A resource recommendation method and device, electronic equipment and storage medium

Abstract

The application discloses a resource recommendation method and device, electronic equipment and storage medium, the method comprises the following steps: taking target recommendation index data, object association data and resource association data as nodes, and constructing recommendation relationship information by taking the association relationship between the node corresponding data as an edge; inputting historical object resource data and the recommendation relationship information into an influence factor analysis model for data influence analysis to obtain an influence factor analysis result; the influence factor analysis result represents the influence distribution information of the historical object resource data on the target recommendation index data; and performing resource recommendation processing based on the influence factor analysis result. The method quantifies the influence of each influence data on the target recommendation index data, improves the accuracy and effectiveness of the influence factor analysis, and enables the influence factor analysis result to be applied to a resource recommendation business scenario, thereby improving the rationality of the resource recommendation business.

Description

Technical Field

[0001] This application relates to the field of recommendation technology, and in particular to a resource recommendation method, apparatus, electronic device, and storage medium. Background Technology

[0002] User experience evaluation is a technology that can improve product functions based on user operation data after product launch. In existing technologies, user experience analysis frameworks are based on indicator information, which cannot provide a connection between user experience data and the overall data of product application. Moreover, the number of indicators is often greater than the user experience data summarized by the actual product business. This leads to the results obtained after analyzing user experience being more idealized, resulting in low accuracy and effectiveness when analyzing the influencing factors of target recommendation indicator data. Summary of the Invention

[0003] This application provides a resource recommendation method, apparatus, electronic device, and storage medium that can improve the accuracy and effectiveness of influencing factor analysis.

[0004] On the one hand, this application provides a resource recommendation method, the method comprising:

[0005] Obtain historical object resource data corresponding to the target object, wherein the historical object resource data includes object association data of the target object and resource association data of historical multimedia resources recommended to the target object;

[0006] Recommendation relationship information is constructed by using target recommendation index data, object association data, and resource association data as nodes, and the association relationships between the corresponding data of the nodes as edges.

[0007] The historical object resource data and the recommendation relationship information are input into the influencing factor analysis model to perform data impact analysis, and the influencing factor analysis results are obtained. The influencing factor analysis results characterize the distribution information of the impact of the historical object resource data on the target recommendation index data.

[0008] Resource recommendation processing is performed based on the analysis results of the aforementioned influencing factors.

[0009] On the other hand, a resource recommendation device is provided, the device comprising:

[0010] The historical data acquisition module is used to acquire historical object resource data corresponding to the target object. The historical object resource data includes object association data of the target object and resource association data of historical multimedia resources recommended to the target object.

[0011] The recommendation relationship information construction module is used to construct recommendation relationship information by using target recommendation index data, the object association data and the resource association data as nodes and the association relationship between the corresponding data of the nodes as edges.

[0012] The data impact analysis module is used to input the historical object resource data and the recommendation relationship information into the impact factor analysis model to perform data impact analysis and obtain the impact factor analysis results. The impact factor analysis results characterize the impact distribution information of the historical object resource data on the target recommendation index data.

[0013] The resource recommendation module is used to perform resource recommendation processing based on the analysis results of the influencing factors.

[0014] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement a resource recommendation method as described above.

[0015] On the other hand, a computer-readable storage medium is provided, the storage medium including a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement a resource recommendation method as described above.

[0016] On the other hand, a computer program product is provided, including a computer program that, when executed by a processor, implements the resource recommendation method described above.

[0017] This application provides a resource recommendation method, apparatus, electronic device, and storage medium. The method constructs recommendation relationship information using target recommendation index data, object-related data, and resource-related data as nodes, and the relationships between corresponding data nodes as edges. Historical object resource data and recommendation relationship information are input into an influencing factor analysis model for data impact analysis, yielding influencing factor analysis results. These results characterize the distribution of the impact of historical object resource data on the target recommendation index data. Resource recommendation processing is then performed based on these influencing factor analysis results. This method quantifies the impact of each influencing data point on the target recommendation index data, thereby allowing the recommendation analysis results to be regressed to the influencing factor analysis results corresponding to different object groups. This improves the accuracy and effectiveness of the influencing factor analysis and enables the results to be applied to resource recommendation business scenarios, enhancing the rationality of resource recommendation services. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating an application scenario of a resource recommendation method provided in an embodiment of this application;

[0020] Figure 2 A flowchart illustrating a resource recommendation method provided in this application embodiment;

[0021] Figure 3 This is a flowchart illustrating the construction of recommendation relationship information in a resource recommendation method provided in an embodiment of this application;

[0022] Figure 4 This is a schematic diagram of recommendation relationship information in a resource recommendation method provided in an embodiment of this application;

[0023] Figure 5 A flowchart illustrating data impact analysis in a resource recommendation method provided in this application embodiment;

[0024] Figure 6 This is a schematic diagram showing the analysis results of influencing factors when the duration of an advertisement video is 0-20 seconds, as provided in an embodiment of this application for a resource recommendation method.

[0025] Figure 7 A flowchart illustrating the process of obtaining and fusing at least one model analysis result in a resource recommendation method provided in this application embodiment;

[0026] Figure 8 A flowchart illustrating feature construction in a resource recommendation method provided in this application embodiment;

[0027] Figure 9 A schematic diagram illustrating a feature construction method in a resource recommendation method provided in an embodiment of this application;

[0028] Figure 10 This application provides a flowchart illustrating the resource recommendation method that performs resource recommendation operations based on the results of influencing factor analysis.

[0029] Figure 11 A schematic diagram of the module executing the resource recommendation method provided in an embodiment of this application;

[0030] Figure 12 This is a schematic diagram of the structure of a resource recommendation device provided in an embodiment of this application;

[0031] Figure 13 This is a schematic diagram of the hardware structure of a device for implementing the method provided in the embodiments of this application. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0033] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. Furthermore, the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein.

[0034] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0035] Please see Figure 1 This illustration shows an application scenario diagram of a resource recommendation method provided in this application embodiment. The application scenario includes a client 110 and a server 120. The server 120 obtains historical object resource data corresponding to the client 110. The historical object resource data includes object association data of the target object and resource association data of historical multimedia resources recommended to the target object. The server 120 constructs recommendation relationship information using target recommendation index data, object association data, and resource association data as nodes and the relationship between corresponding data of nodes as edges. The server 120 inputs the historical object resource data and recommendation relationship information into an influencing factor analysis model to perform data influence analysis, obtains the influencing factor analysis results, and performs resource recommendation processing based on the influencing factor analysis results.

[0036] In this embodiment, client 110 includes physical devices such as smartphones, desktop computers, tablets, laptops, digital assistants, smart wearable devices, and in-vehicle terminals, and may also include software running on the physical device, such as applications. The operating system running on the physical device in this embodiment may include, but is not limited to, Android, iOS, Linux, Unix, and Windows. Client 110 includes a UI (User Interface) layer, through which it provides multimedia resource display and log data collection. Additionally, it sends historical object resource data required for data analysis to server 120 based on API (Application Programming Interface).

[0037] In this embodiment, the server 120 may include a standalone server, a distributed server, or a server cluster consisting of multiple servers. The server 120 may include a network communication unit, a processor, and a memory, etc. Specifically, the server 120 can be used to construct recommendation relationship information and input historical object resource data and recommendation relationship information into an influencing factor analysis model for data influence analysis to obtain the influencing factor analysis results.

[0038] Please see Figure 2 It demonstrates a resource recommendation method that can be applied to the server side, and the method includes:

[0039] S210. Obtain historical object resource data corresponding to the target object. The historical object resource data includes object association data of the target object and resource association data of historical multimedia resources recommended to the target object.

[0040] In some embodiments, object association data can be inherent association data of the target object, such as the target object's age, gender, etc. Resource association data can include the target object's interactive operation data on historical multimedia resources and the inherent association data of historical multimedia resources. Interactive operation data can include video playback duration, playback completeness, likes, forwards, shares, comments, etc., and inherent association data can include the category, content, etc. of historical multimedia resources.

[0041] S220. Using target recommendation index data, object association data, and resource association data as nodes, and the association relationships between corresponding data of nodes as edges, recommendation relationship information is constructed.

[0042] In some embodiments, the association can be a causal relationship. A causal relationship means that among two nodes connected by a directed line in the recommendation relationship information, the node originating from the directed line has a causal effect on the node reaching the directed line; that is, changes in the node originating from the directed line will affect changes in the node reaching the directed line. Specifically, there can be a causal relationship between object-related data and resource-related data, or between object-related data and target recommendation metric data, or between resource-related data and target recommendation metric data. Therefore, both object-related data and resource-related data in the recommendation relationship information have a direct or indirect causal relationship with the target recommendation metric data.

[0043] In some embodiments, see Figure 3 Using target recommendation metric data, object association data, and resource association data as nodes, and the relationships between corresponding data points as edges, the recommendation relationship information is constructed, including:

[0044] S310. Perform recommendation analysis on the recommendation results corresponding to historical object resource information to obtain the unverified impact data associated with the target recommendation index data. The unverified impact data includes object association data that affects the target recommendation index data and resource association data that affects the target recommendation index data.

[0045] S320. Determine the first association between the target recommendation indicator data and the impact data to be verified;

[0046] S340. Determine the second association relationship between object-related data and resource-related data;

[0047] S350. Using target recommendation index data, object association data, and resource association data as nodes, and the first association relationship and the second association relationship as edges, the recommendation relationship information is constructed.

[0048] In some embodiments, when performing recommendation analysis on the recommendation results corresponding to historical object resource information, the historical object resource information in the target application can be compared horizontally with the historical object resource information in other applications of the same type, excluding the target application, to obtain the differences between the historical object resource information in the target application and the historical object resource information in other applications. By analyzing these differences, the unverified impact data related to the target recommendation indicator data can be obtained.

[0049] For example, when conducting recommendation analysis on advertising resources, we analyze the differences between the category, duration, and time sequence exposure distribution of application software 1 and application software 3. We obtain the user consumption distribution across application software 1, application software 2, and application software 3, which is historical object resource data. Comparing the user consumption distribution across these three platforms reveals differences in user consumption distribution. These differences include the fact that in the overall category exposure distribution of application software 1, the proportion of knowledge-related categories is significantly higher than other categories. Simultaneously, comparing application software 2 and application software 3 horizontally, the category exposure distribution of application software 1 shows significant differences. In some categories, such as games, the proportion of application software 1 differs considerably from its competitors. Furthermore, in the overall ad duration exposure distribution of application software 1, ads with a duration of 0-20 seconds are significantly less frequent than those in application software 2 and application software 3, while ads with a duration of 40-60 seconds are significantly more frequent than those in application software 1, whereas the ad duration exposure distributions of application software 2 and application software 3 are relatively similar. Therefore, it can be determined that, overall, factors such as ad category, ad duration, and the distribution of ad duration before and after the user's video consumption time series have an impact on the user's ad consumption, which means that the impact data related to the target recommendation index data has been obtained and needs to be verified.

[0050] In some embodiments, there is a direct causal relationship between the data to be verified and the target recommendation metric data. Changes in the data to be verified directly affect changes in the target recommendation metric data; this is the first association between the target recommendation metric data and the data to be verified. Since a second association exists between object-related data and resource-related data, and the data to be verified is essentially the object-related data and resource-related data that affect the target recommendation metric data, other object-related data or other resource-related data, besides the data to be verified, may have an indirect causal relationship with the target recommendation metric data, indirectly affecting its changes. Furthermore, when there are multiple data points to be verified, a second association also exists between each pair of data points.

[0051] Therefore, using target recommendation metric data, object-related data, and resource-related data as nodes, and the first and second association relationships as edges, recommendation relationship information is constructed. This recommendation relationship information is a causal relationship graph structure, which includes object-related data and resource-related data that have direct and indirect causal relationships with the target recommendation metric data. Nodes related to business experience data, such as video supply distribution and distribution strategies, can also be added to the recommendation relationship information.

[0052] In some embodiments, see Figure 4 ,like Figure 4 The diagram illustrates recommendation relationship information. In this diagram, the target recommendation metric is the average ad completion rate, the associated data includes user distribution, and the associated resource data includes ad video duration and video content. Furthermore, based on business experience data, two additional nodes—supply release and distribution strategy—are added. Based on this diagram, it can be determined that ad video duration distribution and user distribution are data that directly affect the average ad completion rate. User distribution also affects ad video duration distribution; that is, user distribution is a common factor in both ad video duration distribution and average ad completion rate. When determining the impact of user distribution on average ad completion rate, the ad video duration distribution cannot be kept constant, otherwise the calculation of the impact of distribution on average ad completion rate will be inaccurate. In addition, video content also affects ad video duration distribution; that is, video content is data that indirectly affects the average ad completion rate. When verifying the impact of ad video duration distribution on average ad completion rate, video content can also interfere with the verification results. Therefore, historical object resource data can be classified based on recommendation relationship information to determine the historical object resource data associated with the target recommendation indicator data.

[0053] By constructing recommendation relationship information through the causal relationships between the data corresponding to nodes, common causes and interfering factors can be identified based on the recommendation relationship information. This facilitates the data impact analysis of the influencing factor analysis model and improves the accuracy of the data impact analysis.

[0054] S230. Input historical object resource data and recommendation relationship information into the influencing factor analysis model to perform data impact analysis, and obtain the influencing factor analysis results. The influencing factor analysis results characterize the distribution information of the impact of historical object resource data on the target recommendation index data.

[0055] In some embodiments, based on recommendation relationship information, it is possible to determine which variables need to be controlled in the information input into the influencing factor analysis model, thereby grouping historical object resource data.

[0056] In some embodiments, the grouped historical object resource data is input into the influencing factor analysis model. Based on the influencing factor analysis model, data influence analysis is performed on the impact data to be verified corresponding to the historical object resource data to obtain the influencing factor analysis results. The influencing factor analysis results characterize the distribution information of the influence of historical object resource data in different groups on the target recommendation index data, that is, the degree of influence of the impact data to be verified on the target recommendation index data.

[0057] In some embodiments, see Figure 5The influencing factor analysis model includes an interference determination module and an impact analysis module. Historical object resource data and recommendation relationship information are input into the influencing factor analysis model for data impact analysis. The results of the influencing factor analysis include:

[0058] S510. Input the recommendation relationship information into the interference determination module, and determine the interference nodes corresponding to the associated nodes of the target nodes based on the path information between each node in the recommendation relationship information and the target nodes corresponding to the target recommendation index data.

[0059] S520. Input the associated nodes, interference nodes, and historical object resource data into the impact analysis module, perform data impact analysis on the historical object resource data, and obtain the results of the impact factor analysis.

[0060] In some embodiments, based on recommendation relationship information, the associated nodes of the target node corresponding to the target recommendation indicator data can be determined, that is, object-related data or resource-related data that directly affect the target recommendation indicator data. When determining the interfering nodes corresponding to the associated nodes, the interfering nodes can be determined based on the path information between each node in the recommendation relationship information and the target node corresponding to the target recommendation indicator data. If the path information from a certain node to the target node overlaps with the path information from the associated node to the target node, or if there is path overlap or node overlap, the interfering node corresponding to the associated node can be determined. The interfering node is the node that will affect the associated node when determining the target recommendation indicator data.

[0061] For example, if node A can reach node C via node B, and node B can directly reach node C, then the path from node B to node C is an overlapping path. In this case, when node B is the associated node and node C is the target node, node A is the interfering node. Similarly, if node D can reach node F via node E, and node D can also directly reach node F, then node D is an overlapping node for both paths. In this case, when node D is the associated node and node F is the target node, node E is the interfering node.

[0062] When classifying historical object resource data, it is necessary to process the object-related data or resource-related data corresponding to the interfering node. For example, the object-related data or resource-related data corresponding to the interfering node can be kept unchanged, or other methods can be used to remove the interference. Then, based on the object-related data or resource-related data corresponding to the associated node, the historical object resource data can be classified to avoid the influence of the interfering node on the associated node.

[0063] In some embodiments, such as Figure 4In the causal relationship diagram shown, there is an ad video duration distribution along the path from video content to the average completion rate of an ad. Therefore, video content is an interference node of the associated node corresponding to the ad video duration distribution. When determining the impact of the ad video duration distribution on the average completion rate of an ad, it is necessary to keep the video content unchanged. For example, historical object resource data can be classified based on video content to obtain multiple categories such as games and lifestyle. In each category, data related to the ad video duration distribution can be obtained as historical object resource data.

[0064] Furthermore, the ad video duration distribution is located on the path from the user distribution back to the average ad completion rate. The ad video duration distribution is an interference node of the related node corresponding to the user distribution. However, since the user distribution affects the ad video duration distribution, and the ad video duration distribution in turn affects the average ad completion rate, when determining the impact of the user distribution on the average ad completion rate, although the ad video duration distribution is an interference node, it cannot be controlled to remain unchanged. Otherwise, the calculation of the impact of the user distribution on the average ad completion rate will be inaccurate.

[0065] In some embodiments, historical object resource data may include at least two sets of data. Historical object resource data may be a binary data set of experimental group data and control group data, which can be used to verify the influencing factors of a single dimension. Target control resource data may also be a multivariate data set of multiple sets of experimental group data and control group data, or a multivariate data set corresponding to continuous user segments, which can be used to verify the influencing factors of multiple dimensions.

[0066] In some embodiments, the influencing factor analysis model may include various models such as causal forest, X-learner, S-learner, ATE-IPTW, etc., which can introduce more dimensional and information-rich features into the model and continuously iterate and optimize the influencing factor analysis model.

[0067] When using the X-learner model for data impact analysis, the impact analysis module performs function fitting on the control group data in the historical object resource data based on the base learner. This yields the function corresponding to the control group data and its result, which represents the increment of the target recommendation index data. Similarly, the experimental group data in the historical object resource data is fitted with the same function based on the base learner, yielding the corresponding function and its result, which also represents the increment of the target recommendation index data. Based on the function corresponding to the control group data, a first estimated information is calculated for the experimental group data. This first estimated information is then subtracted from the function result for the experimental group data to obtain the first output result. A second estimated information is calculated based on the function corresponding to the control group data. This second estimated information is then subtracted from the function result for the control group data to obtain the second output result. A weighted average of the first and second output results yields the Conditional Average Treatment Effect (CATE) function, which can generate an image corresponding to the Shapley Additive Explanation (Shap) tool. Therefore, the X-learner model can use the observed sample results to predict the unobserved sample results, approximate the increment of the target recommendation index data, and adjust the bias weights of the first and second output results to optimize the approximate results.

[0068] In some embodiments, when using a causal classification model for data impact analysis, a decision tree can be used to group data in the impact analysis module. Then, for each leaf, the average value of the experimental group data is subtracted from the average value of the control group data to obtain the CATE result.

[0069] In some embodiments, when using ATE-IPTW to perform data impact analysis in the impact analysis module, the impact analysis module can assign a probability of acceptance of processing to each historical object resource data, and then weight the results corresponding to each historical object resource data according to its opposite probability. That is, for historical object resource data that is actually processed, the historical object resource data that is predicted to be unprocessed will be given a greater weight than the historical object resource data that is predicted to be processed.

[0070] In some embodiments, when the historical object resource data is an advertisement video, the following table shows the analysis results of the factors affecting the average completion rate of advertisements by users of different age groups when the advertisement video duration is 0-20s.

[0071]

[0072]

[0073] The table shows that when the ad video length is 0-20 seconds, the overall ad completion rate increases. Among them, the increase in completion rate is greater for female users than for male users. Furthermore, the increase in completion rate is highest among women aged 36-45.

[0074] Please see Figure 6 ,like Figure 6 The image shows a SHA-T plot corresponding to the influencing factor analysis results when the historical object resource data is an advertisement video with a duration of 0-20 seconds. Based on Figure 6 The analysis of influencing factors shows that for lifestyle-related advertising videos, such as... Figure 6 As shown in the diagram, increasing the proportion of cat-related videos (0-20 seconds) improves the average completion rate of user ads. However, for game-related videos, this proportion decreases the average completion rate. Female users show a greater interest in 0-20 second ads, while for users over 55 years old, increasing the proportion of 0-20 second ads improves the average completion rate. Therefore, based on the above analysis of influencing factors, adjustments can be made to the resource recommendation process.

[0075] Interference nodes can be identified based on recommendation relationship information. Then, an influencing factor analysis model is used to conduct data impact analysis on historical object resource data to obtain the influencing factor analysis results. By eliminating interference factors in the environment and making causal inferences on the correlation between the data to be verified and the target recommendation index data, the causal relationship between the related data can be verified, thereby improving the accuracy and effectiveness of data impact analysis.

[0076] In some embodiments, see Figure 7 Historical object resource data is input into the influencing factor analysis model for data impact analysis. The results of the influencing factor analysis include:

[0077] S710. Based on the data types in historical object resource data and the preset data impact analysis scenario, determine at least one target impact factor analysis model from multiple preset impact factor analysis models;

[0078] S720. Input historical object resource data into at least one target influencing factor analysis model to perform data impact analysis, and obtain the model analysis results corresponding to each target influencing factor analysis model;

[0079] S730. Perform data fusion processing on the analysis results of each model to determine the analysis results of influencing factors.

[0080] In some embodiments, the preset influencing factor analysis model may include various models such as causal forest, X-learner, and ATE-IPTW. Among them, causal forest is a non-parametric method, and its effectiveness does not deteriorate with the increase of variable dimensionality. Therefore, when the historical object resource data is a multidimensional data set, the causal forest model can be used for data influence analysis, and the asymptotic distribution of the historical object resource data and the construction of confidence intervals can be obtained through the causal forest model.

[0081] The X-learner model remains effective even when the experimental group data is significantly larger than the control group data, and it leverages the functional structure of CATE. The X-learner model can quickly generate CATE results and corresponding images from the Shapley Additive Explanation (Shap). ATE-IPTW can aggregate similar objects even when object association data is missing.

[0082] Therefore, based on the data types in historical object resource data and the pre-defined data impact analysis scenario, at least one target impact factor analysis model can be determined from multiple impact factor analysis models. For example, if the data types in the historical object resource data do not include object-related data types, the model corresponding to the ATE-IPTW matching method can be used. If the data impact analysis scenario requires CATE results and shap diagrams, the X-learner model can be used. By combining multiple impact factor analysis models and aggregating the model analysis results corresponding to different impact factor analysis models, the impact factor analysis results are obtained.

[0083] In some embodiments, multiple influencing factor analysis models can be integrated into a causal inference toolbox. The causal inference toolbox includes three models: data input, model training, and result output. It can determine the applicable influencing factor analysis model for input historical object resource data, perform causal inference processing, and obtain influencing factor analysis results.

[0084] In some embodiments, the target influencing factor analysis model includes a target interference determination module and a target influence analysis module. Recommendation relationship information is input into the target interference determination module. Based on the path information between each node in the recommendation relationship information and the target node corresponding to the target recommendation index data, the interference nodes corresponding to the associated nodes of the target node can be determined. The associated nodes, interference nodes, and historical object resource data are input into the target influencing factor analysis module. Data influence analysis is performed on the historical object resource data to obtain the model analysis results corresponding to the target influencing factor analysis model. See the embodiments corresponding to the above-described method of inputting historical object resource data and recommendation relationship information into the influencing factor analysis model to perform data influence analysis and obtain the influencing factor analysis results.

[0085] By aggregating commonly used influencing factor analysis models, corresponding influencing factor analysis models can be obtained for analysis based on historical object resource data with different structures. This can improve the efficiency of using influencing factor analysis models and make them easier to apply in data impact analysis.

[0086] In some embodiments, see Figure 8 After determining the historical object resource data associated with the target recommendation indicator data from the historical object resource data based on recommendation relationship information, the method further includes:

[0087] S810. Perform feature construction processing on historical object resource data at least once to obtain at least one target feature data;

[0088] Historical object resource data is input into the influencing factor analysis model for data impact analysis. The results of the influencing factor analysis include:

[0089] S820. Input at least one target feature data into the influencing factor analysis model to perform data influence analysis and obtain the influencing factor analysis results.

[0090] In some embodiments, based on the learning objectives and business logic required by the influencing factor analysis model, at least one feature construction process is performed on historical object resource data to obtain at least one target feature data, which is a feature useful for model prediction. See also Figure 9 ,like Figure 9The diagram illustrates various feature construction methods. During feature construction, historical object resource data can be processed using techniques such as aggregate feature construction, simple transformation feature construction, Cartesian product feature construction, genetic programming feature construction, Gradient Boosting Decision Tree (GBDT) feature construction, clustering feature construction, temporal feature construction, time series feature construction, spatial feature construction, text feature construction, and automated feature construction. This process can create new features that are useful for model prediction, thus obtaining the target feature data.

[0091] The target feature data obtained after feature construction is then input into the influencing factor analysis model for data impact analysis to obtain the influencing factor analysis results.

[0092] After performing feature construction processing on historical object resource data, the feature dimensions and information content can be increased, which is beneficial for the influencing factor analysis model to make model predictions, thereby improving the effectiveness of data influence analysis.

[0093] S240. Perform resource recommendation processing based on the results of the influencing factor analysis.

[0094] In some embodiments, based on the results of the influencing factor analysis, the degree of influence of each data point to be verified on the target recommendation index data can be quantified, thereby adjusting the resource recommendation processing business as a whole to perform resource recommendation processing.

[0095] In some embodiments, see Figure 10 The resource recommendation operation based on the results of the influencing factor analysis also includes:

[0096] S1010. Perform the target resource recommendation operation corresponding to the influencing factor analysis results on the sample objects associated with the target object to obtain the sample recommendation results;

[0097] S1020. Based on the sample recommendation results, verify the results of the influencing factor analysis;

[0098] S1030. If the results of the influencing factor analysis are verified, perform the recommended actions based on the results of the influencing factor analysis.

[0099] In some embodiments, sample objects associated with the target object are determined, and the sample objects can be a preset number of target objects. Based on the results of the influencing factor analysis, the resource recommendation processing is adjusted, and the target resource recommendation operation corresponding to the results of the influencing factor analysis is performed on the sample objects to obtain the sample recommendation results.

[0100] When validating the influencing factor analysis results based on sample recommendation results, the first degree of change in the target recommendation indicator data corresponding to the influencing factor analysis results is compared with the second degree of change in the target recommendation indicator data corresponding to the sample recommendation results. If the trends of the first and second degrees of change are consistent—that is, if the first degree of change indicates an increase in the target recommendation indicator data, and the second degree of change also indicates an increase in the target recommendation indicator data—then the target resource recommendation operation corresponding to the influencing factor analysis results has a positive impact on the target recommendation indicator data, and the influencing factor analysis results pass the validation. If the trends of the first and second degrees of change are inconsistent—that is, if the first degree of change indicates an increase in the target recommendation indicator data, but the second degree of change indicates a decrease in the target recommendation indicator data—then the target resource recommendation operation corresponding to the influencing factor analysis results has a negative impact on the target recommendation indicator data, and the influencing factor analysis results fail the validation.

[0101] If the results of the influencing factor analysis are verified, recommendation operations can be performed based on the results of the influencing factor analysis in actual resource recommendation operations.

[0102] Validating the results of the influencing factor analysis, and obtaining consistent conclusions at both the statistical and practical application levels, allows for adjustments to the resource recommendation business, thereby enhancing the credibility of the analysis results.

[0103] In some embodiments, see Figure 11 ,like Figure 11 As shown, in the scenario of applying this resource recommendation method to advertising video recommendations, the modules that execute this method can include a testing module, a causal graph model, a data module, a model module, and a testing module. In the testing module, an experimental questionnaire can be designed and tested with target users. After collecting and analyzing the experimental results, the data to be verified is determined, thus clarifying the analysis object and leading to the causal graph module. In the causal graph module, based on the first association between the data to be verified and the target recommendation index data, and the second association between the object association data and the resource association data, the nodes corresponding to the object association data, resource association data, and target recommendation index data are connected to construct the recommendation relationship information, i.e., the causal graph is constructed. In the data module, multiple application data, i.e., historical object resource data, can be obtained. After filtering out abnormal data in these application data, feature construction processing is performed to generate a feature table, i.e., historical object resource data. The model module integrates a causal recommendation toolkit, including multiple influencing factor analysis models. The historical object resource data and recommendation relationship information are input into the influencing factor analysis models in the model module to obtain the influencing factor analysis results.

[0104] In the testing module, micro-testing experiments can be designed to verify whether the target resource recommendation operations corresponding to the influencing factor analysis results can improve the target recommendation metric data. If the verification is successful, A / B testing can be conducted based on the influencing factor analysis results for further verification. The test results in the first scenario and the second scenario can be compared to determine whether the target resource recommendation operations corresponding to the influencing factor analysis results need to be adopted to adjust the resource recommendation business. For example, if the target audience is divided into users in scenario A and users in scenario B, where scenario A uses the target resource recommendation operations corresponding to the influencing factor analysis results and the user conversion rate is 30%, and scenario B does not use the target resource recommendation operations corresponding to the influencing factor analysis results and the user conversion rate is 20%, then the influencing factor analysis results can improve the target recommendation metric data, and the influencing factor analysis results can be determined to have passed verification.

[0105] In some embodiments, data modules, model modules, and data transfer between data models and model modules can be aggregated into a pipeline. For tasks at a preset level, only the hyperparameters corresponding to each module need to be configured to automatically perform data impact analysis.

[0106] This application provides a resource recommendation method, which includes: constructing recommendation relationship information using target recommendation index data, object association data, and resource association data as nodes, and using the association relationships between corresponding data nodes as edges; inputting historical object resource data and recommendation relationship information into an influencing factor analysis model for data impact analysis to obtain influencing factor analysis results, which characterize the distribution information of the impact of historical object resource data on target recommendation index data; and performing resource recommendation processing based on the influencing factor analysis results. This method quantifies the impact of each influencing data point on the target recommendation index data, thereby allowing the recommendation analysis results to be regressed to the influencing factor analysis results corresponding to different object groups. This improves the accuracy and effectiveness of influencing factor analysis and enables the influencing factor analysis results to be applied to resource recommendation business scenarios, thus enhancing the rationality of resource recommendation business.

[0107] This application also provides a resource recommendation device; please refer to [link to relevant documentation]. Figure 12 The device includes:

[0108] The historical data acquisition module 1210 is used to acquire historical object resource data corresponding to the target object. The historical object resource data includes object association data of the target object and resource association data of historical multimedia resources recommended to the target object.

[0109] The recommendation relationship information construction module 1220 is used to construct recommendation relationship information by using target recommendation index data, object association data and resource association data as nodes and the association relationship between corresponding data of nodes as edges.

[0110] The data impact analysis module 1230 is used to input historical object resource data and recommendation relationship information into the influencing factor analysis model to perform data impact analysis and obtain the influencing factor analysis results. The influencing factor analysis results characterize the distribution information of the impact of historical object resource data on the target recommendation index data.

[0111] The resource recommendation module 1240 is used to perform resource recommendation processing based on the results of influencing factor analysis.

[0112] In some embodiments, the influencing factor analysis model includes an interference determination module and an impact analysis module, wherein the data impact analysis module includes:

[0113] The first interference determination unit is used to input the recommendation relationship information into the interference determination module, and determine the interference nodes corresponding to the associated nodes of the target nodes based on the path information between each node in the recommendation relationship information and the target nodes corresponding to the target recommendation index data.

[0114] The first impact analysis unit is used to input associated nodes, interference nodes, and historical object resource data into the impact analysis module, perform data impact analysis on the historical object resource data, and obtain the results of the impact factor analysis.

[0115] In some embodiments, the data impact analysis module includes:

[0116] The target model determination unit is used to determine at least one target influencing factor analysis model from multiple preset influencing factor analysis models based on the data types in historical object resource data and preset data influence analysis scenarios.

[0117] The model analysis result acquisition unit is used to input historical object resource data and recommendation relationship information into at least one target influencing factor analysis model to perform data influence analysis and obtain the model analysis results corresponding to each target influencing factor analysis model.

[0118] The data fusion unit is used to perform data fusion processing on the analysis results of each model to determine the analysis results of influencing factors.

[0119] In some embodiments, the target influencing factor analysis model includes a target interference determination module and a target influence analysis module, and the model analysis result acquisition unit includes:

[0120] The second interference determination unit is used to input the recommendation relationship information into the target interference determination module, and determine the interference nodes corresponding to the associated nodes of the target nodes based on the path information between each node in the recommendation relationship information and the target nodes corresponding to the target recommendation index data.

[0121] The second impact analysis unit is used to input associated nodes, interference nodes, and historical object resource data into the target impact factor analysis module, perform data impact analysis on the historical object resource data, and obtain the model analysis results corresponding to the target impact factor analysis model.

[0122] In some embodiments, the resource recommendation module includes:

[0123] The sample recommendation result acquisition unit is used to perform target resource recommendation operations corresponding to the influencing factor analysis results on the sample objects associated with the target object, and obtain the sample recommendation results;

[0124] The validation unit is used to validate the results of the influencing factor analysis based on the sample recommendation results.

[0125] The execution unit is used to perform the recommended actions based on the results of the influencing factor analysis, provided that the results of the influencing factor analysis have been validated.

[0126] In some embodiments, the recommendation relationship information construction module includes:

[0127] The recommendation analysis unit is used to perform recommendation analysis on the recommendation results corresponding to historical object resource information to obtain the unverified impact data associated with the target recommendation index data. The unverified impact data includes object association data that affects the target recommendation index data and resource association data that affects the target recommendation index data.

[0128] The first correlation determination unit is used to determine the first correlation between the target recommendation indicator data and the impact data to be verified.

[0129] The second association relationship determination unit is used to determine the second association relationship between object association data and resource association data;

[0130] The recommendation relationship information construction unit is used to construct recommendation relationship information by using target recommendation index data, object association data, and resource association data as nodes and the first association relationship and the second association relationship as edges.

[0131] In some embodiments, the device further includes:

[0132] The feature construction unit is used to perform feature construction processing on historical object resource data at least once to obtain at least one target feature data.

[0133] The data impact analysis module includes:

[0134] The feature analysis unit is used to input at least one target feature data into the influencing factor analysis model for data analysis and obtain the influencing factor analysis results.

[0135] The apparatus provided in the above embodiments can execute the methods provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in the above embodiments can be found in a resource recommendation method provided in any embodiment of this application.

[0136] This embodiment also provides a computer-readable storage medium storing computer-executable instructions, which are loaded by a processor and executed by the resource recommendation method described above in this embodiment.

[0137] This embodiment also provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations recommended in the above resources.

[0138] This embodiment also provides an electronic device, which includes a processor and a memory, wherein the memory stores a computer program adapted to be loaded by the processor and executed as described above in this embodiment of a resource recommendation method.

[0139] The device may be a computer terminal, a mobile terminal, or a server, and may also participate in constituting the apparatus or system provided in the embodiments of this application. For example... Figure 13 As shown, server 13 may include one or more processors 1302 (shown as 1302a, 1302b, ..., 1302n in the figure) (processor 1302 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 1304 for storing data, and a transmission device 1306 for communication functions. In addition, it may also include: input / output interfaces (I / O interfaces) and network interfaces. Those skilled in the art will understand that... Figure 13 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, server 13 may also include... Figure 13 The more or fewer components shown, or having the same Figure 13 The different configurations shown.

[0140] It should be noted that the aforementioned one or more processors 1302 and / or other data processing circuitry are generally referred to herein as “data processing circuitry.” This data processing circuitry may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the server 13.

[0141] The memory 1304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method described in the embodiments of this application. The processor 1302 executes various functional applications and data processing by running the software programs and modules stored in the memory 1304, thereby realizing the above-described method for generating temporal behavior capture boxes based on self-attention networks. The memory 1304 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1304 may further include memory remotely located relative to the processor 1302, and these remote memories can be connected to the server 13 via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0142] The transmission device 1306 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of server 13. In one example, the transmission device 1306 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 1306 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0143] This specification provides the operational steps of the methods described in the embodiments or flowcharts, but more or fewer operational steps may be included based on conventional or non-inventive labor. The steps and order listed in the embodiments are merely one possible execution order among many steps and do not represent the only execution order. In actual system or interrupt product execution, the methods shown in the embodiments or drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0144] The structure shown in this embodiment is only a partial structure related to the solution of this application and does not constitute a limitation on the device to which the solution of this application is applied. Specific devices may include more or fewer components than shown, or combinations of certain components, or arrangements of different components. It should be understood that the methods, apparatuses, etc., disclosed in this embodiment can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or unit modules through some interfaces.

[0145] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0146] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this specification can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0147] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A resource recommendation method, characterized in that, The method includes: Obtain historical object resource data corresponding to the target object, wherein the historical object resource data includes object association data of the target object and resource association data of historical multimedia resources recommended to the target object; The recommendation results corresponding to the historical object resource data are analyzed to obtain the unverified impact data associated with the target recommendation index data. The unverified impact data includes object association data that affects the target recommendation index data and resource association data that affects the target recommendation index data. Determine the first correlation between the target recommendation index data and the impact data to be verified; Determine a second association relationship between the object-related data and the resource-related data; Recommendation relationship information is constructed using the target recommendation index data, the object association data, and the resource association data as nodes, and the first association relationship and the second association relationship as edges; The historical object resource data and the recommendation relationship information are input into the influencing factor analysis model to perform data impact analysis, and the influencing factor analysis results are obtained. The influencing factor analysis results characterize the distribution information of the impact of the historical object resource data on the target recommendation index data. Resource recommendation processing is performed based on the analysis results of the influencing factors.

2. The resource recommendation method according to claim 1, characterized in that, The influencing factor analysis model includes an interference determination module and an impact analysis module. The step of inputting the historical object resource data and the recommendation relationship information into the influencing factor analysis model for data impact analysis, and obtaining the influencing factor analysis results, includes: The recommendation relationship information is input into the interference determination module. Based on the path information between each node in the recommendation relationship information and the target node corresponding to the target recommendation index data, the interference node corresponding to the associated node of the target node is determined. The associated nodes, the interference nodes, and the historical object resource data are input into the impact analysis module to perform data impact analysis on the historical object resource data and obtain the analysis results of the influencing factors.

3. The resource recommendation method according to claim 1, characterized in that, The step of inputting the historical object resource data and the recommendation relationship information into the influencing factor analysis model for data influence analysis and obtaining the influencing factor analysis results includes: Based on the data types in the historical object resource data and the preset data impact analysis scenario, at least one target impact factor analysis model is determined from multiple preset impact factor analysis models. The historical object resource data and the recommendation relationship information are input into at least one target influencing factor analysis model to perform data influence analysis, and the model analysis results corresponding to each target influencing factor analysis model are obtained. Data fusion processing is performed on the analysis results of each model to determine the analysis results of influencing factors.

4. The resource recommendation method according to claim 3, characterized in that, The target influencing factor analysis model includes a target interference determination module and a target influence analysis module. The step of inputting the historical object resource data and the recommendation relationship information into at least one target influencing factor analysis model for data influence analysis, and obtaining the model analysis results corresponding to each target influencing factor analysis model, includes: The recommendation relationship information is input into the target interference determination module. Based on the path information between each node in the recommendation relationship information and the target node corresponding to the target recommendation index data, the interference node corresponding to the associated node of the target node is determined. The associated nodes, the interference nodes, and the historical object resource data are input into the target influencing factor analysis module. Data influence analysis is performed on the historical object resource data to obtain the model analysis results corresponding to the target influencing factor analysis model.

5. The resource recommendation method according to claim 1, characterized in that, The resource recommendation operation based on the analysis results of the influencing factors includes: Perform the target resource recommendation operation corresponding to the influencing factor analysis results on the sample objects associated with the target object to obtain the sample recommendation results; Based on the sample recommendation results, the analysis results of the influencing factors are verified; If the results of the influencing factor analysis are verified, the recommended operation based on the results of the influencing factor analysis is performed.

6. The resource recommendation method according to claim 1, characterized in that, Before inputting the historical object resource data and the recommendation relationship information into the influencing factor analysis model for data influence analysis and obtaining the influencing factor analysis results, the method further includes: Perform at least one feature construction process on the historical object resource data to obtain at least one target feature data; The step of inputting the historical object resource data and the recommendation relationship information into the influencing factor analysis model for data influence analysis and obtaining the influencing factor analysis results includes: The at least one target feature data and the recommendation relationship information are input into the influencing factor analysis model to perform data influence analysis, and the influencing factor analysis results are obtained.

7. A resource recommendation device, characterized in that, The device includes: The historical data acquisition module is used to acquire historical object resource data corresponding to the target object. The historical object resource data includes object association data of the target object and resource association data of historical multimedia resources recommended to the target object. A recommendation relationship information construction module is used to construct recommendation relationship information using target recommendation index data, object-related data, and resource-related data as nodes, and the relationship between the corresponding data of the nodes as edges. The recommendation relationship information construction module includes: a recommendation analysis unit, used to perform recommendation analysis on the recommendation results corresponding to the historical object-resource data to obtain unverified impact data associated with the target recommendation index data, the unverified impact data including object-related data affecting the target recommendation index data and resource-related data affecting the target recommendation index data; a first relationship determination unit, used to determine a first relationship between the target recommendation index data and the unverified impact data; a second relationship determination unit, used to determine a second relationship between the object-related data and the resource-related data; and a recommendation relationship information construction unit, used to construct the recommendation relationship information using the target recommendation index data, object-related data, and resource-related data as nodes, and the first and second relationships as edges. The data impact analysis module is used to input the historical object resource data and the recommendation relationship information into the impact factor analysis model to perform data impact analysis and obtain the impact factor analysis results. The impact factor analysis results characterize the impact distribution information of the historical object resource data on the target recommendation index data. The resource recommendation module is used to perform resource recommendation processing based on the analysis results of the influencing factors.

8. The apparatus according to claim 7, characterized in that, The influencing factor analysis model includes an interference determination module and an impact analysis module. The data impact analysis module includes: The first interference determination unit is used to input the recommendation relationship information into the interference determination module, and determine the interference nodes corresponding to the associated nodes of the target nodes based on the path information between each node in the recommendation relationship information and the target nodes corresponding to the target recommendation index data. The first impact analysis unit is used to input associated nodes, interference nodes, and historical object resource data into the impact analysis module, perform data impact analysis on the historical object resource data, and obtain the results of the impact factor analysis.

9. The apparatus according to claim 7, characterized in that, The data impact analysis module includes: The target model determination unit is used to determine at least one target influencing factor analysis model from multiple preset influencing factor analysis models based on the data types in historical object resource data and preset data influence analysis scenarios. The model analysis result acquisition unit is used to input historical object resource data and recommendation relationship information into at least one target influencing factor analysis model to perform data influence analysis and obtain the model analysis results corresponding to each target influencing factor analysis model. The data fusion unit is used to perform data fusion processing on the analysis results of each model to determine the analysis results of influencing factors.

10. The apparatus according to claim 9, characterized in that, The target influencing factor analysis model includes a target interference determination module and a target influence analysis module. The model analysis result acquisition unit includes: The second interference determination unit is used to input the recommendation relationship information into the target interference determination module, and determine the interference nodes corresponding to the associated nodes of the target nodes based on the path information between each node in the recommendation relationship information and the target nodes corresponding to the target recommendation index data. The second impact analysis unit is used to input associated nodes, interference nodes, and historical object resource data into the target impact factor analysis module, perform data impact analysis on the historical object resource data, and obtain the model analysis results corresponding to the target impact factor analysis model.

11. The apparatus according to claim 7, characterized in that, The resource recommendation module includes: The sample recommendation result acquisition unit is used to perform target resource recommendation operations corresponding to the influencing factor analysis results on the sample objects associated with the target object, and obtain the sample recommendation results; The validation unit is used to validate the results of the influencing factor analysis based on the sample recommendation results. The execution unit is used to perform the recommended actions based on the results of the influencing factor analysis, provided that the results of the influencing factor analysis have been validated.

12. The apparatus according to claim 7, characterized in that, The device further includes: The feature construction unit is used to perform feature construction processing on historical object resource data at least once to obtain at least one target feature data. The data impact analysis module includes: The feature analysis unit is used to input at least one target feature data into the influencing factor analysis model for data analysis and obtain the influencing factor analysis results.

13. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement a resource recommendation method as described in any one of claims 1-6.

14. A computer-readable storage medium, characterized in that, The storage medium includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement a resource recommendation method as described in any one of claims 1-6.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the resource recommendation method according to any one of claims 1-6.

Images