the target multimedia data within the target time window, entity recognition is performed, and the entity mention fragments in each barrage information are identified. Further entity mining is performed on the entity mention fragments to identify entities such as events, people, products, video names, drama titles, song names, movie names and places, that is, the first entity corresponding to each interactive information is obtained.

[0114] Furthermore, the first entity corresponding to each interactive information is matched with each candidate interest point included in the dictionary tree to determine whether each first entity can be successfully matched with each candidate interest point included in the dictionary tree. If the match is successful, it indicates that the successfully matched first entity can be used as the interest point of the target multimedia data within the target time window.

[0115] In one embodiment, after identifying the points of interest in each interactive information, each point of interest is further highlighted (such as highlighted, displayed in bold font, and displayed in a striking font color, etc.), and each object accessing the target multimedia data is detected, and the click, search, and other behaviors performed on each point of interest are detected, so as to find related multimedia data or provide corresponding search results based on the click and search behaviors of each object, thereby deepening the interactivity between the object and the multimedia data platform, and at the same time achieving the purpose of attracting traffic, providing more access traffic for the multimedia data platform, etc.

[0116] Step S206: Determine the relevance between each point of interest and the interactive information set.

[0117] Specifically, for each point of interest and the multiple interactive information included in the interactive information set, it is necessary to first determine the first correlation between each point of interest and each interactive information, and then further determine the second correlation between each point of interest and the interactive information set through the first correlation between each point of interest and each interactive information.

[0118] The first correlation represents the degree of association between each point of interest and each piece of interactive information, while the second correlation represents the degree of association between each point of interest and the entire interactive information set. For example, if the target multimedia data has five points of interest within the target time window, and the interactive information set includes 15 pieces of interactive information, then the first correlation between each of the five points of interest and each of the 15 pieces of interactive information needs to be calculated. Based on the first correlation between each point of interest and these 15 pieces of interactive information, the second correlation between each point of interest and the complete interactive information set consisting of 15 pieces of interactive information needs to be calculated.

[0119] Furthermore, the trained deep semantic matching model can be used to calculate the first correlation between each point of interest and each interactive information, and further calculate the second correlation between the point of interest and the entire interactive information set based on the first correlation corresponding to each point of interest.

[0120] Step S208 : Filtering target interest points within the target time window from the interest points based on the relevance between the interest points and the interactive information set.

[0121] Specifically, based on a first preset relevance threshold and a second relevance between each POI and the interactive information set, target POIs within the target time window are screened from each POI. Specifically, POIs whose second relevance is greater than the first preset relevance threshold are screened from each POI and determined as target POIs for the target multimedia data within the target time window. The first preset relevance threshold can be adjusted or modified based on actual needs and is not limited to a specific value or values.

[0122] Step S210: determining target recommendation information based on the target point of interest.

[0123] Specifically, the target recommendation information may include multimedia recommendation data and product recommendation data, that is, for the target interest point of the target multimedia data within the target time window, the multimedia recommendation data associated with the target interest point can be determined, such as recommended video data, recommended audio data, and recommended image data, etc. It can also be product recommendation data associated with the target interest point, such as specific recommended products of certain shopping platforms or jump links to recommended products, etc.

[0124] In the above-described method for determining target recommendation information, by obtaining a set of interaction information for target multimedia data within a target time window, the interaction information set comprising each interaction information, it is possible to instantly identify points of interest within the target multimedia data within the target time window and continuously track changes in the points of interest within the target multimedia data within different time windows. Based on each interaction information, points of interest within the target time window can be identified, and the correlation between each point of interest and the interaction information set can be determined. Furthermore, based on the correlation between each point of interest and the interaction information set, target points of interest within the target time window can be selected from the points of interest, and ultimately, target recommendation information can be determined based on the target points of interest. By calculating the correlation between the interaction information set and each point of interest, the characteristics of the target multimedia data itself and the interaction information can be comprehensively considered, thereby improving the closeness between the determined target points of interest and the interaction information set, thereby improving the recommendation accuracy and recommendation hit rate of the target recommendation information associated with the target points of interest.

[0125] In one embodiment, Figure 4 As shown, the steps of determining the relevance between each point of interest and the interactive information set specifically include:

[0126] Step S402: determining a first relevance between each point of interest and each piece of interactive information.

[0127] Specifically, the first correlation is used to represent the degree of association between each point of interest and each interactive information. This can be done by obtaining each first text feature vector corresponding to each point of interest and each second text feature vector corresponding to each interactive information, and mapping each first text feature vector and each second text feature vector respectively to obtain the corresponding first semantic vector and second semantic vector. Then, based on each first semantic vector and each second semantic vector, semantic correlation calculation is performed respectively to obtain the first correlation between each point of interest and each interactive information.

[0128] In one embodiment, Figure 5As shown, a deep semantic matching model for correlation calculation is provided, which can be specifically adopted as follows Figure 5 The trained deep semantic matching model shown in FIG. 1 realizes the calculation of the first relevance, wherein the deep semantic matching model includes an input layer, a representation layer, a matching layer, and an output layer. Figure 5 It can be seen that the input layer of the deep semantic matching model can be used to perform feature mapping processing on the first text feature vectors corresponding to each point of interest and the second text feature vectors corresponding to each interactive information, and the text feature vectors can be mapped into low-dimensional embedding vectors, that is, the first semantic vector corresponding to the first text feature vector and the second semantic vector corresponding to the second text feature vector can be obtained.

[0129] Specifically, refer to Figure 5 It can be seen that the semantic features corresponding to the first semantic vector and the second semantic vector, including the first semantic feature and the second semantic feature, can be learned and extracted through the representation layer of the deep semantic matching model. Then, the matching layer of the deep semantic matching model can be used to calculate the semantic correlation between the first semantic feature and the second semantic feature to obtain the first correlation between each interest point and each interactive information. Finally, the corresponding first correlation is output through the output layer of the deep semantic matching model.

[0130] The representation layer of the deep semantic matching model can be a fully connected layer, a CNN (convolutional neural network) layer, a Transformer layer, and various RNN (recurrent neural network) layers, and can be a stack of multiple layers. RNN (recurrent neural network) layers can include, for example, LSTM (long short-term memory), Bi-LSTM (bidirectional long short-term memory), and GRU (gated recurrent unit).

[0131] Furthermore, the matching layer of the deep semantic matching model can be a fully connected layer, which specifically calculates the first correlation between each point of interest and each interactive information by using methods such as dot product, cosine similarity, Gaussian distance, and similarity matrix. Among them, after the matching layer of the deep semantic matching model calculates the first correlation between each point of interest and each interactive information, it specifically uses an activation function (such as a Sigmoid activation function) to map the first correlation to the probability space of [0,1] to obtain a specific probability value of the first correlation, which can be different values ​​in the probability space of [0,1], such as 0.2, 0.4, and 0.8. The specific probability value of the first correlation is limited, and different values ​​can be taken according to actual conditions.

[0132] Step S404: determining a second correlation between each interest point and the interactive information set based on the first correlation between each interest point and each interactive information.

[0133] Specifically, by summing up the first correlation between each point of interest and each interactive information, the total correlation between the corresponding point of interest and each interactive information is obtained, and the total number of interactive information in the interactive information set is obtained. Then, based on the total number of interactive information, the total correlation is averaged to obtain the second correlation between the corresponding point of interest and the interactive information set.

[0134] Furthermore, the first correlation between each point of interest X and each interactive information Y in the interactive information set S (where Y∈S) is expressed as Match(X, Y), and the second correlation between each point of interest X and the interactive information set S is expressed as Match(X, S). The second correlation Match(X, S) is calculated using the following formula (1):

[0135]

[0136] Wherein, |S| represents the total number of interactive information in the interactive information set S, Match(X, Y) represents the first correlation between each point of interest and each interactive information, and Match(X, S) represents the second correlation between each point of interest and the interactive information set.

[0137] For example, if the interactive information of the target multimedia data within the target time window includes: "The voice of subject A is so ethereal", "Subject A's Hanfu is very beautiful", "The combination of Hanfu and headdress when singing each song is very suitable", "Subject B has also sung "XXX Song"", and "The scenery of XX city mentioned in "XXX Song" is very nice", etc., then based on multiple candidate interest points in the dictionary tree, each interactive information is identified, and the interest points of the target multimedia data within the target time window are obtained as: "Subject A", "Hanfu", "headdress" and "XXX Song".

[0138] Specifically, for the interest points of the target multimedia data within the target time window, including "object A", "Hanfu", "headwear" and "XXX song", the first correlation Match(X, Y) between each interest point X and each interactive information Y is calculated, and the first correlation Match(X, Y) is summed to obtain the total correlation ∑ between the corresponding interest point X and each interactive information Y Y∈s Match(X, Y), and obtain the total number of interactive information |S| in the interactive information set, and then average the total correlation according to the total number of interactive information |S| to obtain the second correlation Match(X, S) between the corresponding interest point and the interactive information set.

[0139] Furthermore, for example, for each point of interest, the calculated Match("A object", S) = 0.8, Match("Hanfu", S) = 0.4, Match("Headwear", S) = 0.2, Match("XXX song", S) = 0.6, and then according to the first preset correlation threshold and the second correlation between each point of interest and the interactive information set, the target point of interest within the target time window can be filtered out from each point of interest.

[0140] For example, if the first preset relevance threshold is set to 0.5, each point of interest is screened based on the first preset relevance threshold, and points of interest with a second relevance less than 0.5 are eliminated. Points of interest with a second relevance greater than 0.6 are obtained, i.e., point of interest "A object" with Match("Object A", S) = 0.8, and point of interest "XXX song" with Match("XXX song", S) = 0.6, which are determined as target points of interest. The first preset relevance threshold can be adjusted or modified according to actual needs and is not limited to one or more specific values.

[0141] In this embodiment, by determining the first correlation between each point of interest and each interactive information, and by determining the second correlation between each point of interest and the interactive information set through the first correlation between each point of interest and each interactive information, it is achieved that on the basis of calculating the first correlation between the point of interest and the interactive information, the second correlation between each point of interest and the entire interactive information set is calculated to further determine the point of interest with a higher degree of correlation with the entire interactive information set, and eliminate irrelevant points of interest with a low degree of correlation, so that the target point of interest finally determined is more accurate, thereby improving the recommendation accuracy and recommendation hit rate of the target recommendation information when recommending the target recommendation information associated with the target point of interest.

[0142] In one embodiment, the step of obtaining a dictionary tree corresponding to the target multimedia data specifically includes:

[0143] Acquire all interactive information and multimedia attribute information corresponding to the target multimedia data, where the multimedia attribute information includes title data and classification label data; perform entity recognition on all interactive information and multimedia attribute information to obtain corresponding second entities; determine each second entity as a candidate point of interest, and construct a dictionary tree corresponding to the target multimedia data based on each candidate point of interest.

[0144] Specifically, by obtaining all interactive information corresponding to the target multimedia data, as well as title data and classification label data, and based on all interactive information, such as all barrage information of the target multimedia data, as well as title data and classification label data, entity recognition is performed to identify entity mention fragments in each barrage information, and further entity mining is performed on the entity mention fragments to identify events, people, products, video names, drama titles, song names, movie names and places, etc., to obtain the corresponding second entity.

[0145] Furthermore, each second entity obtained by performing entity recognition on all interactive information, as well as title data and classification label data, can be understood as an entity that the object accessing the target multimedia data may be interested in, and then each second entity is determined as a candidate point of interest, and based on each candidate point of interest, a dictionary tree corresponding to the target multimedia data is constructed.

[0146] A trie, also known as a prefix tree or single-lookup tree, is an ordered tree data structure used in information retrieval. Pre-retrieved data (such as objects or entities) is stored in the trie for subsequent search and retrieval. A hash table can also be used to implement the trie's functionality, storing data and providing data lookup.

[0147] In this embodiment, all interactive information and multimedia attribute information corresponding to the target multimedia data is obtained, and entity recognition is performed on all interactive information and multimedia attribute information to obtain corresponding second entities. Each second entity is then determined as a candidate point of interest, and a dictionary tree corresponding to the target multimedia data is constructed based on each candidate point of interest. This achieves comprehensive consideration of the interactive information and multimedia attribute information related to the target multimedia data, avoiding omissions of points of interest, making the obtained candidate points of interest more complete and comprehensive, and at the same time, the dictionary tree constructed based on the candidate points of interest is more complete, so that when subsequently identifying points of interest based on the dictionary tree, omissions of points of interest in the interactive information are avoided, and the determined target points of interest are more accurate.

[0148] In one embodiment, Figure 6 As shown in FIG, the steps of establishing the interest point index structure specifically include:

[0149] Step S602: performing entity recognition on multimedia attribute information of each multimedia data in the multimedia database to obtain a third entity of each multimedia data.

[0150] Among them, the multimedia attribute information includes title data and classification label data. The title data represents the title of the multimedia data, such as "Subject A sings the song "XXX"," "Subject A wears Hanfu and performs the program "XXX"," etc., while the classification label data of the multimedia data represents the interest tags pre-added to the multimedia data, such as "celebrity (such as Subject A)", "Song "XXX", "Singing", "Hanfu" and other interest tags.

[0151] Specifically, entity recognition is performed on the title data and classification label data of each multimedia data in the multimedia database, and entity mention fragments in the title data and classification label data are identified. Entity mining is further performed on the entity mention fragments to identify entities such as events, people, products, video names, drama titles, song names, movie names, and places, and obtain the third entity of each multimedia data.

[0152] Step S604: Match each third entity with each candidate interest point in the dictionary tree, determine the target entity in the third entity and the target candidate interest point in the dictionary tree, and match the target entity with the target candidate interest point.

[0153] Specifically, by matching the third entities of each multimedia data with each candidate interest point included in the dictionary tree, it is determined whether each third entity can be successfully matched with each candidate interest point included in the dictionary tree. If the match is successful, it indicates that the successfully matched third entity can be used as the target entity that matches the target candidate interest point among the candidate interest points. The purpose of matching each third entity with each candidate interest point in the dictionary tree is to determine mutually matching target entities and target candidate interest points, that is, to determine that the target entity determined from the third entity and the target candidate interest point determined from the dictionary tree are mutually matched.

[0154] Step S606: establishing a corresponding interest point index structure based on the target candidate interest points and the multimedia data identifier corresponding to each target entity.

[0155] Specifically, multimedia data identifiers corresponding to each target entity are obtained, and a corresponding point of interest index structure is established based on the target candidate points of interest and the multimedia data identifiers corresponding to each target entity. The point of interest index structure includes an index relationship between the points of interest and the multimedia data identifiers. That is, based on the point of interest index structure, the corresponding multimedia data identifiers can be found using the points of interest, and the corresponding multimedia recommendation data can be obtained from the multimedia database based on the found multimedia data identifiers.

[0156] Furthermore, if Figure 7 As shown, an interest point index structure is provided, referring to Figure 7As can be seen, a single target candidate point of interest can correspond to multiple multimedia data identifiers, and a single multimedia data identifier can also correspond to multiple target candidate points of interest. In other words, multiple multimedia data identifiers can be determined based on a single target candidate point of interest, and then multimedia recommendation data corresponding to each multimedia data identifier can be searched from the multimedia database. Similarly, a single multimedia data identifier can be searched for different target candidate points of interest, meaning that the same multimedia recommendation data may be recommended for different target candidate points of interest.

[0157] For example, refer to Figure 7 It can be seen that, for example, the target candidate interest points include "A object", "Hanfu", "singing", "XXX song", etc., and the multimedia database includes multimedia data with multimedia data identifiers of 01, 02, 03, 04, 05, and 06. Among them, the multimedia data identifiers corresponding to the target entity matching the target candidate interest point "A object" include 01, 02, 04 and 06, the multimedia data identifiers corresponding to the target entity matching the target candidate interest point "Hanfu" include 02 and 03, the multimedia data identifiers corresponding to the target entity matching the target candidate interest point "singing" include 03 and 04, and the multimedia data identifiers corresponding to the target entity matching the target candidate interest point "XXX song" include 01, 05 and 06. Then, according to each target candidate interest point and the multimedia data identifiers corresponding to each target entity matching each target candidate interest point, the following can be constructed. Figure 7 The POI index structure shown.

[0158] It is understood that the index relationship between the target candidate interest points and the multimedia data identifiers corresponding to the target entity, as well as the constructed interest point index structure, are not limited to those described in this embodiment or Figure 7 The specific situation shown in needs to be determined according to different situations in actual application scenarios. It can be different forms of interest point index structures and different index relationships. This application does not specifically limit this.

[0159] In one embodiment, the target recommendation information includes multimedia recommendation data, and the step of determining the target recommendation information based on the target point of interest specifically includes:

[0160] In the established interest point index structure, candidate interest points matching each target interest point are searched, and the interest point index structure includes an index relationship between the interest points and the multimedia data identifiers; based on the index relationship of the interest point index structure, the multimedia data identifiers corresponding to the candidate interest points are determined; and the multimedia recommendation data corresponding to each multimedia data identifier is determined as the target recommendation information.

[0161] Among them, the interest point index structure includes the index relationship between the interest point and the multimedia data identifier, that is, based on the interest point index structure, the corresponding multimedia data identifier can be found using the interest point, and the multimedia data identifier corresponds to the multimedia recommendation data, that is, through the found multimedia data identifier, the corresponding multimedia recommendation data can be obtained from the multimedia database.

[0162] Specifically, after establishing a corresponding POI index structure based on the target candidate POIs and the multimedia data identifiers corresponding to each target entity, the constructed POI index structure can be used to search for candidate POIs that match each target POI. Specifically, the POI index structure can be used to determine whether each target POI is stored in the POI index structure. If a match is found, it indicates that a candidate POI that matches the target POI exists in the POI index structure. Furthermore, based on the index relationship between the POIs and the multimedia data identifiers, the multimedia data identifiers corresponding to the candidate POIs that match each target POI can be obtained.

[0163] Furthermore, corresponding multimedia recommendation data is acquired from a multimedia database through the found multimedia data identifiers, and the multimedia recommendation data corresponding to each multimedia data identifier is determined as target recommendation information.

[0164] In one embodiment, the target recommendation information also includes product recommendation data. The target recommendation information is determined based on the target point of interest, including:

[0165] Perform product recognition processing on each target point of interest to generate corresponding product recognition results; based on the product recognition results, determine corresponding product recommendation data, and determine the product recommendation data as target recommendation information.

[0166] The target recommendation information may include multimedia recommendation data and product recommendation data. The multimedia recommendation data can be retrieved from the multimedia database based on the multimedia data identifier after searching for the multimedia data identifier based on the interest point index structure. The multimedia recommendation data may include recommended video data, recommended audio data, and recommended image data.

[0167] Product recommendation data refers to products, commodities, items or businesses related to the target interest points, such as specific recommended products and jump links of recommended products on certain shopping platforms, specific businesses of certain business platforms (such as different types of financial businesses, training businesses of online teaching platforms, etc.), and jump links of corresponding businesses.

[0168] In this embodiment, candidate points of interest that match each target point of interest can be searched within the established point of interest index structure. Based on the index relationships within the point of interest index structure, multimedia data identifiers corresponding to the candidate points of interest are determined, and multimedia recommendation data corresponding to each multimedia data identifier is determined as target recommendation information. This allows direct search based on the pre-established point of interest index structure to determine candidate points of interest that match the target point of interest, as well as their corresponding multimedia data identifiers. This improves the search speed and accuracy of multimedia data identifiers, thereby increasing the correlation between subsequently acquired multimedia data and the target point of interest. This improves the accuracy and hit rate of target recommendation information associated with the target point of interest.

[0169] In one embodiment, Figure 8 As shown, a method for determining target recommendation information is provided, which specifically includes the following steps:

[0170] Step S801: Acquire all interactive information and multimedia attribute information corresponding to target multimedia data, where the multimedia attribute information includes title data and classification label data.

[0171] Step S802: Perform entity recognition on all interactive information and multimedia attribute information to obtain a corresponding second entity.

[0172] Step S803: Determine each second entity as a candidate interest point, and construct a dictionary tree corresponding to the target multimedia data based on each candidate interest point.

[0173] Step S804: performing entity recognition on the multimedia attribute information of each multimedia data in the multimedia database to obtain a third entity of each multimedia data.

[0174] Step S805: Match each third entity with each candidate interest point in the dictionary tree respectively, determine the target entity in the third entity and the target candidate interest point in the dictionary tree, and match the target entity with the target candidate interest point.

[0175] Step S806: establishing a corresponding interest point index structure based on each candidate interest point and the multimedia data identifier corresponding to each target entity.

[0176] Step S807 : obtaining an interaction information set of the target multimedia data within the target time window, and performing entity recognition processing on each interaction information in the interaction information set to obtain a first entity corresponding to each interaction information.

[0177] Step S808: Acquire each candidate point of interest included in the dictionary tree, and perform point of interest matching on each first entity with each candidate point of interest included in the dictionary tree to obtain each point of interest within the target time window.

[0178] Step S809: Obtain each first text feature vector corresponding to each point of interest and each second text feature vector corresponding to each interactive information, and perform mapping processing on each first text feature vector and each second text feature vector to obtain a corresponding first semantic vector and second semantic vector.

[0179] In step S810 , semantic relevance calculations are performed based on the first semantic vectors and the second semantic vectors to obtain first relevance between each point of interest and each interactive information.

[0180] Step S811 : performing summation processing on the first relevance between each interest point and each interactive information to obtain the total relevance between the corresponding interest point and each interactive information.

[0181] Step S812: Obtain the total number of interactive information in the interactive information set, and average the total relevance according to the total number of interactive information to obtain a second relevance between the corresponding interest point and the interactive information set.

[0182] Step S813 , based on the second correlation between each interest point and the interactive information set, select a target interest point within the target time window from each interest point.

[0183] Step S814: searching for candidate interest points that match the target interest points in the established interest point index structure. The interest point index structure includes index relationships between interest points and multimedia data identifiers.

[0184] Step S815 : determining multimedia data identifiers corresponding to candidate interest points based on the index relationship of the interest point index structure, and determining multimedia recommendation data corresponding to each multimedia data identifier as target recommendation information.

[0185] In the above-described method for determining target recommendation information, by obtaining a set of interaction information for target multimedia data within a target time window, the interaction information set comprising each interaction information, it is possible to instantly identify points of interest within the target multimedia data within the target time window and continuously track changes in the points of interest within the target multimedia data within different time windows. Based on each interaction information, points of interest within the target time window can be identified, and the correlation between each point of interest and the interaction information set can be determined. Furthermore, based on the correlation between each point of interest and the interaction information set, target points of interest within the target time window can be selected from the points of interest, and ultimately, target recommendation information can be determined based on the target points of interest. By calculating the correlation between the interaction information set and each point of interest, the characteristics of the target multimedia data itself and the interaction information can be comprehensively considered, thereby improving the closeness between the determined target points of interest and the interaction information set, thereby improving the recommendation accuracy and recommendation hit rate of the target recommendation information associated with the target points of interest.

[0186] In one embodiment, Figure 9 As shown, an information recommendation method is provided, which specifically includes the following steps:

[0187] Step S902: Acquire portrait data of a target object for accessing target multimedia data.

[0188] Specifically, by detecting the target object accessing the target multimedia data and obtaining the portrait data of the target object, the target multimedia data can be video data, audio data or image data, such as video data of video viewing platforms, short video platforms and live broadcast platforms, etc., or audio data of audio listening platforms, etc., or image data of image viewing platforms, etc.

[0189] Among them, the target object can be understood as the user who accesses the target multimedia data, and the portrait data of the target object can include the target object's historical access (or search) records, followed points of interest, and click or search behavior on points of interest.

[0190] Step S904, filtering out directional interest points that match the portrait data from the target interest points within the target time window of the target multimedia data; the target interest points within the target time window are filtered out from the various interest points based on the correlation between the various interest points within the target time window of the target multimedia data and the interactive information set; the various interest points within the target time window are obtained based on interest point identification of the various interactive information in the interactive information set.

[0191] Specifically, the third correlation between each target interest point within the target time window of the target multimedia data and the portrait data of the target object can be calculated, and based on the third correlation between each target interest point and the portrait data, and the corresponding second preset correlation threshold, the directional interest points that match the portrait data can be screened out from each target interest point.

[0192] Furthermore, a trained deep semantic matching model is specifically used to calculate the third correlation between each target point of interest and the portrait data. Specifically, it is necessary to obtain each third text feature vector corresponding to each target point of interest, and the fourth text feature vector corresponding to the portrait data, and use the trained deep semantic matching model to map each third text feature vector and fourth text feature vector respectively to obtain the corresponding third semantic vector and fourth semantic vector. Then, based on each third semantic vector and fourth semantic vector, semantic correlation calculation is performed respectively to obtain the third correlation between each target point of interest and the portrait data.

[0193] Specifically, from each target interest point, directional interest points having a third relevance greater than a second preset relevance threshold are selected. It is understood that the second preset relevance threshold can be adjusted or modified according to actual needs and is not limited to one or more specific values.

[0194] For example, the target interest points include: "Object A" and "XXX Song", and the third correlation between each target interest point M and the portrait data N is represented by Match(M, N). Then, for each target interest point, the calculated third correlation may include: Match("Object A", portrait data) = 0.7, Match("XXX Song", portrait data) = 0.4. For example, if the second preset correlation threshold is set to 0.6, then according to the second preset correlation threshold, from each target interest point, the directional interest points whose third correlation is greater than the second preset correlation threshold of 0.6 are screened out. Since Match("Object A", portrait data) = 0.7 is greater than the second preset correlation threshold of 0.6, the target interest point "Object A" can be determined as a directional interest point.

[0195] In one embodiment, before selecting directional points of interest that match the portrait data from target points of interest within a target time window of target multimedia data, it is necessary to first obtain an interaction information set for the target multimedia data within the target time window, identify points of interest within the target time window based on each interaction information included in the interaction information set, and further calculate the correlation between each point of interest within the target time window and the interaction information set. Then, based on the correlation between each point of interest within the target time window of the target multimedia data and the interaction information set, target points of interest are selected from the points of interest.

[0196] When calculating the correlation between each point of interest and the multiple pieces of interactive information included in the interactive information set within the target time window, it is necessary to first determine the first correlation between each point of interest and each piece of interactive information. Then, based on the first correlation between each point of interest and each piece of interactive information, a second correlation between each point of interest and the interactive information set is further determined. Specifically, the first correlation between each point of interest and each piece of interactive information is summed to obtain the total correlation between the corresponding point of interest and each piece of interactive information. The total number of interactive information in the interactive information set is then obtained, and the total correlations are averaged based on the total number of interactive information to obtain the second correlation between the corresponding point of interest and the interactive information set.

[0197] Furthermore, when identifying points of interest for each interactive information in the interactive information set, a pre-established dictionary tree corresponding to the target multimedia data is obtained. Based on the candidate points of interest included in the dictionary tree, a point of interest matching is performed on the first entity in each interactive information included in the interactive information set to obtain the points of interest of the target multimedia data within the target time window. Based on a first preset relevance threshold and a second relevance between each point of interest and the interactive information set, the target point of interest within the target time window can be filtered out from the points of interest.

[0198] Step S906: Push the target recommendation information associated with the directional interest point to the target object.

[0199] Specifically, the target recommendation information associated with the targeted interest point may include multimedia recommendation data and product recommendation data, wherein the multimedia recommendation data may be recommended video data, recommended audio data, and recommended image data, etc., while the product recommendation data represents the products, commodities, items or services associated with the targeted interest point, such as specific recommended products and jump links of recommended products on certain shopping platforms, specific services of certain business platforms (such as different types of financial services, training services of online teaching platforms, etc.), and jump links of corresponding services, etc.

[0200] Furthermore, recommended video data, recommended audio data, and recommended image data associated with the targeted point of interest, or products, commodities, items, or services associated with the targeted point of interest, can be pushed to the target object, that is, pushed to the terminal device where the target object is located for display.

[0201] In one embodiment, when multimedia recommendation data associated with a targeted point of interest is pushed to a terminal device where a target object is located for display, the determined complete multimedia recommendation data can be pushed to the terminal device where the target object is located and displayed, or the target segments associated with the targeted point of interest can be filtered out after segmentation based on the multimedia recommendation data, and the filtered target segments can be pushed to the terminal device where the target object is located and displayed.

[0202] Among them, when the multimedia recommendation data is divided into segments, the multimedia recommendation data can be divided according to the preset segment length to obtain multiple segments of the same length, or the multimedia recommendation data can be randomly divided to obtain multiple segments (which can be multiple segments of the same or different lengths), or the multimedia recommendation data can be divided according to the data content and actual scenarios of the multimedia recommendation data by first identifying the data content and actual scenarios of the multimedia recommendation data, to ensure the content coherence and integrity of each segment obtained by division, without focusing on the segment length of each segment.

[0203] Furthermore, after the multimedia recommendation data is segmented to obtain multiple segments, for example, based on the data content and actual scenarios of the multimedia recommendation data, it is necessary to further calculate a fourth correlation between the directional interest point and each segment, and then select target segments from the multiple segments whose fourth correlation exceeds a third preset correlation threshold for recommendation. The third preset correlation threshold can be adjusted or modified based on actual needs and is not limited to a specific value or values.

[0204] For example, taking multimedia recommendation data as video recommendation data, for example, if the directional interest point is "object A and object B in the same frame", then for video recommendation data, such as a performance video of a certain evening party, which includes clips of different performance types such as solo, chorus, sketches and recitations by multiple different objects, after segmentation, it can be divided into multiple segments such as "object A solo "XX" song", "object C performs "XX" sketch", "D team performs chorus "XX"" and "object A and object B sing "XX" song" and so on. Then, according to the fourth correlation between the directional interest point and each segment, the corresponding target segment can be screened out from each segment, that is, the target segment corresponding to the directional interest point "object A and object B in the same frame" is determined to be the video segment of "object A and object B sing "XX" song".

[0205] Similarly, when the directional interest point is "object A and object B in the same frame", for example, the video recommendation data is a certain movie, then the movie includes different segments such as solo performances of multiple characters, rivalry between two or more characters, and group performances, such as "a solo performance of object A in XX scene", "a solo performance of object B in XX scene", "a diagonal performance between object A and object C", "a group performance between object A, object B, and object C", and "a diagonal performance between object A and object B". Then, based on the fourth degree of association between the directional interest point and each segment, the corresponding target segment can be screened out from each segment, that is, the target segment corresponding to the directional interest point "object A and object B in the same frame" is determined, including the video segment of "a group performance between object A, object B, and object C", and "a diagonal performance between object A and object B".

[0206] In the above-mentioned information recommendation method, by obtaining the portrait data of the target object accessing the target multimedia data, and screening the directional interest points that match the portrait data from the target interest points within the target time window of the target multimedia data, the target recommendation information associated with the directional interest points is pushed to the target object. This method achieves the goal of screening the target interest points from the various interest points within the target time window based on the correlation between the various interest points of the target multimedia data and the interactive information set, and screening the various target interest points based on a comprehensive consideration of the characteristics of the target multimedia data itself and the interactive information. The target interest points are then further screened based on the portrait data of the target object to determine the directional interest points that match the portrait data, making the association between the determined directional interest points and the portrait data closer, thereby improving the recommendation accuracy and recommendation hit rate of the target recommendation information when recommending the target recommendation information associated with the directional interest points.

[0207] It should be understood that, although the steps in the flowcharts of the above embodiments are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flowcharts of the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily to be performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0208] Based on the same inventive concept, the embodiments of the present application also provide a device for determining target recommendation information for implementing the target recommendation information determination method involved above, and an information recommendation device for implementing the information recommendation method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution described in the above method, so the specific limitations in the embodiments of one or more target recommendation information determination devices provided below can refer to the limitations of the target recommendation information determination method above. Similarly, the specific limitations in the embodiments of one or more target information recommendation devices provided below can refer to the limitations of the information recommendation method above, and will not be repeated here.

[0209] In one embodiment, Figure 10 As shown, a device for determining target recommendation information is provided, comprising: an interactive information set acquisition module 1002, an interest point identification module 1004, a relevance determination module 1006, a target interest point screening module 1008, and a target recommendation information determination module 1010, wherein:

[0210] The interactive information set acquisition module 1002 is configured to acquire an interactive information set of target multimedia data within a target time window, where the interactive information set includes various interactive information.

[0211] The interest point identification module 1004 is configured to identify interest points within a target time window based on the interaction information.

[0212] The relevance determination module 1006 is used to determine the relevance between each point of interest and the interactive information set.

[0213] The target interest point screening module 1008 is configured to screen target interest points within a target time window from the various interest points based on the relevance between the various interest points and the interactive information set.

[0214] The target recommendation information determination module 1010 is configured to determine target recommendation information based on the target point of interest.

[0215] The aforementioned device for determining target recommendation information obtains an interactive information set for target multimedia data within a target time window. The interactive information set includes each interactive information item. This allows for real-time identification of points of interest within the target multimedia data within the target time window, and allows for full tracking of changes in points of interest within the target multimedia data across different time windows. Based on each interactive information item, points of interest within the target time window can be identified, and the correlation between each point of interest and the interactive information set can be determined. Furthermore, based on the correlation between each point of interest and the interactive information set, target points of interest within the target time window can be selected from the points of interest, ultimately determining target recommendation information based on the target points of interest. By calculating the correlation between the interactive information set and each point of interest, the characteristics of the target multimedia data itself and the interactive information can be comprehensively considered, thereby increasing the closeness between the determined target points of interest and the interactive information set. This improves the accuracy and hit rate of the target recommendation information associated with the target points of interest.

[0216] In one embodiment, the relevance determination module is further configured to:

[0217] Determine a first correlation between each point of interest and each interactive information; and determine a second correlation between each point of interest and the interactive information set based on the first correlation between each point of interest and each interactive information.

[0218] In one embodiment, the relevance determination module is further configured to:

[0219] Obtain each first text feature vector corresponding to each point of interest, and each second text feature vector corresponding to each interactive information; perform mapping processing on each first text feature vector and each second text feature vector, respectively, to obtain a corresponding first semantic vector and second semantic vector; perform semantic relevance calculation based on each first semantic vector and each second semantic vector, respectively, to obtain a first relevance between each point of interest and each interactive information.

[0220] In one embodiment, the relevance determination module is further configured to:

[0221] The first correlation between each point of interest and each interactive information is summed to obtain the total correlation between the corresponding point of interest and each interactive information; the total number of interactive information in the interactive information set is obtained; and the total correlation is averaged according to the total number of interactive information to obtain the second correlation between the corresponding point of interest and the interactive information set.

[0222] In one embodiment, the point of interest identification module is further configured to:

[0223] Obtain a dictionary tree corresponding to the target multimedia data; the dictionary tree includes each candidate point of interest; perform entity recognition processing on each interactive information to obtain a first entity corresponding to each interactive information; perform point of interest matching on each first entity with each candidate point of interest included in the dictionary tree to obtain each point of interest within the target time window.

[0224] In one embodiment, a device for determining target recommendation information is provided, further comprising a dictionary tree construction module for:

[0225] Acquire all interactive information and multimedia attribute information corresponding to the target multimedia data, where the multimedia attribute information includes title data and classification label data; perform entity recognition on all interactive information and multimedia attribute information to obtain corresponding second entities; determine each second entity as a candidate point of interest, and construct a dictionary tree corresponding to the target multimedia data based on each candidate point of interest.

[0226] In one embodiment, a device for determining target recommendation information is provided, further comprising a module for establishing an interest point index structure, configured to:

[0227] Entity recognition is performed on the multimedia attribute information of each multimedia data in the multimedia database to obtain the third entity of each multimedia data, where the multimedia attribute information includes title data and classification label data; each third entity is matched with each candidate interest point in the dictionary tree respectively, and the target entity in the third entity and the target candidate interest point in the dictionary tree are determined, and the target entity is matched with the target candidate interest point; based on each target candidate interest point and the multimedia data identifier corresponding to each target entity, a corresponding interest point index structure is established.

[0228] In one embodiment, the target recommendation information determination module is further configured to:

[0229] In the established interest point index structure, candidate interest points matching each target interest point are searched, and the interest point index structure includes an index relationship between the interest points and the multimedia data identifiers; based on the index relationship of the interest point index structure, the multimedia data identifiers corresponding to the candidate interest points are determined; and the multimedia recommendation data corresponding to each multimedia data identifier is determined as the target recommendation information.

[0230] In one embodiment, the target recommendation information determination module is further configured to:

[0231] Perform product recognition processing on each target point of interest to generate corresponding product recognition results; based on the product recognition results, determine corresponding product recommendation data, and determine the product recommendation data as target recommendation information.

[0232] In one embodiment, Figure 11As shown, an information recommendation device is provided, including: a portrait data acquisition module 1102, a directional interest point screening module 1104 and a target recommendation information push module 1106, wherein:

[0233] The portrait data acquisition module 1102 is used to acquire the portrait data of the target object for accessing the target multimedia data.

[0234] The directional interest point screening module 1104 is used to screen the directional interest points that match the portrait data from the target interest points within the target time window of the target multimedia data; the target interest points within the target time window are screened from the various interest points based on the correlation between the various interest points of the target multimedia data within the target time window and the interactive information set, and the various interest points within the target time window are obtained based on the interest point identification of each interactive information in the interactive information set.

[0235] The target recommendation information pushing module 1106 is configured to push the target recommendation information associated with the directional interest point to the target object.

[0236] In the above-mentioned information recommendation device, by acquiring the portrait data of the target object accessing the target multimedia data, and screening the directional interest points that match the portrait data from the target interest points within the target time window of the target multimedia data, the target recommendation information associated with the directional interest points is pushed to the target object. This achieves the goal of screening out the target interest points from the various interest points within the target time window based on the correlation between the various interest points of the target multimedia data and the interactive information set, and screening out each target interest point based on a comprehensive consideration of the characteristics of the target multimedia data itself and the interactive information. The target interest points are then further screened based on the portrait data of the target object to determine the directional interest points that match the portrait data, making the association between the determined directional interest points and the portrait data closer, thereby improving the recommendation accuracy and recommendation hit rate of the target recommendation information when recommending the target recommendation information associated with the directional interest points.

[0237] The aforementioned target recommendation information determination device and the various modules within the information recommendation device may be implemented in whole or in part via software, hardware, or a combination thereof. Each of the aforementioned modules may be embedded in or independent of a processor within a computer device in the form of hardware, or may be stored in a computer device memory in the form of software, so that the processor can call and execute the corresponding operations of each of the aforementioned modules.

[0238] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 12As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data such as target multimedia data, interactive information, interactive information sets, points of interest, the correlation between points of interest and interactive information sets, target points of interest, target recommendation information, portrait data of target objects, directional points of interest, and target objects. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a method for determining target recommendation information and an information recommendation method.

[0239] Those skilled in the art will understand that Figure 12 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0240] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0241] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0242] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above-mentioned method embodiments when executed by a processor. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions. Among them, the user can reject the pushed products or advertisements or can easily reject the push information of advertisements or products.

[0243] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, and the like.

[0244] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0245] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for determining target recommendation information, characterized in that: The method comprises: Acquire an interaction information set of target multimedia data within a target time window, wherein the interaction information set includes each piece of interaction information; Acquire a dictionary tree corresponding to the target multimedia data; the dictionary tree includes each candidate point of interest; Performing entity recognition processing on each interactive information to obtain a first entity corresponding to each interactive information, performing point of interest matching on each first entity with each candidate point of interest included in the dictionary tree to obtain each point of interest within the target time window; Determining the relevance of each of the interest points to the interactive information set; Filtering target interest points within the target time window from the interest points according to the relevance between the interest points and the interactive information set; Target recommendation information is determined based on the target point of interest.

2. The method according to claim 1, characterized in that The determining of the relevance between each of the interest points and the interactive information set includes: Determining a first relevance between each of the interest points and each of the interactive information; The second relevance between each of the interest points and the set of interactive information is determined based on the first relevance between each of the interest points and each of the interactive information.

3. The method according to claim 2, characterized in that The determining of the first relevance between each of the interest points and each of the interactive information includes: Obtaining each first text feature vector corresponding to each point of interest and each second text feature vector corresponding to each interactive information; Performing mapping processing on each of the first text feature vectors and each of the second text feature vectors to obtain a corresponding first semantic vector and a second semantic vector; Based on each of the first semantic vectors and each of the second semantic vectors, semantic relevance calculations are performed to obtain first relevance between each of the interest points and each of the interactive information.

4. The method according to claim 2, characterized in that The determining, by using the first relevance between each of the interest points and each of the interactive information, a second relevance between each of the interest points and the interactive information set includes: performing summation processing on the first correlation between each of the interest points and each of the interactive information to obtain a total correlation between the corresponding interest point and each of the interactive information; Obtaining the total number of interactive information in the interactive information set; According to the total number of the interactive information, the total correlation is averaged to obtain a second correlation between the corresponding interest point and the interactive information set.

5. The method according to claim 1, wherein The method of obtaining the dictionary tree corresponding to the target multimedia data includes: Acquire all interactive information and multimedia attribute information corresponding to the target multimedia data, wherein the multimedia attribute information includes title data and classification label data; Performing entity recognition on all the interactive information and multimedia attribute information to obtain a corresponding second entity; Each of the second entities is determined as a candidate point of interest, and a dictionary tree corresponding to the target multimedia data is constructed based on each of the candidate points of interest.

6. The method according to claim 1, characterized in that The target recommendation information includes multimedia recommendation data; and determining the target recommendation information based on the target point of interest includes: searching for candidate points of interest that match each of the target points of interest in an established point of interest index structure, wherein the point of interest index structure includes an index relationship between the points of interest and multimedia data identifiers; Determining multimedia data identifiers corresponding to the candidate points of interest based on index relationships of the points of interest index structure; The multimedia recommendation data corresponding to each multimedia data identifier is determined as target recommendation information.

7. The method according to claim 6, characterized in that The method for establishing the interest point index structure includes: Performing entity recognition on multimedia attribute information of each multimedia data in a multimedia database to obtain a third entity of each multimedia data, wherein the multimedia attribute information includes title data and classification label data; Matching each of the third entities with each of the candidate interest points in the dictionary tree, determining a target entity in the third entity and a target candidate interest point in the dictionary tree, wherein the target entity matches the target candidate interest point; Based on each of the candidate target interest points and the multimedia data identifier corresponding to each of the target entities, a corresponding interest point index structure is established.

8. The method according to any one of claims 1 to 4, characterized in that The target recommendation information includes product recommendation data; and determining the target recommendation information based on the target point of interest includes: Performing product recognition processing on each of the target points of interest to generate corresponding product recognition results; Based on the product identification result, corresponding product recommendation data is determined, and the product recommendation data is determined as target recommendation information.

9. An information recommendation method, characterized in that: The method comprises: Obtaining portrait data of a target object accessing target multimedia data; Filtering directional interest points that match the portrait data from target interest points within a target time window of the target multimedia data, wherein the target interest points within the target time window are filtered from the interest points based on the correlation between the interest points within the target time window and the interactive information set of the target multimedia data; the interest points within the target time window are obtained by performing interest point identification on each interactive information in the interactive information set; the interest point identification based on each interactive information in the interactive information set comprises: obtaining a dictionary tree corresponding to the target multimedia data, wherein the dictionary tree includes each candidate interest point; performing entity recognition processing on each interactive information to obtain a first entity corresponding to each interactive information, performing interest point matching on each first entity with each candidate interest point included in the dictionary tree to obtain each interest point within the target time window; Push target recommendation information associated with the directional interest point to the target object.

10. A device for determining target recommendation information, characterized in that: The device comprises: An interactive information set acquisition module, configured to acquire an interactive information set of target multimedia data within a target time window, wherein the interactive information set includes various interactive information; an interest point identification module, configured to obtain a dictionary tree corresponding to the target multimedia data, the dictionary tree including each candidate interest point; perform entity recognition processing on each interactive information to obtain a first entity corresponding to each interactive information; perform interest point matching on each first entity with each candidate interest point included in the dictionary tree to obtain each interest point within the target time window; A relevance determination module, configured to determine the relevance between each of the interest points and the interactive information set; a target interest point screening module, configured to screen target interest points within the target time window from the various interest points according to the relevance between the various interest points and the interactive information set; The target recommendation information determination module is used to determine target recommendation information based on the target point of interest.

11. The device according to claim 10, characterized in that The relevance determination module is further configured to: Determine a first correlation between each of the interest points and each of the interactive information; and determine a second correlation between each of the interest points and the interactive information set based on the first correlation between each of the interest points and each of the interactive information.

12. The device according to claim 11, characterized in that The relevance determination module is further configured to: Obtain each first text feature vector corresponding to each of the interest points, and each second text feature vector corresponding to each of the interactive information; perform mapping processing on each of the first text feature vectors and each of the second text feature vectors to obtain a corresponding first semantic vector and second semantic vector; perform semantic relevance calculation based on each of the first semantic vectors and each of the second semantic vectors to obtain a first relevance between each of the interest points and each of the interactive information.

13. The device according to claim 11, characterized in that The relevance determination module is further configured to: performing summation processing on the first correlation between each of the interest points and each of the interactive information to obtain a total correlation between the corresponding interest point and each of the interactive information; and obtaining a total number of interactive information in the interactive information set; According to the total number of the interactive information, the total correlation is averaged to obtain a second correlation between the corresponding interest point and the interactive information set.

14. The device according to claim 10, characterized in that The device further includes a dictionary tree construction module, which is used to: Acquire all interactive information and multimedia attribute information corresponding to the target multimedia data, wherein the multimedia attribute information includes title data and classification label data; perform entity recognition on the all interactive information and multimedia attribute information to obtain corresponding second entities; determine each second entity as a candidate point of interest, and construct a dictionary tree corresponding to the target multimedia data based on each candidate point of interest.

15. The device according to claim 10, characterized in that The target recommendation information determination module is further configured to: In the established point of interest index structure, candidate points of interest that match each of the target points of interest are searched, the point of interest index structure including an index relationship between the points of interest and the multimedia data identifiers; based on the index relationship of the point of interest index structure, the multimedia data identifiers corresponding to the candidate points of interest are determined; and the multimedia recommendation data corresponding to each of the multimedia data identifiers is determined as target recommendation information.

16. The device according to claim 15, characterized in that The device further includes a point of interest index structure establishing module, which is used to: Performing entity recognition on multimedia attribute information of each multimedia data in a multimedia database to obtain a third entity of each multimedia data, wherein the multimedia attribute information includes title data and classification label data; matching each third entity with each candidate interest point in the dictionary tree, and determining a target entity in the third entity and a target candidate interest point in the dictionary tree, wherein the target entity matches the target candidate interest point; Based on each of the candidate target interest points and the multimedia data identifier corresponding to each of the target entities, a corresponding interest point index structure is established.

17. The device according to any one of claims 10 to 13, characterized in that The target recommendation information includes product recommendation data; the target recommendation information determination module is further configured to: Perform product recognition processing on each of the target interest points to generate a corresponding product recognition result; determine corresponding product recommendation data based on the product recognition result, and determine the product recommendation data as target recommendation information.

18. An information recommendation device, characterized in that: The device comprises: A portrait data acquisition module, used to acquire portrait data of a target object accessing target multimedia data; a directional interest point screening module, configured to screen directional interest points that match the portrait data from target interest points within a target time window of the target multimedia data; the target interest points within the target time window are screened from each interest point according to the correlation between each interest point and the interactive information set of the target multimedia data within the target time window; each interest point within the target time window is obtained by performing interest point identification on each interactive information in the interactive information set; the performing interest point identification on each interactive information in the interactive information set comprises: obtaining a dictionary tree corresponding to the target multimedia data, the dictionary tree including each candidate interest point; performing entity recognition processing on each interactive information to obtain a first entity corresponding to each interactive information, performing interest point matching on each first entity with each candidate interest point included in the dictionary tree to obtain each interest point within the target time window; The target recommendation information pushing module is used to push the target recommendation information associated with the directional interest point to the target object.

19. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

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

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

Images