Multimedia content push method, device, equipment and storage medium

Abstract

The present application discloses a method, apparatus, device and storage medium for pushing multimedia content, which belongs to the field of artificial intelligence technology. The method includes: obtaining service location information and service content information associated with multimedia content; inputting the service location information and service content information into a crowd-oriented joint model to obtain the service geographic features and service preference features of the multimedia content; fusing the service geographic features and service preference features with the account features of each user account to obtain the conversion tendency parameters of each user account for multimedia content; and pushing multimedia content according to the conversion tendency parameters. The technical solution provided by the embodiment of the present application trains the crowd-oriented joint model through sorting consistency conditions, and only requires the service location and service content to determine the conversion tendency parameters of the multimedia content and perform crowd targeting and content push, thereby improving the quality of crowd targeting and the efficiency of multimedia content push.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and storage medium for pushing multimedia content. Background Art

[0002] Before releasing multimedia content, multimedia content publishers need to identify the target audience as a targeting condition on the delivery side. Audience targeting is the first step in releasing multimedia content.

[0003] In related technologies, for new publishers who need to deliver multimedia content, such as local catering businesses or local supermarkets, since new publishers of multimedia content lack historical behavior groups, they can only extract people who have visited the city in the past as the target group.

[0004] In related technologies, the quality of crowd targeting is low and the efficiency of multimedia content push is low. Summary of the Invention

[0005] The embodiments of the present application provide a method, apparatus, device, and storage medium for pushing multimedia content, which can improve the quality of crowd targeting, reduce dependence on historical data, and improve the efficiency of multimedia content pushing.

[0006] According to one aspect of an embodiment of the present application, a method for pushing multimedia content is provided, the method comprising:

[0007] Obtaining service location information and service content information associated with multimedia content;

[0008] Inputting the service location information and the service content information into a crowd-oriented joint model to obtain service geographic features and service preference features of the multimedia content;

[0009] fusing the service geographic feature and the service preference feature with the account feature of each user account to obtain a conversion tendency parameter of each user account for the multimedia content;

[0010] pushing the multimedia content according to the conversion tendency parameter;

[0011] The crowd-oriented joint model is trained based on a ranking consistency constraint, where the ranking consistency constraint means that the conversion data of the user account for the target multimedia content is positively correlated with the target conversion tendency parameter of the user account for the target multimedia content.

[0012] According to one aspect of an embodiment of the present application, a device for pushing multimedia content is provided, the device comprising:

[0013] A service information acquisition module, used to obtain service location information and service content information associated with multimedia content;

[0014] A service feature determination module, configured to input the service location information and the service content information into a crowd-oriented joint model to obtain a service geographic feature and a service preference feature of the multimedia content;

[0015] a conversion parameter prediction module, configured to fuse the service geographic feature and the service preference feature with the account feature of each user account to obtain a conversion tendency parameter of each user account for the multimedia content;

[0016] a content push module, configured to push the multimedia content according to the conversion tendency parameter;

[0017] The crowd-oriented joint model is trained based on a ranking consistency constraint, where the ranking consistency constraint means that the conversion data of the user account for the target multimedia content is positively correlated with the target conversion tendency parameter of the user account for the target multimedia content.

[0018] According to one aspect of an embodiment of the present application, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the above-mentioned method for pushing multimedia content.

[0019] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, wherein the storage medium stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the above-mentioned multimedia content pushing method.

[0020] According to one aspect of an embodiment of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the above-described method for pushing multimedia content.

[0021] The technical solutions provided in the embodiments of the present application can bring the following beneficial effects:

[0022] By setting a sorting consistency constraint condition that the conversion data of the user account for the same multimedia content is positively correlated with the conversion tendency parameter, the crowd targeting joint model is trained so that the crowd targeting joint model only needs the service location and service content of the multimedia content to determine the service geographic characteristics and service preference characteristics of the multimedia content, and can combine the above-mentioned service geographic characteristics, service preference characteristics and account characteristics to predict the parameters that measure the user's conversion to the multimedia content, and then perform crowd targeting and push multimedia content based on the parameters. The above-mentioned sorting consistency constraint condition can enable new users to perform relatively accurate crowd targeting for newly released multimedia content even when there is no historical data or the historical data is relatively sparse, thereby improving the quality of crowd targeting and reducing dependence on historical data, thereby improving the efficiency of multimedia content push, avoiding waste of computing resources, and reducing equipment operation pressure. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0024] Figure 1 is a schematic diagram of an application program operating environment provided by an embodiment of the present application;

[0025] Figure 2 This is a flowchart of a method for pushing multimedia content provided by one embodiment of the present application;

[0026] Figure 3 This is a flowchart of a method for pushing multimedia content provided by one embodiment of the present application;

[0027] Figure 4 The following is an exemplary diagram of a training flow chart for a crowd-oriented joint model;

[0028] Figure 5 is a flowchart of a method for pushing multimedia content provided by another embodiment of the present application;

[0029] Figure 6 A schematic diagram of a process of pushing multimedia content is exemplified;

[0030] Figure 7 This is a block diagram of a multimedia content push device provided by one embodiment of the present application;

[0031] Figure 8 This is a structural block diagram of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0032] The multimedia content push method provided in the embodiment of the present application involves artificial intelligence technology and cloud technology, which are briefly described below to facilitate understanding by those skilled in the art.

[0033] Cloud computing refers to the delivery and usage model of IT (Internet Technology) infrastructure, enabling on-demand, scalable access to resources over the internet. In a broader sense, cloud computing refers to the delivery and usage model of services, enabling on-demand, scalable access to services over the internet. These services can be IT-related, software-related, internet-related, or other services. Cloud computing is the product of the convergence of traditional computer and network technologies, including grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing.

[0034] Cloud computing has rapidly grown with the development of the internet, real-time data streams, and the diversification of connected devices, as well as the growing demand for search services, social networks, mobile commerce, and open collaboration. Unlike conventional parallel distributed computing, the emergence of cloud computing will fundamentally revolutionize the entire internet model and enterprise management model. In the multimedia content push method provided in the embodiments of this application, a cloud server can be used to target and push multimedia content to a specific audience.

[0035] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to 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 seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0036] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, as well as machine learning / deep learning, autonomous driving, and smart transportation.

[0037] Natural language processing (NLP) is a key area of ​​research in computer science and artificial intelligence. It studies the theories and methods that enable effective communication between humans and computers using natural language. Natural language processing (NLP) integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language we use in everyday life—and is closely linked to the study of linguistics. Natural language processing technologies typically include text processing, semantic understanding, machine translation, robotic question answering, and knowledge graphs.

[0038] 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.

[0039] With the research and progress 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, automatic driving, drones, robots, smart medical care, smart customer service, Internet of Vehicles, automatic driving, smart transportation, 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.

[0040] The multimedia content push method provided in the embodiments of the present application involves the above-mentioned artificial intelligence natural language processing, machine learning and other technologies, and is specifically illustrated by the following embodiments.

[0041] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0042] Please refer to Figure 1 , which shows a schematic diagram of an application program running environment provided by an embodiment of the present application. The application program running environment may include: a terminal 10 and a server 20.

[0043] The terminal 10 may be an electronic device such as a mobile phone, a tablet computer, a game console, an e-book reader, a multimedia player, a wearable device, a PC (Personal Computer), etc. A client of an application program may be installed in the terminal 10 .

[0044] In an embodiment of the present application, the above-mentioned application can be any application that can push multimedia content. Typically, the application is an information flow content service application. Of course, in addition to information flow content service applications, other types of applications can also provide services for pushing multimedia content. For example, news applications, social applications, interactive entertainment applications, browser applications, shopping applications, content sharing applications, virtual reality (VR) applications, augmented reality (AR) applications, etc., which are not limited in the embodiment of the present application. In addition, for different applications, the multimedia content pushed will also be different, and the corresponding functions will also be different. This can be pre-configured according to actual needs, which is not limited in the embodiment of the present application. Optionally, a client of the above-mentioned application is running in the terminal 10.

[0045] The server 20 is used to provide background services for the client of the application in the terminal 10. For example, the server 20 can be the background server of the above-mentioned application. The server 20 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the server 20 provides background services for applications in multiple terminals 10 at the same time.

[0046] Optionally, the terminal 10 and the server 20 may communicate with each other via a network 30. The terminal 10 and the server 20 may be directly or indirectly connected via wired or wireless communication, which is not limited in this application.

[0047] Please refer to Figure 2, which shows a flowchart of a method for pushing multimedia content provided by an embodiment of the present application. The method can be applied to a computer device, which refers to an electronic device with data calculation and processing capabilities. For example, the execution subject of each step can be Figure 1 The server 20 in the application running environment is shown. The method may include the following steps (210-240).

[0048] Step 210: Acquire service location information and service content information associated with multimedia content.

[0049] The service location associated with the multimedia content may also be the geographic location of the publisher of the multimedia content. The service location information includes the latitude and longitude coordinates of the location. The service content information may be text information within the multimedia content, or it may be tag information obtained by extracting features from the multimedia content. For example, if the multimedia content includes an image, semantic recognition may be performed on the image to obtain a corresponding classification tag as the service content information.

[0050] In one possible implementation, the multimedia content is a local life service advertisement, and the service location information associated with the multimedia content may be the latitude and longitude data of the service address corresponding to the local life service advertisement. The service content information may be the advertising copy of the local life service advertisement, or the industry label corresponding to the local life service advertisement, or a label added by the advertiser based on the life service content.

[0051] Step 220 : Input the service location information and service content information into a crowd-oriented joint model to obtain the service geographic features and service preference features of the multimedia content.

[0052] Among them, the crowd-oriented joint model is trained based on the sorting consistency constraint condition, and the sorting consistency constraint condition means that the conversion data of the user account for the target multimedia content is positively correlated with the target conversion tendency parameter of the user account for the target multimedia content.

[0053] The aforementioned service geographic features are used to measure the influence of multimedia content on different service areas in a city. The aforementioned service preference features are used to measure the similarity of conversion behaviors between different multimedia content.

[0054] The conversion tendency parameter can be determined through step 230 , that is, the conversion tendency parameter is determined based on the service geographic characteristics and the service preference characteristics.

[0055] For example, if the number of conversions of user a to multimedia content b is greater than the number of conversions of user c to multimedia content b, then the conversion tendency parameter value of user a to multimedia content b should also be greater than the conversion tendency parameter value of user c to multimedia content b.

[0056] In a possible implementation, the service geographic feature is a vector representation of the multimedia content in the target geographic feature space, that is, a geographic feature vector of the multimedia content.

[0057] In a possible implementation, the service preference feature is a vector representation of the multimedia content in the target preference feature space, that is, a preference feature vector of the multimedia content.

[0058] In an exemplary embodiment, as Figure 3 As shown, Figure 3 The flowchart of the multimedia content push method provided by one embodiment of the present application is shown. The above step 220 includes the following sub-steps (221-223).

[0059] Step 221 : Compare the service location information with the regional location information of each service area to obtain the service geographical features.

[0060] The above-mentioned regional location information includes the location of the regional center, such as the latitude and longitude coordinates corresponding to the regional center. The service location is compared with the regional center location to obtain the geographical feature vector of the multimedia content.

[0061] In one possible implementation, the regional center location of each service area is obtained; based on the service location and the regional center location of each service area, a geographic feature vector of the multimedia content is determined, where the geographic feature vector includes a probability of a user moving from the kth regional center to the service location.

[0062] Optionally, the method for determining the geographic feature vector of the multimedia content may refer to the method for determining the geographic feature vector of the multimedia content in the following embodiment. j The method will not be described here.

[0063] Step 222: Update at least one multimedia content semantic graph based on the service location information and the service content information to obtain an updated multimedia content semantic graph.

[0064] At least one multimedia content semantic graph is used to represent the degree of association between each multimedia content and different semantic node sets.

[0065] Optionally, at least one multimedia content semantic graph includes the content and content semantic graph G in the following embodiment. LL , content and conversion time semantic graph G LT , content and label semantic graph G LW , content and region semantic graph G LR , content and neighboring user semantic graph G LU At least one of them.

[0066] Correspondingly, the above-mentioned semantic node set can be the set L of all multimedia contents, the conversion time set T, the set W of all semantic tags, the area set R, and the set U of user accounts that have conversion behavior on multimedia content within 200 meters of the service location of the specified multimedia content during the target period in the embodiments below.

[0067] In a possible implementation, the multimedia content is newly released, and the service location information and service content information of the multimedia content may be added to the multimedia content semantic graph to update the graph structure data of the multimedia content semantic graph.

[0068] Step 223 : determining the service preference characteristics of the multimedia content based on the graph structure data corresponding to the updated multimedia content semantic graph.

[0069] In one possible implementation, a graph representation learning method is used to extract features from the graph structure data corresponding to the semantic graph of multimedia content to determine the preference feature vector of the multimedia content. Optionally, the preference feature vector of the multimedia content can be v in the following embodiment. j .

[0070] In an exemplary embodiment, as Figure 4 As shown, Figure 4 The training flow chart of a crowd-oriented joint model is shown as an example. The training process of the crowd-oriented joint model includes the following steps (410-430).

[0071] Step 410: Obtain sample data log.

[0072] The sample data log includes a behavior data record of a sample user account for sample multimedia content. The behavior data record includes tag information of the sample multimedia content. The tag information is conversion data of the sample user account for the sample multimedia content.

[0073] In some embodiments, the sample data log includes a first log record and a second log record. The first log record may be recorded in the format of (sample user account, sample multimedia content identification number, number of exposures, number of clicks, number of conversions). The second log record may be recorded in the format of (sample user account, sample multimedia content identification number, conversion time).

[0074] Step 420 : Determine service information of sample multimedia content, user profile data of sample user accounts, and at least one semantic graph of the multimedia content sample based on the sample data log.

[0075] In a possible implementation, the service information of the sample multimedia content can be directly extracted from the sample data log, the user profile data of the sample user account can be determined by the historical conversion data of the sample user account, and the method for constructing the semantic graph of the multimedia content sample can refer to the content and content semantic graph G. LL , content and conversion time semantic graph G LT , content and label semantic graph G LW , content and region semantic graph G LR , content and neighboring user semantic graph G LU The construction method.

[0076] Step 430 , based on the service information of the sample multimedia content, the user portrait data of the sample user account and at least one semantic graph of the multimedia content sample, and in accordance with the joint constraint conditions, the crowd-oriented joint model is trained until the output result of the crowd-oriented joint model satisfies the joint constraint conditions.

[0077] Among them, the joint constraints include sorting consistency constraints and semantic graph loss conditions.

[0078] In one possible implementation, the above-mentioned joint constraint condition can be expressed by the mathematical expression (18) in the following embodiment, and the above-mentioned joint constraint condition is the final loss function Satisfy the loss conditions.

[0079] In a possible implementation, the above-mentioned sorting consistency constraint condition can be expressed by the mathematical expression (12) in the following embodiment.

[0080] In one possible implementation, the above semantic graph loss condition can be expressed by the mathematical expression (17) in the following example.

[0081] Step 230 : The service geographic feature and the service preference feature are respectively integrated with the account feature of each user account to obtain a conversion tendency parameter of each user account for multimedia content.

[0082] In a possible implementation, the conversion tendency parameter is a value used to represent the probability of each user account performing a conversion behavior on the multimedia content.

[0083] The above fusion can be the calculation of feature distance or the inner product of vectors between feature vectors, which is not limited in the embodiment of the present application.

[0084] In an exemplary embodiment, as Figure 3 As shown, the above method further includes the following steps (250-270).

[0085] Step 250: Obtain data log.

[0086] The data log includes the behavior data records of each user account for each multimedia content.

[0087] In some embodiments, the data log includes a first log record and a second log record. The first log record and the second log record are both behavioral data records of each user account for each multimedia content, and the recorded data may be different. The recording format of the first log record may be (user account, multimedia content identification number, number of exposures, number of clicks, number of conversions). The recording format of the second log record may be (user account, multimedia content identification number, conversion time).

[0088] Step 260: Generate at least one multimedia content semantic graph based on the behavior data record.

[0089] Among them, at least one multimedia content semantic graph includes at least one of a content and content semantic graph, a content and conversion time semantic graph, a content and label semantic graph, a content and area semantic graph, and a content and neighboring user semantic graph.

[0090] Optionally, the specific implementation process of the above step 260 can refer to the following content and content semantic graph G LL , content and conversion time semantic graph G LT , content and label semantic graph G LW , content and region semantic graph G LR , content and neighboring user semantic graph G LU The construction method.

[0091] Step 270 : Determine account characteristics of each user account based on the data log and at least one multimedia content semantic graph.

[0092] In an exemplary embodiment, as Figure 5 As shown, Figure 5 The above-mentioned account characteristics include account geographical characteristics and account preference characteristics, and the above-mentioned step 270 includes the following sub-steps (271-272).

[0093] In step 271 , the user portrait data of each user account is input into a crowd-oriented joint model for feature extraction to obtain the account preference features of each user account.

[0094] The user profile data is generated based on data logs. The account preference features are used to measure the similarity of conversion behaviors between different user accounts. Optionally, the account preference features can be the user preference feature vector v in the following embodiment. i .

[0095] In one possible implementation, the user portrait data of each user account is input into a pre-trained neural network model for feature extraction to obtain a user preference feature vector. The pre-trained neural network model can be composed of a three-layer neural network, and the parameters of the neural network model are training parameters learned by the model, and the neural network model is subject to the above-mentioned joint constraints.

[0096] Step 272 : Inputting the graph structure data corresponding to at least one multimedia content semantic graph into the crowd-oriented joint model to obtain the account geographic features of each user account.

[0097] The account geographical feature is used to measure the influence of the user account on different service areas in the city. In a possible implementation, the account geographical feature can be the geographical feature vector of the user account. Alternatively, the account geographical feature can be the user's geographical feature vector g j , including the expected number of conversions of multimedia content by user accounts in each service area.

[0098] Correspondingly, such as Figure 5 As shown, the above step 230 includes the following sub-steps (231-233).

[0099] In step 231 , for each user account, a geographic feature parameter is determined based on the service geographic feature and the account geographic feature of the user account.

[0100] In a possible implementation, the geographic feature parameter may be the vector inner product of the geographic feature vector of the multimedia content and the geographic feature vector of the user account, obtained by performing a dot product of the geographic feature vector of the multimedia content and the geographic feature vector of the user account.

[0101] Step 232 : Determine preference characteristic parameters based on the service preference characteristic and the account preference characteristic of the user account.

[0102] In a possible implementation, the preference feature parameter may be the vector inner product of the preference feature vector of the multimedia content and the preference feature vector of the user account, obtained by performing a dot product of the preference feature vector of the multimedia content and the preference feature vector of the user account.

[0103] Step 233 : Determine the conversion tendency parameter of the user account for multimedia content based on the geographic feature parameter and the preference feature parameter.

[0104] In one possible implementation, the conversion tendency parameter is the sum of the inner product of the geographic feature vector of the multimedia content and the geographic feature vector of the user account, and the inner product of the preference feature vector of the multimedia content and the preference feature vector of the user account, that is, the sum of the inner products of the two vectors.

[0105] Alternatively, the conversion tendency parameter may be the conversion tendency score y in the following examples. ij .

[0106] Finally, conversion tendency parameters for each user account with respect to multimedia content are generated.

[0107] Step 240: Push multimedia content according to the conversion tendency parameter.

[0108] In an exemplary embodiment, as Figure 3 As shown, the above step 240 includes the following sub-steps (241-242).

[0109] Step 241 : Determine a target user account corresponding to a target conversion tendency parameter that meets the conversion tendency parameter condition.

[0110] In one possible implementation, the conversion tendency parameter condition is a threshold condition, such as a conversion tendency parameter value being greater than a preset parameter threshold. Accordingly, the target conversion tendency parameter is a conversion tendency parameter whose parameter value is greater than the preset parameter threshold.

[0111] The target user account is the user account corresponding to the target conversion tendency parameter. The users corresponding to the target user account can constitute a user group, that is, the result of population targeting.

[0112] Step 242: Push multimedia content to the target user account.

[0113] When the target user account is connected to the target traffic domain, multimedia content is pushed to the target user account.

[0114] The target traffic domain may be a traffic domain for delivering multimedia content, such as a traffic domain of a certain social application.

[0115] In an exemplary embodiment, as Figure 5 As shown, the above method also includes the following steps (280-310).

[0116] Step 280: Obtain the conversion user account corresponding to the multimedia content.

[0117] In some embodiments, after new multimedia content is delivered or pushed, a conversion behavior may occur, so the conversion user account corresponding to the multimedia content can be obtained. The above-mentioned conversion user account refers to the user account that has converted the multimedia content.

[0118] Step 290 : Average the account characteristics of the converted user accounts to obtain the average account characteristics of the converted user accounts.

[0119] In a possible implementation, the preference feature vector of the user account is converted and average pooling is performed to obtain an average preference feature vector on the user side.

[0120] Step 300 : Determine the real-time service preference characteristics of multimedia content based on the average account characteristics and the service preference characteristics.

[0121] In a possible implementation, the average preference feature vector is added to the preference feature vector of the multimedia content to obtain the real-time preference feature vector of the multimedia content.

[0122] Step 310: Update the conversion tendency parameter according to the real-time service preference characteristics.

[0123] In one possible implementation, the conversion tendency parameter is recalculated based on the real-time preference feature vector of the multimedia content to update the user group corresponding to the target user account, thereby improving the quality of the crowd targeting and achieving better multimedia content delivery effects.

[0124] In summary, the technical solution provided by the embodiment of the present application trains a crowd-targeting joint model by setting a sorting consistency constraint condition that the conversion data of the user account for the same multimedia content is positively correlated with the conversion tendency parameter, so that the crowd-targeting joint model only needs the service location and service content of the multimedia content to determine the service geographic characteristics and service preference characteristics of the multimedia content, and can combine the above-mentioned service geographic characteristics, service preference characteristics and account characteristics to predict the parameters for measuring the user's conversion of multimedia content, and then target the crowd and push the multimedia content according to the parameters. Through the above-mentioned sorting consistency constraint condition, new users can still perform relatively accurate crowd targeting for newly released multimedia content even when there is no historical data or the historical data is relatively sparse, thereby improving the quality of crowd targeting and reducing dependence on historical data, thereby improving the efficiency of multimedia content push, avoiding waste of computing resources, and alleviating equipment operation pressure.

[0125] In one example, if Figure 6 As shown, it exemplifies a process diagram for pushing multimedia content. The offline process (steps 1 to 4) is used to generate feature vectors related to users and multimedia content every day and write them into the data engine for fast retrieval. The online process (step 5) is used to: 1) generate crowd targeting when creating a local life service advertisement for the first time when multimedia content is published; 2) generate new crowd targeting conditions for the publisher of multimedia content being delivered during automatic updates.

[0126] Step 1: Obtain data logs of multimedia content.

[0127] In one possible implementation, the data log includes interaction data between user accounts and historical multimedia content, including exposure data, click data, and conversion data for the historical multimedia content. Exposure data may include the number of user accounts that were able to view the multimedia content. Click data may include the number of user actions on the multimedia content, such as clicks and drags. Conversion data may include data generated by user conversion behaviors, such as purchases, phone calls, inquiries, downloads, form submissions, and so on, after a user clicks on an ad.

[0128] In a possible implementation, data logs of historical multimedia content within a target period are extracted. Optionally, the target period is the most recent month.

[0129] The data log includes a first log record and a second log record. The first log record may be recorded in the format of (user account, multimedia content identification number, number of exposures, number of clicks, number of conversions). The second log record may be recorded in the format of (user account, multimedia content identification number, conversion time).

[0130] In some embodiments, the multimedia content is a local life service advertisement. Local life service advertisers, that is, local life service providers, can publish local life service advertisements in advertising delivery applications. In this case, records related to local life service advertisers in the past month can be extracted from the logs of advertising delivery applications, and the user account (user ID) and advertising identification number (advertising ID) can be used as key fields to aggregate the record data to obtain a first log record. The record format of the above-mentioned first log record can be (user ID, advertising ID, number of exposures, number of clicks, number of conversions). The above-mentioned number of exposures, number of clicks, and number of conversions can be used as value fields corresponding to the key fields.

[0131] At the same time, a conversion flow record of a historical advertisement (ie, a second log record) may also be extracted. The format of the conversion flow record may be (user ID, advertisement ID, conversion time).

[0132] Step 2: Generate account geographic features of the user account and service geographic features of the multimedia content.

[0133] The above-mentioned geographic features are used to measure the spatial proximity between the user account and the multimedia content. The account geographic features of the user account can be the geographic feature vector of the user account, and the service geographic features of the multimedia content can be the geographic feature vector of the multimedia content.

[0134] For a multimedia content j , its geographic feature vector g j Is a K-dimensional vector. Optionally, the above gj The mathematical expression (1) is as follows:

[0135] g j =[f(d(ω j ,ω1)),...,f(d(ω j ,ω K ))] T (1)

[0136] Where K represents the number of regions. For example, a city can be divided into K regions, namely service areas. In some embodiments, the default value of K is 50; ω j Represents multimedia content j The service location, such as ω j It is a two-dimensional coordinate composed of the longitude and latitude of the location; ω1 represents the geometric center coordinate of the first area, and the geometric center coordinate of the service location coordinates of all multimedia content in the Kth area is ω K .

[0137] In one possible implementation, given a set of all multimedia content released in a city in the past year, the K-means algorithm is used to cluster the multimedia content using the service location coordinates of the multimedia content as clustering features to obtain K subsets of multimedia content. Each subset is defined as an area, and the geometric center coordinates of the service location coordinates of all multimedia content in the Kth area are ω K .

[0138] d(*) represents the Euclidean distance, such as the Euclidean distance between the service location coordinates of the multimedia content and the geometric center coordinates of the region. j The kth value of the vector represents the user moving from the center of the kth region to the multimedia content l j The probability of the service location coordinates is calculated by the function f(*). The above probability can be the probability of the user accessing the above multimedia content, which is used to represent the possibility of the user accessing the multimedia content.

[0139] The above f(*) can be any function that can generate a probability value. In one possible implementation, the Pareto distribution is used to model the relationship between access probability and coordinate distance. Specifically, the geographic feature vector g of the media content is j The mathematical expression (2) is:

[0140] g j =[(1+d(ω j ,ω1)) -α ,...,(1+d(ω j ,ω K )) -α ]T (2)

[0141] Here, α is the shape parameter of the Pareto distribution, which can be obtained by utilizing the user's conversion sequence between different multimedia contents and based on the Maximum Likelihood Estimate (MLE) method.

[0142] For user u i The geographic feature vector g i , g i The mathematical expression (3) is:

[0143] g i =[γ i，1 , γ i，2 ,...,γ i，K ] T (3)

[0144] Among them, γ i，K To represent the expected number of conversions of the user to multimedia content in the Kth region, the geographic feature vector corresponding to the above user can be obtained by model training.

[0145] Step 3: Create a semantic graph of multimedia content.

[0146] Constructing a multimedia content semantic graph: To improve the expressive power of multimedia content in the feature space and accurately generate corresponding preference feature vectors for newly created multimedia content, it is necessary to construct a multimedia content semantic graph to express the semantic similarity of multimedia content. This semantic similarity is then used to generate corresponding preference feature vectors for newly created multimedia content. Each multimedia content semantic graph is a weighted bipartite graph.

[0147] In one possible implementation, five types of multimedia content semantic graphs need to be constructed, namely, content and content semantic graph, content and conversion time semantic graph, content and label semantic graph, content and region semantic graph, and content and neighboring user semantic graph.

[0148] 1. Constructing content and content semantic graph. In one possible implementation, the content and content semantic graph G LL The mathematical expression (4) is as follows:

[0149] G LL =(L, L, E LL , W LL ) (4)

[0150] The node sets on both sides of this bipartite graph are L, where L represents the set of all multimedia content. The edge set is E LLUsing the conversion flow records extracted above, if a user has conversion behaviors on two multimedia contents in a week, then an edge is established between the two multimedia content nodes. The weight of the edge is the number of times the two multimedia contents are continuously visited by the same user within a week. The weight set of the edge is W LL Content and content semantic graphs are used to represent the sequential access relationship between multimedia contents.

[0151] 2. Constructing a semantic graph of content and conversion time. In a possible implementation, the semantic graph of content and conversion time G LT The mathematical expression (5) is as follows:

[0152] G LT =(L, T, E LT , W LT ) (5)

[0153] One side of this bipartite graph is the multimedia content set L, and the other side is the conversion time set T accurate to the hour. Optionally, T has 24 nodes, each representing 24 hours. LT Indicates that multimedia content has conversion records within a certain hour, and the edge weight W LT Indicates the number of conversions of a multimedia content in a certain hour.

[0154] 3. Constructing a content and label semantic graph. In one possible implementation, the content and label semantic graph G LW The mathematical expression (6) is as follows:

[0155] G LW =(L, W, E Lw , W LW ) (6)

[0156] One side of this bipartite graph is the multimedia content set L, and the other side is the set of all semantic tags W. Optionally, a semantic tag can be a word contained in the textual copy of the service text associated with the multimedia content, or it can be the primary and secondary industry tags of the account that published the multimedia content. For example, if the primary industry tag is local life services, the corresponding secondary industry tags are sub-category tags within the primary industry, such as catering, hairdressing, and pets. Edge weights represent the number of times a word appears.

[0157] 4. Constructing content and region semantic graphs. In one possible implementation, the content and region semantic graph G LR The mathematical expression (7) is as follows:

[0158] G LR =(L, R, E LR , WLR )(7)

[0159] One side of this bipartite graph is the multimedia content set L, and the other side is the region set R, E LR is an edge set, where the edge represents the location information corresponding to a multimedia content belonging to a certain regional node. LR is a set of weights, all of which are 1.

[0160] 5. Constructing a semantic graph of content and neighboring users. In one possible implementation, the semantic graph of content and neighboring users G LU The mathematical expression (8) is as follows:

[0161] G LU =(L, U, E LU , W LU ) (8)

[0162] One side of this bipartite graph is the multimedia content set L, and the other side is the user account set U that has converted to multimedia content within 200 meters of the service location of the specified multimedia content during the target period (for example, the last month). The edge weight W LU Represents the number of conversions.

[0163] Step 4: Train the crowd orientation joint model

[0164] First, users and multimedia content need to be modeled in the feature space. This solution uses vector Represents user u i The preference feature vector of .

[0165] In one possible implementation, the user portrait stored in the application can be used as the user feature, and the user feature can be input into a pre-trained neural network model for feature extraction to obtain the user's preference feature vector. The above-mentioned pre-trained neural network model can be composed of a 3-layer neural network, and the parameters of the neural network model are training parameters learned by the model.

[0166] Multimedia content preference feature vector v j It is completely learned by the model. Then define the user's conversion tendency score for multimedia content as y ij , the above conversion propensity score y ij The mathematical expression (9) is as follows:

[0167] y ij =v i ·v j +g i ·g j (9)

[0168] This calculation method can simultaneously combine the geographic feature vectors and preference feature vectors corresponding to users and multimedia content, and automatically learn the importance of the similarity between preference feature space and geographic feature space in determining user conversion behavior.

[0169] Sorting consistency constraint: For a multimedia content, users with more conversions have higher conversion propensity scores than users with fewer conversions, which can be expressed as formula (10):

[0170]

[0171] in Represents a multimedia content conversion user u i A collection of Indicates multimedia content l j Conversions C i′j Less than user u i Number of conversions C ij User u i′ The set of Θ={v i , v j , g i |u i ∈U,l j ∈L} represents the set of training parameters.

[0172] P((y ij -y i′j )>0|Θ) represents user u i The conversion propensity score of user u i′ The probability of having a high conversion propensity score.

[0173] In a possible implementation, the above probability can be calculated using formula (11), which is as follows:

[0174]

[0175] and define is the loss function corresponding to the set L of all multimedia content, and the mathematical expression (12) is as follows:

[0176]

[0177] where λ||Θ|| 2 is the Gaussian prior parameter used for regularization. Because the loss function used in this embodiment It is defined by comparing the conversion propensity scores of different users. Compared with the classification model that can only use converted users as positive samples and non-converted users as negative samples, it can obtain more training samples and alleviate the problem of reduced model prediction ability caused by insufficient training samples.

[0178] Semantic graph loss condition in graph representation learning: For ease of expression, this embodiment uses formula (13) to represent one of the multiple multimedia content semantic graphs constructed in the above process. Optionally, formula (13) is as follows:

[0179] G LS =(L, S, E LS , W Ls ) (13)

[0180] Then, the LINE (Large-scale Information Network Embedding) model is used to minimize the KL divergence (Kullback-Leibler Divergence, relative entropy) between the empirical prediction probability and the model prediction probability.

[0181] Specifically, define the semantic node s k Can be used by multimedia content j The probability represented by Optionally, The mathematical expression (14) is as follows:

[0182]

[0183] where w jk is the weight of the edge. At the same time, the probability of using the model to express the preference feature vector using multimedia content and semantic nodes is p(s k |l j ). Optionally, p(s k |l j ) is expressed as follows:

[0184]

[0185] After simplifying the KL divergence formula, we can get the loss function representing a semantic graph of multimedia content: Optionally, the above The mathematical expression (16) is as follows:

[0186]

[0187] In this embodiment, because five kinds of multimedia content semantic graphs are used for representation learning, the objective function (i.e., loss function) used in the graph representation learning part is Optionally, The mathematical expression (17) is as follows:

[0188]

[0189] in It is a collection of five semantic nodes.

[0190] A crowd-targeted joint model is jointly trained based on sorting consistency constraints and semantic graph loss conditions: The multimedia content in the graph representation learning can be either already delivered multimedia content or newly created multimedia content. Newly created multimedia content contains textual material, regions, and neighboring users, so it can be represented in a biased feature space, namely, a preference feature vector. Furthermore, newly created multimedia content includes service location information, such as longitude and latitude coordinates, which can generate a representation of the newly created multimedia content in a geographic feature space, namely, a geographic feature vector.

[0191] This embodiment enables the loss function of the multimedia content semantic graph and the ranking consistency loss function to share the embedded features of the multimedia content in the preference space, achieving the purpose of joint training. The final loss function is Optionally, The mathematical expression (18) is as follows:

[0192]

[0193] Where β is the coefficient.

[0194] In one possible implementation, the parameters of the above model may be iteratively optimized using a stochastic gradient descent method.

[0195] Step 5: Model prediction and online recall of multimedia content.

[0196] The preference feature vectors and geographic feature vectors corresponding to the user and multimedia content generated in the above process can be written to the data engine on a daily basis. When the multimedia content publisher needs to perform audience targeting, they can first retrieve the geographic feature vectors and preference feature vectors corresponding to the multimedia content from the data engine. Then, using nearest neighbor retrieval technology, they can search the data engine for users with the highest conversion propensity scores to form a targeted audience of the size specified by the multimedia content publisher. Relevant tags are then added to the multimedia content and written into the multimedia content recall system. When a user in the targeted audience accesses Tencent traffic, the recall system can retrieve the user and be hit by the relevant tag. Then, the multimedia content associated with the tag is extracted as a candidate set for exposure, completing the recall process.

[0197] In some application scenarios, the solution provided by this embodiment can be applied to multimedia content delivery applications, such as advertising delivery applications. Optionally, a multimedia content designated population targeting module is created in the multimedia content delivery application, and this module can execute the solution provided by this embodiment. Taking the advertising delivery application as an example, as long as the advertiser specifies the location information corresponding to the local life service advertisement to be promoted and the advertising category information of the local life service advertisement, which can be the advertising copy or keywords, the population targeting for acquiring new customers can be extracted on the advertising management platform (Data Management Platform, DMP) for advertising delivery. In some embodiments, the above-mentioned advertising management platform can provide advertisers with different population extraction methods, such as basic attribute intersection and difference combination or historical advertising behavior population extraction method, which is mainly used for population targeting. After the advertiser obtains the population targeting from the DMP, it can bind different population targeting to different advertisements on the delivery end for advertising delivery.

[0198] If multimedia content publishers need to update the preference feature vectors of multimedia content on an hourly basis, they can first use the crowd extraction function for multimedia content to extract the users who converted within the last 24 hours. They can then retrieve the preference feature vectors of these users and perform an average pooling operation to obtain an average user preference feature vector. This average user preference vector is then added to the multimedia content preference vector to obtain a real-time multimedia content preference vector. This vector can then be used to search the data engine for targeted users.

[0199] In some application scenarios, the solution provided in this embodiment can also be used to provide automatic update functionality to local life service advertisers. Local life service advertisers that have already launched ads can use the automatic update feature of the ad placement application to modify the target audience of their ads on time, thereby improving the effectiveness of their placement.

[0200] In some application scenarios, the solution provided in this embodiment can be used to target local life service ads, thereby identifying the target user groups for local life service ads. Compared with other solutions, the proportion of new advertisers who fail to achieve advertising volume is reduced by 20%, and the overall service consumption of local life service ads is increased by 10%.

[0201] The joint population targeting model proposed in this embodiment can provide population targeting for local life service advertisers by combining ranking consistency and graph representation learning methods. Advertisers only need to specify the location information corresponding to the local life service to be promoted and the advertising category information of the local life service, which can be advertising copy or keywords. Then, they can extract the population targeting used to acquire new customers on the advertising management platform for advertising delivery. In this application scenario, the ranking consistency constraint means that the converted population of a local service advertisement is more likely to convert again than the non-converted population of the same advertisement. This ranking consistency constraint can provide more model training samples and alleviate the problem of sparse training samples for local life service advertisements. To measure the conversion tendency between users and local life service advertisements, this solution uses geographic feature vectors and preference feature vectors to compress and represent users and local life service advertisements. The geographic feature vector measures the influence of users and local life advertisements on different service areas in a city. The preference feature vector measures the similarity of conversion behavior between different users and different local life service advertisements. In order to learn the preference feature vectors of new local life service advertisements, this embodiment proposes constructing different weighted bipartite graphs to represent the different semantic information of advertisements, and then using a joint training method to integrate the semantic information of advertisements into the learning of the preference feature vector representation of advertisements.

[0202] Compared with the crowd targeting solutions of some related technologies, this embodiment can improve the local life service advertising delivery effect of both new and existing advertisers. It can not only improve the delivery effect of local life service advertisers with existing delivery records, but also enable new advertisers to generate crowd targeting for local life service advertisements, increase the speed of new advertisers' advertising growth, achieve the goal of rapid growth, and thus increase advertisers' delivery consumption on the advertising delivery platform.

[0203] The following are device embodiments of the present application, which can be used to implement the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0204] Please refer to Figure 7 , which shows a block diagram of a multimedia content push device provided by one embodiment of the present application. The device has the function of implementing the above-mentioned multimedia content push method, and the function can be implemented by hardware or by hardware executing corresponding software. The device can be a computer device or can be set in a computer device. The device 700 can include: a service information acquisition module 710, a service feature determination module 720, a conversion parameter prediction module 730, and a content push module 740.

[0205] The service information acquisition module 710 is configured to acquire service location information and service content information associated with multimedia content.

[0206] The service feature determination module 720 is configured to input the service location information and the service content information into a crowd-oriented joint model to obtain the service geographic features and service preference features of the multimedia content.

[0207] The conversion parameter prediction module 730 is configured to fuse the service geographic feature and the service preference feature with the account feature of each user account to obtain a conversion tendency parameter of each user account for the multimedia content.

[0208] The content push module 740 is configured to push the multimedia content according to the conversion tendency parameter.

[0209] The crowd-oriented joint model is trained based on a ranking consistency constraint, where the ranking consistency constraint means that the conversion data of the user account for the target multimedia content is positively correlated with the target conversion tendency parameter of the user account for the target multimedia content.

[0210] In an exemplary embodiment, the service feature determination module 720 includes: a service geographic feature determination unit, a semantic graph updating unit, and a service preference feature determination unit.

[0211] The service geographic feature determination unit is configured to compare the service location information with the area location information of each service area to obtain the service geographic feature.

[0212] A semantic graph updating unit is used to update at least one multimedia content semantic graph based on the service location information and the service content information to obtain an updated multimedia content semantic graph, wherein the at least one multimedia content semantic graph is used to represent the degree of association between each multimedia content and different semantic node sets.

[0213] The service preference feature determination unit is configured to determine the service preference feature of the multimedia content based on the graph structure data corresponding to the updated multimedia content semantic graph.

[0214] In an exemplary embodiment, the training process of the crowd-oriented joint model includes:

[0215] Obtaining a sample data log, the sample data log including a behavior data record of a sample user account for sample multimedia content, the behavior data record including tag information of the sample multimedia content, the tag information being conversion data of the sample user account for the sample multimedia content;

[0216] Determining, based on the sample data log, service information of the sample multimedia content, user portrait data of the sample user account, and at least one semantic graph of the multimedia content sample;

[0217] Based on the service information of the sample multimedia content, the user profile data of the sample user account, and the semantic graph of the at least one multimedia content sample, the crowd-oriented joint model is trained according to the joint constraint condition until an output result of the crowd-oriented joint model satisfies the joint constraint condition;

[0218] The joint constraint condition includes the sorting consistency constraint condition and the semantic graph loss condition.

[0219] In an exemplary embodiment, the apparatus 700 further includes: a log acquisition module, a semantic graph generation module, and an account feature determination module.

[0220] The log acquisition module is used to acquire data logs, where the data logs include behavioral data records of each user account for each multimedia content.

[0221] A semantic graph generation module is used to generate the at least one multimedia content semantic graph based on the behavior data record.

[0222] An account feature determination module is configured to determine the account features of each user account based on the data log and the at least one multimedia content semantic graph.

[0223] The at least one multimedia content semantic graph includes at least one of a content and content semantic graph, a content and conversion time semantic graph, a content and label semantic graph, a content and region semantic graph, and a content and neighboring user semantic graph.

[0224] In an exemplary embodiment, the account characteristics include account geographical characteristics and account preference characteristics, and the account characteristic determination module includes: an account preference characteristic determination unit and an account geographical characteristic determination unit.

[0225] An account preference feature determination unit is used to input the user portrait data of each user account into the crowd-oriented joint model for feature extraction processing to obtain the account preference features of each user account, wherein the user portrait data is generated based on the data log.

[0226] The account geographic feature determination unit is configured to input the graph structure data corresponding to the at least one multimedia content semantic graph into the crowd-oriented joint model to obtain the account geographic features of each user account.

[0227] In an exemplary embodiment, the conversion parameter prediction module 730 includes: a geographic feature parameter determination unit, a preference feature parameter determination unit, and a conversion tendency parameter determination unit.

[0228] The geographic feature parameter determination unit is configured to determine, for each user account, a geographic feature parameter based on the service geographic feature and the account geographic feature of the user account.

[0229] The preference characteristic parameter determining unit is configured to determine the preference characteristic parameter based on the service preference characteristic and the account preference characteristic of the user account.

[0230] The conversion tendency parameter determining unit is configured to determine a conversion tendency parameter of the user account for the multimedia content based on the geographic feature parameter and the preference feature parameter.

[0231] In an exemplary embodiment, the apparatus 700 further includes: a conversion account acquisition module, an average account feature determination module, a service preference feature update module, and a conversion tendency parameter update module.

[0232] A conversion account acquisition module, configured to acquire a conversion user account corresponding to the multimedia content;

[0233] an average account feature determination module, configured to average the account features of the converted user accounts to obtain the average account features of the converted user accounts;

[0234] a service preference feature updating module, configured to determine a real-time service preference feature of the multimedia content based on the average account feature and the service preference feature;

[0235] A conversion tendency parameter updating module is used to update the conversion tendency parameter according to the real-time service preference feature.

[0236] In summary, the technical solution provided by the embodiment of the present application trains a crowd-targeting joint model by setting a sorting consistency constraint condition that the conversion data of the user account for the same multimedia content is positively correlated with the conversion tendency parameter, so that the crowd-targeting joint model only needs the service location and service content of the multimedia content to determine the service geographic characteristics and service preference characteristics of the multimedia content, and can combine the above-mentioned service geographic characteristics, service preference characteristics and account characteristics to predict the parameters for measuring the user's conversion of multimedia content, and then target the crowd and push the multimedia content according to the parameters. Through the above-mentioned sorting consistency constraint condition, new users can still perform relatively accurate crowd targeting for newly released multimedia content even when there is no historical data or the historical data is relatively sparse, thereby improving the quality of crowd targeting and reducing dependence on historical data, thereby improving the efficiency of multimedia content push, avoiding waste of computing resources, and alleviating equipment operation pressure.

[0237] It should be noted that the apparatus provided in the above embodiments, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0238] Please refer to Figure 8 , which shows a block diagram of a computer device provided by an embodiment of the present application. The computer device may be a server for executing the above-mentioned method for pushing multimedia content. Specifically:

[0239] Computer device 800 includes a central processing unit (CPU) 801, a system memory 804 including a random access memory (RAM) 802 and a read-only memory (ROM) 803, and a system bus 805 connecting system memory 804 and CPU 801. Computer device 800 also includes a basic input / output system (I / O system) 806 that facilitates information transfer between various components within the computer, and a mass storage device 807 for storing an operating system 813, application programs 814, and other program modules 815.

[0240] The basic input / output system 806 includes a display 808 for displaying information and an input device 809, such as a mouse and keyboard, for user input. Both the display 808 and the input device 809 are connected to the central processing unit 801 via an input / output controller 810 connected to the system bus 805. The basic input / output system 806 may also include an input / output controller 810 for receiving and processing input from a variety of other devices, such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller 810 also provides output to a display screen, printer, or other types of output devices.

[0241] The mass storage device 807 is connected to the central processing unit 801 via a mass storage controller (not shown) connected to the system bus 805. The mass storage device 807 and its associated computer-readable media provide non-volatile storage for the computer device 800. In other words, the mass storage device 807 may include a computer-readable medium (not shown) such as a hard disk or a CD-ROM (Compact Disc Read-Only Memory) drive.

[0242] Without loss of generality, computer-readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other solid-state storage technologies, CD-ROM, DVD (Digital Video Disc) or other optical storage, tape cassettes, magnetic tape, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that computer storage media are not limited to the above-mentioned ones. The above-mentioned system memory 804 and mass storage device 807 can be collectively referred to as memory.

[0243] According to various embodiments of the present application, the computer device 800 may also be connected to a remote computer on a network such as the Internet for operation. That is, the computer device 800 may be connected to a network 812 via a network interface unit 811 connected to the system bus 805, or the network interface unit 811 may be used to connect to other types of networks or remote computer systems (not shown).

[0244] The memory further includes a computer program, which is stored in the memory and configured to be executed by one or more processors to implement the above-mentioned method for pushing multimedia content.

[0245] In an exemplary embodiment, a computer-readable storage medium is also provided, in which at least one instruction, at least one program, a code set or an instruction set is stored. When the at least one instruction, the at least one program, the code set or the instruction set is executed by a processor, the method for pushing the above-mentioned multimedia content is implemented.

[0246] Optionally, the computer-readable storage medium may include: ROM (Read Only Memory), RAM (Random Access Memory), SSD (Solid State Drives), or an optical disk, etc. Among them, the random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0247] In an exemplary embodiment, a computer program product or computer program is also provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the above-described method for pushing multimedia content.

[0248] It should be understood that the "multiple" mentioned in this article refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. In addition, the step numbers described in this article only illustrate a possible execution sequence between the steps. In some other embodiments, the above steps may not be executed in the order of the numbers, such as two steps with different numbers are executed at the same time, or two steps with different numbers are executed in the opposite order to the diagram. The embodiments of the present application do not limit this.

[0249] The above description is merely an exemplary embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for pushing multimedia content, characterized in that: The method comprises: Obtaining service location information and service content information associated with multimedia content; Inputting the service location information and the service content information into a crowd-oriented joint model to obtain service geographic features and service preference features of the multimedia content; fusing the service geographic feature and the service preference feature with the account feature of each user account to obtain a conversion tendency parameter of each user account for the multimedia content; pushing the multimedia content according to the conversion tendency parameter; Among them, the crowd-oriented joint model is trained based on a sorting consistency constraint condition, and the sorting consistency constraint condition means that the number of conversions corresponding to the conversion data of the user account for the target multimedia content is positively correlated with the target conversion tendency parameter of the user account for the target multimedia content.

2. The method according to claim 1, characterized in that Inputting the service location information and the service content information into a crowd-oriented joint model to obtain the service geographic features and service preference features of the multimedia content includes: Comparing the service location information with the regional location information of each service area to obtain the service geographical characteristics; updating at least one multimedia content semantic graph based on the service location information and the service content information to obtain an updated multimedia content semantic graph, wherein the at least one multimedia content semantic graph is used to represent the degree of association between each multimedia content and different semantic node sets; Based on the graph structure data corresponding to the updated multimedia content semantic graph, a service preference feature of the multimedia content is determined.

3. The method according to claim 2, characterized in that The training process of the crowd-oriented joint model includes: Obtaining a sample data log, the sample data log including a behavior data record of a sample user account for sample multimedia content, the behavior data record including tag information of the sample multimedia content, the tag information being conversion data of the sample user account for the sample multimedia content; Determining, based on the sample data log, service information of the sample multimedia content, user portrait data of the sample user account, and at least one semantic graph of the multimedia content sample; Based on the service information of the sample multimedia content, the user profile data of the sample user account, and the semantic graph of the at least one multimedia content sample, the crowd-oriented joint model is trained according to the joint constraint condition until an output result of the crowd-oriented joint model satisfies the joint constraint condition; The joint constraint condition includes the sorting consistency constraint condition and the semantic graph loss condition.

4. The method according to claim 2, characterized in that The method further comprises: Obtaining a data log, wherein the data log includes a behavior data record of each user account for each multimedia content; generating the at least one multimedia content semantic graph according to the behavior data record; determining account characteristics of each of the user accounts based on the data log and the at least one multimedia content semantic graph; The at least one multimedia content semantic graph includes at least one of a content and content semantic graph, a content and conversion time semantic graph, a content and label semantic graph, a content and region semantic graph, and a content and neighboring user semantic graph.

5. The method according to claim 4, characterized in that The account characteristics include account geographic characteristics and account preference characteristics. Determining the account characteristics of each user account based on the data log and the at least one multimedia content semantic graph includes: Inputting the user portrait data of each user account into the crowd-oriented joint model for feature extraction processing to obtain account preference features of each user account, wherein the user portrait data is generated based on the data log; The graph structure data corresponding to the at least one multimedia content semantic graph is input into the crowd-oriented joint model to obtain the account geographic features of each user account.

6. The method according to claim 5, characterized in that The fusing of the service geographic feature and the service preference feature with the account feature of each user account to obtain the conversion tendency parameter of each user account for the multimedia content includes: For each user account, determining a geographic feature parameter based on the service geographic feature and the account geographic feature of the user account; determining a preference characteristic parameter based on the service preference characteristic and the account preference characteristic of the user account; A conversion tendency parameter of the user account for the multimedia content is determined according to the geographic feature parameter and the preference feature parameter.

7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: Obtaining a conversion user account corresponding to the multimedia content; Averaging the account characteristics of the converted user accounts to obtain average account characteristics of the converted user accounts; determining a real-time service preference characteristic of the multimedia content based on the average account characteristic and the service preference characteristic; The conversion tendency parameter is updated according to the real-time service preference feature.

8. A multimedia content push device, characterized in that: The device comprises: A service information acquisition module, used to acquire service location information and service content information associated with multimedia content; A service feature determination module, configured to input the service location information and the service content information into a crowd-oriented joint model to obtain the service geographic features and service preference features of the multimedia content; a conversion parameter prediction module, configured to fuse the service geographic feature and the service preference feature with the account feature of each user account to obtain a conversion tendency parameter of each user account for the multimedia content; a content push module, configured to push the multimedia content according to the conversion tendency parameter; Among them, the crowd-oriented joint model is trained based on a sorting consistency constraint condition, and the sorting consistency constraint condition means that the number of conversions corresponding to the conversion data of the user account for the target multimedia content is positively correlated with the target conversion tendency parameter of the user account for the target multimedia content.

9. The device according to claim 8, characterized in that The service feature determination module includes: a service geographic feature determination unit, a semantic graph update unit, and a service preference feature determination unit; a service geographic feature determination unit, configured to compare the service location information with the regional location information of each service area to obtain the service geographic feature; a semantic graph updating unit, configured to update at least one multimedia content semantic graph based on the service location information and the service content information to obtain an updated multimedia content semantic graph, wherein the at least one multimedia content semantic graph is used to represent the degree of association between each multimedia content and different semantic node sets; The service preference feature determination unit is configured to determine the service preference feature of the multimedia content based on the graph structure data corresponding to the updated multimedia content semantic graph.

10. The device according to claim 9, characterized in that The device further comprises: A sample log acquisition module is configured to acquire a sample data log, wherein the sample data log includes a behavioral data record of a sample user account for sample multimedia content, wherein the behavioral data record includes tag information of the sample multimedia content, and the tag information is conversion data of the sample user account for the sample multimedia content; a sample information determination module, configured to determine, based on the sample data log, service information of the sample multimedia content, user portrait data of the sample user account, and at least one semantic graph of the multimedia content sample; A model training module is used to train the crowd-oriented joint model based on the service information of the sample multimedia content, the user portrait data of the sample user account and the semantic graph of at least one multimedia content sample, and in accordance with the joint constraint conditions until the output result of the crowd-oriented joint model meets the joint constraint conditions; wherein the joint constraint conditions include the sorting consistency constraint conditions and the semantic graph loss conditions.

11. The device according to claim 9, characterized in that The device further comprises: a log acquisition module, a semantic graph generation module and an account feature determination module; A log acquisition module, configured to acquire a data log, wherein the data log includes a behavior data record of each user account for each multimedia content; a semantic graph generation module, configured to generate a semantic graph of the at least one multimedia content according to the behavior data record; An account feature determination module is configured to determine the account features of each user account based on the data log and the at least one multimedia content semantic graph; wherein the at least one multimedia content semantic graph includes at least one of a content-to-content semantic graph, a content-to-conversion time semantic graph, a content-to-tag semantic graph, a content-to-region semantic graph, and a content-to-neighboring user semantic graph.

12. The device according to claim 11, characterized in that The account characteristics include account geographical characteristics and account preference characteristics, and the account characteristic determination module includes: an account preference characteristic determination unit and an account geographical characteristic determination unit; an account preference feature determination unit, configured to input the user portrait data of each user account into the crowd-oriented joint model for feature extraction processing to obtain the account preference features of each user account, wherein the user portrait data is generated based on the data log; The account geographic feature determination unit is configured to input the graph structure data corresponding to the at least one multimedia content semantic graph into the crowd-oriented joint model to obtain the account geographic features of each user account.

13. The device according to claim 12, characterized in that The conversion parameter prediction module includes: a geographical feature parameter determination unit, a preference feature parameter determination unit and a conversion tendency parameter determination unit; a geographic feature parameter determination unit, configured to determine, for each user account, a geographic feature parameter based on the service geographic feature and the account geographic feature of the user account; a preference feature parameter determining unit, configured to determine a preference feature parameter based on the service preference feature and the account preference feature of the user account; The conversion tendency parameter determining unit is configured to determine a conversion tendency parameter of the user account for the multimedia content based on the geographic feature parameter and the preference feature parameter.

14. The device according to any one of claims 8 to 13, characterized in that: The apparatus further comprises: a conversion account acquisition module, an average account feature determination module, a service preference feature update module, and a conversion tendency parameter update module; A conversion account acquisition module, configured to acquire a conversion user account corresponding to the multimedia content; an average account feature determination module, configured to average the account features of the converted user accounts to obtain the average account features of the converted user accounts; a service preference feature updating module, configured to determine a real-time service preference feature of the multimedia content based on the average account feature and the service preference feature; A conversion tendency parameter updating module is used to update the conversion tendency parameter according to the real-time service preference feature.

15. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the method for pushing multimedia content according to any one of claims 1 to 7.

16. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the method for pushing multimedia content according to any one of claims 1 to 7.

17. A computer program product comprising a computer program, characterized in that When the computer program is executed by an electronic device, the method for pushing multimedia content according to any one of claims 1 to 7 is implemented.

Images