Information pushing method and device

Abstract

The invention provides an information pushing method and device, and the method comprises the steps: obtaining information pushing data which comprises first related data corresponding to a target object and second related data corresponding to to-be-pushed information; according to the first related data and the second related data, generating a predicted matching score corresponding to the to-be-pushed information, the predicted matching score being associated with a target matching degree and a target activeness, the target matching degree being used for representing a matching degree between the to-be-pushed information and the target object, the target activeness is used for representing the to-be-pushed information and the platform activeness of the target object; and determining a pushing strategy of the to-be-pushed information based on the predicted matching score corresponding to the to-be-pushed information. Therefore, the information pushing accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the general field of data processing technology, and in particular to an information push method and device. Background Art

[0002] At present, before pushing information (such as advertising push, information push, etc.), manufacturers will analyze the degree of correlation between the information to be pushed and the object based on the information of the object, so as to determine whether to push the information to be pushed to the object. However, simply judging the information push based on the degree of correlation between the two will limit the information push to the object's information itself, and for objects such as new platform users who do not have rich basic information, the accuracy of information push will be low. Summary of the invention

[0003] The present application provides an information push method and device, which introduces the parameters of the platform activity of the information to be pushed, so that when determining the information to be pushed, it is no longer limited to the data of the target object itself, and thus more accurate information push can be achieved even when the target object itself has less data.

[0004] In a first aspect, the present application provides an information push method, the method comprising:

[0005] Acquire information push data, the information push data including first relevant data corresponding to the target object and second relevant data corresponding to the information to be pushed;

[0006] Generate a predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data, the predicted matching score being associated with a target matching degree and a target activity degree, the target matching degree being used to characterize the matching degree between the information to be pushed and the target object, and the target activity degree being used to characterize the platform activity degree of the information to be pushed and the target object;

[0007] The push strategy of the information to be pushed is determined based on the predicted matching score.

[0008] It can be seen that in this application, while determining the degree of match between the information to be pushed and the target object, the platform activity of each of the two is also determined, and the predicted matching score is determined by combining the degree of match and the platform activity. The conventional practice is to determine the information to be pushed only by the degree of match between the information to be pushed and the target object, but this will result in lower accuracy in information push when there is less relevant data on the target object. Therefore, compared with the former, the present application can no longer be limited to the target object's own data when determining the information to be pushed, and in the case of a high degree of platform activity for the information to be pushed, the information to be pushed is more likely to obtain feedback from the target object, and thus, when the target object itself has less data, more accurate information push can be achieved.

[0009] In a second aspect, the present application provides an information push device, the device comprising:

[0010] An acquisition unit, used to acquire information push data, the information push data including first relevant data corresponding to the target object and second relevant data corresponding to the information to be pushed;

[0011] a processing unit, configured to generate a predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data, wherein the predicted matching score is associated with a target matching degree and a target activity degree, wherein the target matching degree is used to characterize a matching degree between the information to be pushed and the target object, and the target activity degree is used to characterize a platform activity degree of the information to be pushed and the target object;

[0012] The push unit is used to determine a push strategy for the information to be pushed based on the predicted matching score.

[0013] In a third aspect, the present application provides an electronic device comprising a processor, a memory, and a communication interface. The processor, the memory, and the communication interface are interconnected and perform communication with each other. The memory stores executable program code, the communication interface is used for wireless communication, and the processor is used to call the executable program code stored in the memory to execute, for example, some or all of the steps described in any method of the first aspect.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium, in which electronic data is stored. When the electronic data is executed by a processor, it is used to execute the electronic data to implement some or all of the steps described in the first aspect of the present application.

[0015] In a fifth aspect, the present application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps described in the first aspect of the present application. The computer program product may be a software installation package. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 paying creative work.

[0017] Figure 1 A schematic diagram of the structure of an information push system provided in an embodiment of the present application;

[0018] Figure 2 A flowchart of an information push method provided in an embodiment of the present application;

[0019] Figure 3 A flowchart of a method for calculating a predicted matching score provided in an embodiment of the present application;

[0020] Figure 4 A schematic diagram of a flow chart of a method for determining a prediction matching score based on a model provided in an embodiment of the present application;

[0021] Figure 5 A flowchart of a method for calculating a prediction matching score based on multiple platforms provided in an embodiment of the present application;

[0022] Figure 6 A schematic diagram of another information push method provided in an embodiment of the present application;

[0023] Figure 7 A block diagram of the functional units of an information push device provided in an embodiment of the present application;

[0024] Figure 8 A block diagram of the functional units of another information push device provided in an embodiment of the present application;

[0025] Fig. 9 A structural block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0027] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps is not limited to the listed steps, but optionally includes steps that are not listed, or optionally includes other steps inherent to these processes, methods, products or devices.

[0028] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0029] At present, the analysis of the degree of association between the information to be pushed and the object is an indispensable part before the information is pushed, which enables the information to be pushed to obtain a higher degree of feedback after being pushed, and enables the object to obtain a better browsing experience when browsing different platforms. However, the general analysis of the degree of association between the information to be pushed and the object is limited to the matching degree between the object information and the information to be pushed. For example, in general, after the object logs in to the platform, the platform will obtain the basic information of the object, and determine whether to push the push information to the object based on the matching degree between the basic information and the information to be pushed. However, the push of information in this way will be limited to the basic information of the object itself. If the basic information of the object is less, it is difficult to achieve effective information push.

[0030] Based on this, the present application provides an information push method. When analyzing the degree of association between the information to be pushed and the object, not only the matching degree between the object's own data and the information to be pushed is taken into consideration, but also the activity of the object and the information to be pushed on the platform. This makes the push of information no longer limited to the object's own data.

[0031] The following is a detailed description of the embodiments of the present application:

[0032] See also Figure 1 , Figure 1 A schematic diagram of the structure of an information push system provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the information push system includes a server 101 and a terminal device 102 .

[0033] The server 101 is used to obtain relevant data, calculate the predicted matching score after the relevant data is obtained, and formulate a corresponding push strategy according to the predicted matching score. Optionally, the server 101 can be a server, a server cluster, a cloud server, a cloud computing service center, or other forms of equipment with computing capabilities.

[0034] The terminal device 102 is used to represent the device corresponding to the target object. Optionally, the terminal device 102 can be a desktop computer, a laptop computer, a tablet computer, a smart phone, etc.

[0035] Specifically, in the present application, the server 101 obtains the first relevant data corresponding to the target object through the terminal device 102, and simultaneously obtains the second relevant data corresponding to the information to be pushed, and then generates the predicted matching score corresponding to the information to be pushed according to the second relevant data and the first relevant data. In the process of generating the predicted matching score, the matching degree between the information to be pushed and the target object, and the platform activity corresponding to the information to be pushed and the target object are respectively generated according to the second relevant data and the first relevant data, and the predicted matching score is generated according to the matching degree between the information to be pushed and the target object, and the platform activity corresponding to the information to be pushed and the target object. In this way, when analyzing the degree of association between the information to be pushed and the target object, the factors of the platform activity of the information to be pushed and the target object itself are introduced, so that the push of information is no longer limited to the data of the target object itself. In the case of high platform activity of the information to be pushed, the possibility of the information to be pushed to get feedback from the target object is high, and thus, when the data of the target object itself is less, more accurate information push can be achieved.

[0036] Based on this, an embodiment of the present application provides an information push method, and the embodiment of the present application is described in detail below in conjunction with the accompanying drawings.

[0037] See also Figure 2 , Figure 2 A flow chart of an information push method provided in an embodiment of the present application is provided, and the method is applied to the above-mentioned server, such as Figure 2 As shown, the method comprises the following steps:

[0038] Step S201, obtaining information push data.

[0039] Among them, the information push data includes the second relevant data corresponding to the information to be pushed and the first relevant data corresponding to the target object. The first relevant data can be divided into first behavioral data, first text data and first geographic data. If the information to be pushed is commodity information such as advertisements, the corresponding first behavioral data can also be divided into behavioral data and social data. The behavioral data refers to the browsing behavior, purchasing behavior, clicking behavior, etc. for commodities extracted through database query or log analysis; the social data may refer to the target object's interactive records for the social part obtained by using the social media interface (Application Programming Interface, API) or crawling technology, such as likes, comments, and sharing behaviors. The first text data may be the target object's own label obtained through database query, such as gender, interests, personal profile, etc. The first geographic data may be the location information of the target object obtained from GPS data or the platform's geographic location record.

[0040] It can be understood that the platform shown in this application refers to a software application platform.

[0041] The second related data can also be divided into second behavior data, second text data and second geographic data. The second behavior data refers to the click-through rate (CTR) and conversion rate (CVR) of the information to be pushed obtained through the corresponding interface or database query. The second text data refers to the title, description, keywords, etc. of the information to be pushed obtained through the corresponding interface or database query. The second geographic data refers to the location information associated with the information to be pushed obtained through the corresponding interface or database query, such as the location of the product.

[0042] Step S202: Generate a predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data.

[0043] Among them, the predicted matching score is associated with the target matching degree and the target activity. The target matching degree is used to characterize the matching degree between the information to be pushed and the target object, and the target activity is used to characterize the platform activity of the information to be pushed and the target object.

[0044] Specifically, the composition of the target matching degree is described in detail below:

[0045] In a feasible embodiment, the target matching degree is determined according to text similarity, behavior similarity, and geographic similarity between the first relevant data and the second relevant data.

[0046] The matching degree between the first relevant data and the second relevant data can be obtained by similarity calculation. Since the first relevant data and the second relevant data may include multiple types of data, a single similarity calculation cannot be performed when determining the matching degree between the two. Therefore, this embodiment jointly determines the matching degree between the first relevant data and the second relevant data through text similarity, behavior similarity, and geographic similarity.

[0047] Text similarity refers to the similarity between the first related data and the second related data in terms of content, meaning or expression. The similarity between the two can be determined by a vector-based method, such as converting text into vectors through TF-IDF (term frequency-inverse document frequency) and then calculating the similarity between vectors; or by a semantic-based method, such as using word embedding (such as Word2Vec, BERT, etc.) to convert words into vectors in a high-dimensional space. These vectors can capture the semantic relationship between words, and then calculate the similarity between texts.

[0048] Behavior similarity refers to the similarity between the behavior patterns of the target object and the information to be pushed. The behavior of the target object may include browsing behavior, clicking behavior, purchasing behavior, etc. of the pushed information; the behavior corresponding to the information to be pushed may include display behavior, click behavior, etc. The similarity between the behavior patterns of the target object and the information to be pushed can also be determined based on vector methods, such as converting the behavior data of the target object and the information to be pushed into feature vectors, and then performing similarity between the vectors.

[0049] Geographic similarity refers to the degree of proximity in spatial distribution between the geographic location corresponding to the target object and the geographic location corresponding to the information to be pushed, which can be determined by the longitude and latitude coordinates between the two.

[0050] Exemplarily, the formula corresponding to the target matching degree is as follows: Among them, Match refers to the target matching degree, α e Refers to the priority corresponding to the similarity e, Similarity e refers to the similarity e, and the similarity e can refer to the aforementioned text similarity, behavior similarity and geographical similarity, α e Can be determined based on experience.

[0051] In the present application, since the first relevant data and the second relevant data may include multiple types of data, the target matching degree is comprehensively determined by the text similarity, behavioral similarity and geographical similarity between the first relevant data and the second relevant data, so that corresponding similarity calculations can be performed for different types of data, and the calculated matching degree can be more accurate.

[0052] Specifically, in a feasible embodiment, the first related data includes first text data, first behavior data and first geographic data, and the second related data includes second text data, second behavior data and second geographic data, wherein text similarity is determined based on the first text data and the second text data; behavior similarity is determined based on the first behavior data and the second behavior data; and geographic similarity is determined based on the first geographic data and the second geographic data.

[0053] The first related data and the second related data are divided into behavior data, text data and geographic data, respectively, so as to facilitate the calculation of text similarity, behavior similarity and geographic similarity, respectively.

[0054] Exemplarily, when calculating text similarity, cosine similarity may be used for calculation, and the calculation formula includes: Wherein, a and b represent feature vectors corresponding to the first text data and the second text data respectively.

[0055] When calculating behavior similarity, you can use Euclidean distance to calculate. The calculation formula includes: Euclidean distance Among them, ui and vi represent the feature vectors corresponding to the first behavior data and the second behavior data respectively.

[0056] When calculating geographic similarity, the great circle distance (Haversine formula) can be used for calculation. The calculation formula includes: Here, φ1 and λ1 refer to the latitude and longitude included in the first geographic data, φ2 and λ2 refer to the latitude and longitude included in the second geographic data, and r is the radius of the earth.

[0057] In the present application, the first related data and the second related data are divided into three types of data, respectively, so as to facilitate the corresponding similarity calculation based on the same type of data between the two.

[0058] Specifically, the composition of target activity is described in detail below:

[0059] In a feasible embodiment, the first related data includes first behavior data, and the second related data includes the average click-through rate and the average conversion rate of the information to be pushed, wherein the platform activity of the target object in the target activity is determined based on the feature value corresponding to the feature included in the first behavior data; the platform activity of the information to be pushed in the target activity is determined based on the average click-through rate and the average conversion rate of the information to be pushed.

[0060] The platform activity corresponding to the target object refers to the target object's software usage activity on the data source platform. The platform activity corresponding to the target object is mainly determined by the feature value corresponding to the feature included in the first behavior data of the target object in the platform. Exemplarily, the calculation formula for the platform activity of the target object is as follows: Among them, wi refers to the feature weight corresponding to the feature value, fi refers to the feature value corresponding to the feature included in the first behavior data, and the feature weight can be determined based on experience. Assume there are two features: user browsing times 10 times, user purchase times 2 times, feature values: user browsing times f1 = 0.1, user purchase times f2 = 0.02, feature weights: w1 = 0.5, w2 = 0.5.

[0061] The platform activity corresponding to the information to be pushed refers to the degree of activity of the information to be pushed being triggered on the data source platform. The platform activity corresponding to the information to be pushed is mainly obtained through the average click-through rate and average conversion rate of the target object on the platform, and the average click-through rate and average conversion rate can belong to the second behavior data of the aforementioned information to be pushed. Exemplarily, the calculation formula for the platform activity of the information to be pushed is as follows: Platform activity (information to be pushed) = α1×CTR+β1×CVR, where CTR refers to the average click-through rate, CVR refers to the average conversion rate, and α1 and β1 are adjustable parameters used to balance the importance of CTR and CVR. The adjustment of α1 and β1 can be based on actual business goals to determine the importance of CTR and CVR respectively. For example, if the current information to be pushed is an advertisement, if the click-through rate is more important to the advertising revenue, CTR can be given a higher weight α; if the conversion rate can directly bring actual revenue such as sales or registrations, the weight β of CVR is increased.

[0062] After obtaining the platform activity corresponding to the target object and the platform activity corresponding to the information to be pushed, the target activity can be obtained by the following formula: Target activity = α2 × platform activity (object) + β2 × platform activity (information to be pushed). Similarly, α2 and β2 are adjustable parameters used to balance the importance of the platform activity of the target object and the platform activity of the information to be pushed. The adjustment of α2 and β2 can also be made according to the actual business goals. For example, if more attention is paid to the interests of users, α2 can be increased; if more attention is paid to the feedback of users on the pushed information, β2 can be increased.

[0063] In the present application, the user's platform activity is determined by the first behavior data corresponding to the target object, so that the target object's activity on the platform can be accurately understood. The higher the activity, the more willing the target object is to learn about the information pushed on the platform. At the same time, the platform activity of the information to be pushed is determined based on the average click-through rate and average conversion rate of the information to be pushed on the platform, so the audience level of the information to be pushed can be accurately understood. Similarly, the higher the audience level of the information to be pushed, the more likely the target object is to be interested in the pushed information. Therefore, the target activity determined by combining the two can better reflect the target object's interest in the pushed information on the platform.

[0064] Combine the following Figure 3 The above prediction matching score calculation process is described in detail:

[0065] For example, see Figure 3 , Figure 3 A flowchart of a method for calculating a predicted matching score provided in an embodiment of the present application is shown in FIG. Figure 3As shown, it includes: obtaining the platform activity corresponding to the target object and the platform activity corresponding to the information to be pushed according to the first relevant data and the second relevant data, and obtaining the text similarity, behavior similarity and geographic similarity respectively. The platform activity corresponding to the target object is determined according to the feature value corresponding to the feature included in the first behavior data in the first relevant data; the platform activity corresponding to the information to be pushed is obtained according to the average click rate and the average conversion rate in the second relevant data. The text similarity is obtained by calculating the similarity between the first text data in the first relevant data and the second text data in the second relevant data; the behavior similarity is obtained by calculating the similarity between the first behavior data in the first relevant data and the second behavior data in the second relevant data; the geographic similarity is obtained by calculating the similarity between the first geographic data in the first relevant data and the second geographic data in the second relevant data. Then, the target activity is obtained according to the platform activity corresponding to the target object and the platform activity corresponding to the information to be pushed, and then the target activity is obtained according to the text similarity, behavior similarity and geographic similarity. Finally, the predicted matching score is obtained according to the target activity and the target matching.

[0066] In addition, the aforementioned method of generating predicted matching scores can also be applied to the model, and the initial model configuration is performed through the aforementioned calculation process, so that the model obtained by subsequent training can be directly applied to the generation of predicted matching scores.

[0067] Specifically, in a feasible embodiment, before generating a predicted matching score corresponding to the information to be pushed based on the first relevant data and the second relevant data, the method also includes: obtaining training data, the training data including first historical relevant data, second historical relevant data and a matching score, the first historical relevant data being the historical relevant data corresponding to the target push information, the second historical relevant data being the historical relevant data corresponding to the target object, and the matching score being determined based on the click-through rate and conversion rate of the target object to the target push information; performing model training on the initial model based on the training data to obtain a matching score prediction model after the training is completed; generating a predicted matching score corresponding to the information to be pushed based on the second relevant data and the first relevant data, including: inputting the second relevant data and the first relevant data into the matching score prediction model to obtain a predicted matching score corresponding to the information to be pushed.

[0068] The first historical related data and the second historical related data may be obtained in the manner described in the above embodiment. The above data may be divided into behavioral data, social data, geographic data, and text data, and social data refers to the related data corresponding to the target object. Social data may also be text data including interest tags, or may include social graphs. The target push information corresponding to the first historical related data refers to the push information of the target object that has been previously clicked and converted.

[0069] After obtaining the corresponding training data, data preprocessing can be performed first. First, data cleaning can be performed. For example, it can include the removal of noise and outliers, specifically using statistical methods (such as the 3σ principle, box plots) to detect and remove outliers; it can also include missing value processing, specifically for different types of data, using mean filling, front and back value interpolation or deletion and other strategies. Then data integration can be performed. For example, it can include standardization processing, specifically normalizing numerical data so that different features have similar numerical ranges. It can also include target object data merging processing, specifically based on the ID of the target object, merging data from different sources to build a unified user portrait.

[0070] The data above needs to be encoded to facilitate model training. Data encoding can be performed in different ways according to the above data types. Behavioral data encoding can include: statistical feature extraction, specifically calculating the number of clicks, purchase frequency, browsing time and other features of each user. Time series features, specifically, using time interval models (such as RNN, LSTM) to extract behavioral sequence features, formula: h t =f(h t-1 ,x t ), where ht represents the hidden state at time t, and xt represents the input feature. Text data encoding may include: term frequency-inverse document frequency (TF-IDF) encoding, specifically TF-IDF encoding of the advertising text to generate a sparse matrix, formula: TF-IDF(g,k)=TF(g,k)×IDF(g), where TF(g,k) is the term frequency of term g in document k, and IDF(g) is the inverse document frequency of term g. Word embedding method, specifically, using BERT to convert text into a dense vector. Formula (BERT): h=BERT(X), where X is the input text sequence and h is the generated text embedding.

[0071] Geographic data encoding may include: Region encoding: Mapping geographic coordinates to region codes, and representing them through clustering (such as K-means), formula (K-means): c(i) = argmin||xi-μk||, where xi is the geographic coordinate and μk is the cluster center. Social data encoding may include: Graph embedding: Using graph embedding methods (such as Node2Vec) to convert social graphs into vector representations, formula: maximize∑(u,v)log(σ(z u ·z v )), where zu and zv are the embeddings of nodes u and v, σ is the sigmoid function, and nodes u and v are usually node pairs generated by a random walk process.

[0072] After the training data is processed as described above, feature sets can be established according to the data feature type (behavioral data, social data, geographic data, text data). The features can then be aggregated using feature combination or dimensionality reduction techniques (such as PCA). Formula (PCA): Z = XW, where X is the input feature matrix and W is the dimensionality reduction matrix.

[0073] At this time, the model training is performed again through the predicted matching score calculation process shown in the above embodiment, and optimization is performed according to the matching scores in the training data. The initial model can be a gradient boosted decision tree (GBDT) model or other neural network model.

[0074] It can be understood that since the matching score refers to the click-through rate and conversion rate of the target object to the target push information, the target push information is the push information that the target object has clicked on before. In the process of model training, the platform activity of the target object and the target push information are obtained respectively according to the first historical related data corresponding to the target object in the training data and the second historical related data corresponding to the target push information, and then the comprehensive activity is determined according to the platform activity, that is, the aforementioned target activity. At the same time, the text similarity, behavioral similarity and geographical similarity between the target object and the target push information are calculated according to the first historical related data and the second historical related data, and then the comprehensive matching degree is obtained according to the three similarities, that is, the aforementioned target matching degree. Finally, the predicted matching score is obtained based on the aforementioned comprehensive activity and comprehensive matching degree. The predicted matching score is optimized according to the matching score in the training data, and the training goal is to minimize the matching error.

[0075] At the same time, during the model training process, efficient layer traversal training can be performed through layer traversal technology. Specifically, the efficient LLVM series layer traversal technology can be introduced to simulate the lookback mechanism through recursive convolutional layers (such as RNN). Formula (Transformer): H = MultiHead (Q, K, V), where Q, K, and V are query, key, and value matrices, respectively. The query matrix and key matrix are used to calculate the attention score, which reflects the correlation or importance between different positions in the input sequence. The value matrix contains the actual information of the input sequence and is weighted and summed according to the attention score to generate the output representation of the self-attention mechanism.

[0076] In the optimization process, key neurons can also be identified based on gradient strategies to perform forgetting and learning. Neuron identification can be done by using gradient analysis to identify key neurons related to human preferences. The formula is: Where L is the loss function and w is the weight. Based on this, alignment optimization is performed, and the alignment process is decomposed into forgetting and learning stages to update the model parameters. The formula (optimization) is: Among them, W newis the updated weight, W old is the weight before updating, η refers to the learning rate, and η can be set according to the empirical value. It can be understood that the weight here can be used to represent the weight in the calculation formula included in the calculation process of the predicted matching score listed in the above embodiment, and the process of model optimization can be the process of updating the above weight.

[0077] The aforementioned alignment process essentially refers to matching the model output with the actual behavior and preferences of the user. The essence of key neuron identification is to find out the neurons that play a key role in the preferences of the target object. Forgetting and learning essentially refers to adjusting the model to forget the information that is inconsistent with the preferences of the target object and learn new alignment information. Identifying key neurons can be based on the size of the gradient value obtained by the aforementioned formula to identify the neurons that have the greatest impact on the loss change, select a threshold (such as the top 5% of the gradient value) to select the most important neurons, and subsequent optimization adjustments are mainly for key neurons. The forgetting stage can be in the alignment process, "forgetting" information that is inconsistent with the preferences of the target object. For the misaligned neurons, specifically, the weights of these neurons can be updated using the identified gradient values ​​to weaken their influence. At the same time, the regularization term can be added to make the output of the misaligned neurons approach zero. The learning stage can be learning new alignment information while forgetting irrelevant information, such as strengthening the weights of neurons related to alignment based on the preference data of the target object.

[0078] After the model training is performed in the manner shown in the above embodiment, a corresponding matching score prediction model is obtained. When it is necessary to determine the matching score between the information to be pushed and the target object later, the first relevant data and the second relevant data can be input into the matching score prediction model, and the matching score prediction model obtains the predicted matching score corresponding to the information to be pushed after calculating and processing the first relevant data and the second relevant data according to the corresponding indicators obtained after training.

[0079] In the present application, a matching score prediction model is introduced, and model training is performed based on the historical relevant data of the target object and the historical relevant data of the target push information with which the target object has previously interacted, so that the trained model has strong adaptability to the target object, and can be used for calculating the predicted matching scores between other information to be pushed and the target object with higher accuracy.

[0080] The following is a detailed description of the model reasoning process:

[0081] Further, in a feasible embodiment, the first relevant data includes first behavior data, and the matching score prediction model includes historical platform activity of the target object, and the historical platform activity of the target object is determined based on the historical relevant data corresponding to the target object, wherein the platform activity of the target object in the target activity is determined based on the feature value corresponding to the feature included in the first behavior data, including: the platform activity of the target object in the target activity is determined based on the feature value corresponding to the feature included in the first behavior data, and the historical platform activity.

[0082] In the process of model reasoning, since the model records the historical platform activity of the target object after training, if the historical platform activity of the target object is directly re-determined based on the first relevant data after the first relevant data is input, the user's historical situation cannot be fully considered. If the user's previous platform activity was high, but based on certain circumstances, the user's current platform activity is low, it cannot be simply determined that the user's activity on the platform is low. Therefore, in the process of model reasoning in this embodiment, the determination of the platform activity of the target object in the target activity is based on the historical platform activity and the current platform activity. The current platform activity is obtained based on the feature value corresponding to the feature included in the first behavior data.

[0083] The calculation formula of platform activity can be: updated platform activity = γ × historical platform activity + (1-γ) × current platform activity, where γ is the update coefficient. γ can be determined based on empirical values ​​or based on the degree of change in the platform activity of the target object. If the platform activity of the target object changes more steadily, γ can be set higher; if the platform activity of the target object changes more dramatically, γ can be set lower.

[0084] At the same time, the feedback data can be updated later through the actual clicks, conversions, and other behaviors of the target object, and the model parameters can be updated using the new feedback data through online learning or batch update methods. Formula (online learning): Similarly, L is the loss function, w is the weight, and η refers to the learning rate.

[0085] In the present application, during the model reasoning process, the platform activity of the target object will be comprehensively determined based on the current target object's behavioral data and the historical platform activity obtained through prior training, thereby taking into account both the impact of real-time changes in the target object's activity and the impact of the target object's historical platform activity, thereby making the obtained platform activity of the target object more accurate.

[0086] The above process is described in detail below through diagrams:

[0087] For example, see Figure 4 , Figure 4 A flow chart of a method for determining a prediction matching score based on a model provided in an embodiment of the present application is shown in FIG. Figure 4 As shown, it includes: firstly, data preparation is performed, which may specifically include data cleaning and data integration of the acquired data. Subsequently, data encoding is performed, which may specifically include different types of encoding for different types of data. Model training is then performed based on the encoded data, specifically, the corresponding predicted matching score is obtained according to the formula shown in the aforementioned embodiment, and the predicted matching score is optimized according to the matching score in the training data. In essence, each weight in the formula is optimized, with minimizing the matching deviation as the training goal, so as to obtain a matching score prediction model after the training is completed. Subsequently, the model reasoning is performed on the relevant data corresponding to the information to be pushed and the target object determined by the predicted matching score as needed, so as to obtain the predicted matching score between the information to be pushed and the target object. At the same time, real-time feedback will be performed later, specifically, the feedback data is updated according to the actual click, conversion and other behaviors of the target object, and the model parameters are updated according to the new feedback data.

[0088] The above process is described in detail below through a specific example:

[0089] Exemplarily, advertising features (A) include: title "Comfortable running shoes", description "sports shoes suitable for long-distance running", keywords "sports", "running", "shoes", average CTR of 0.15, average CVR of 0.02, and geographic location in Shanghai; user features (U) include: browsing history of 20 sports shoe products, purchase history of 2 pairs of running shoes, interest tags of "running" and "fitness", and geographic location in Shanghai.

[0090] Data encoding is performed on the above data. Text features: The title, description, and keywords in the advertisement (A) are merged into "comfortable running shoes suitable for long-distance running sports shoes"; the interest tags in the user (U) are merged into "running fitness sports shoes"; after TF-IDF encoding, the TF-IDF vector of the advertisement (A) is [0.6, 0.4, 0.8], and the TF-IDF vector of the user (A) is [0.7, 0.3, 0.5]. Behavioral features: The advertisement CTR is 0.15, the user browsing history includes 2 times, the advertisement CVR is 0.02, and the number of user purchases is 2 times; after Min-Max normalization, the advertisement CTR: 0.15, browsing history: 0.2, advertisement CVR: 0.02, and the number of purchases: 0.2. Geographic features: The geographical location is Shanghai->(31.2304, 21.4737).

[0091] After model training, the above data are combined into training samples to obtain the feature vector: [0.6, 0.4, 0.8, 0.15, 0.02, 31.2304, 121.4737, 0.7, 0.3, 0.5, 0.2, 0.2, 31.2304, 121.4737]. Assume that there are multiple samples to build a feature matrix for training. The model can choose the gradient boosting decision tree (GBDT). At this time, the input features are [0.6, 0.4, 0.8, 0.15, 0.02, 31.2304, 121.4737, 0.7, 0.3, 0.5, 0.2, 0.2, 31.2304, 121.4737], and the target variable is the matching score (obtained based on the historical click-through rate and conversion rate of user U for ad A). The training data is the aforementioned feature matrix and target variable, the training goal is to minimize the matching error, and the output is the predicted matching score between the advertisement and the user. The model structure may include an input layer, a hidden layer, and an output layer. The input layer corresponds to the feature dimension; the hidden layer is a number of fully connected layers, and also includes the activation function ReLU; the output layer corresponds to the matching score, which can be between 0 and 1; the loss function can use mean square error (MSE) or cross entropy (depending on the label type); the optimizer can use Adam or SGD; the training process uses the gradient descent method to update the weights.

[0092] After the model training is completed, new sample features are input, including the advertisement feature vector and the user feature vector, which are combined and input into the trained model to output the predicted matching score (eg, 0.92).

[0093] Furthermore, in a feasible embodiment, the predicted matching score is the sum of multiple sub-predicted matching scores, and the multiple sub-predicted matching scores are generated based on information push data from multiple different platforms.

[0094] In addition to determining the predicted match score based on the data of a single platform, the predicted match score can also be comprehensively considered in combination with the data of multiple platforms. Specifically, the predicted match score calculation can be performed based on the data of multiple platforms to obtain the sub-predicted match scores corresponding to the multiple platforms, and finally the total predicted match score can be obtained based on the multiple sub-predicted match scores.

[0095] The calculation formula of the predicted matching score is as follows: predicted matching score = ∑ d Activity (d) × Match (d, Ad), where Activity (d) refers to the target activity corresponding to platform d, and Match (d, Ad) refers to the target matching degree corresponding to the platform. Since the result of multiplication is not only affected by the size of the two numbers, but also due to the "amplification" effect, it may be much larger than the two numbers themselves. In this embodiment, the target activity and the target matching degree are multiplied to determine the sub-prediction matching score, which can suddenly reflect the impact of both on the predicted matching score.

[0096] Furthermore, if it is necessary to select one piece of information to be pushed from a batch of information to be pushed, the information with the highest predicted matching score in the batch can be selected. Combining the above formula, the formula can be obtained as follows: selected information = argmax∑ d Activity(d)×Match(d,Ad).

[0097] The calculation of prediction matching scores based on data from multiple platforms can also be applied to the above model, that is, feature matrices are established for data from different platforms, and sub-prediction matching scores are calculated separately, and finally multiple sub-prediction matching scores are aggregated to obtain the total prediction matching score.

[0098] In the present application, the predicted matching scores obtained by integrating the data of multiple platforms are used to obtain the total predicted matching score, so that the obtained user portrait is more complete, and the accuracy of the predicted matching score obtained thereby will be higher.

[0099] Combine the following Figure 5 The process of obtaining predicted matching scores for data from multiple platforms is described in detail:

[0100] For example, see Figure 5 , Figure 5 The following is a flow chart of a method for calculating a prediction matching score based on multiple platforms provided in an embodiment of the present application. Figure 5 As shown, it includes: acquiring data from multiple platforms (platform 1, platform 2, platform 3) respectively, so as to obtain relevant data corresponding to each platform respectively. The relevant data corresponding to each platform should include relevant data corresponding to the target object and the information to be pushed respectively. Then, the matching score calculation is performed on the relevant data corresponding to each platform respectively, so as to obtain the sub-prediction matching score corresponding to each platform. Then, the total prediction matching score is obtained according to the sub-prediction matching scores corresponding to the multiple platforms respectively.

[0101] Step S203: determining a push strategy for the information to be pushed based on the predicted matching score.

[0102] After obtaining the predicted matching score corresponding to the information to be pushed, the push strategy for the information to be pushed to the target object can be determined accordingly, for example, the push frequency, the number of pushes, etc.

[0103] The following is a detailed description of the current steps:

[0104] In a feasible embodiment, a push strategy for the information to be pushed is determined based on the predicted matching score, including: when the predicted matching score is higher than a preset score, determining the push frequency of the information to be pushed according to the predicted matching score, the higher the predicted matching score, the higher the push frequency; and pushing the information to be pushed to the target object according to the push frequency.

[0105] Among them, before pushing, it is necessary to determine whether the push information should be pushed to the target object based on the predicted matching score. If the predicted matching score is too low, it preliminarily indicates that the target object may have a low interest in the information to be pushed. At this time, after pushing the information to be pushed to the target object, the possibility of obtaining feedback from the target object is low, and it will also affect the target object's software platform usage experience. At this time, you can choose not to push the push information to the target object. If the predicted matching score is too high, it preliminarily indicates that the target object may have a high interest in the information to be pushed. At this time, after pushing the information to be pushed to the target object, the possibility of obtaining feedback from the target object is high, so you can choose to push the information to be pushed to the target object.

[0106] After determining to push the information to be pushed to the target object, the push frequency can be determined according to the predicted match score. The higher the predicted match score, the higher the push frequency. At the same time, multiple score intervals can be divided according to the score range between the preset score and the full score, and the corresponding push frequency can be determined according to the interval corresponding to the predicted match score. Similarly, the higher the minimum score corresponding to the score interval, the higher the push frequency corresponding to the score interval. The push frequency can represent the number of pushes per day. And for daily pushes, you can also choose a time period when the target object is more active for push.

[0107] In this application, the information to be pushed is pushed to the target object only when the predicted matching score is higher than the preset score, which can increase the possibility of receiving feedback from the target object after the information to be pushed is pushed. At the same time, different push frequencies are determined according to the matching score, which further increases the possibility of receiving feedback from the target object after the information to be pushed is pushed, thereby improving the efficiency of information push for the information to be pushed.

[0108] Combine the following Figure 6 The application of this application is described in detail:

[0109] For example, see Figure 6 , Figure 6 This is a schematic diagram of another information push method provided in an embodiment of the present application. Figure 6As shown, it includes: determining multiple information to be pushed (information to be pushed 1, information to be pushed 2, information to be pushed 3) that need to be pushed, respectively inputting the relevant data corresponding to each information to be pushed into the matching score prediction model in combination with the relevant data corresponding to the target object, and obtaining the predicted matching score corresponding to each information to be pushed (predicted matching score 1, predicted matching score 2, predicted matching score 3), and then determining the highest score from the multiple predicted matching scores to obtain the target information to be pushed corresponding to the highest score, and finally pushing the target information to be pushed to the target object according to the corresponding push strategy.

[0110] It can be seen that in the embodiment of the present application, while determining the degree of match between the information to be pushed and the target object, the platform activity of each of the two is also determined, and the predicted matching score is determined by combining the degree of match and the platform activity. The conventional practice is to determine the information to be pushed only by the degree of match between the information to be pushed and the target object, but this will result in a lower accuracy of information push when there is less relevant data on the target object. Therefore, compared with the former, the present application can make it no longer limited to the target object's own data when determining the information to be pushed, and in the case of a high platform activity of the information to be pushed, the information to be pushed is more likely to obtain feedback from the target object, and thus, when the target object itself has less data, more accurate information push can be achieved.

[0111] In accordance with the above-mentioned embodiments, please refer to Figure 7 , Figure 7 This is a functional unit block diagram of an information push device provided in an embodiment of the present application. The information push device is the above-mentioned server or a part of the server, such as Figure 7 As shown, the information push device 70 includes:

[0112] The acquisition unit 701 is used to acquire information push data, where the information push data includes first related data corresponding to the target object and second related data corresponding to the information to be pushed;

[0113] A processing unit 702 is used to generate a predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data, where the predicted matching score is associated with a target matching degree and a target activity degree, where the target matching degree is used to characterize a matching degree between the information to be pushed and the target object, and the target activity degree is used to characterize a platform activity degree of the information to be pushed and the target object;

[0114] The push unit 703 is used to determine a push strategy for the information to be pushed based on the predicted matching score.

[0115] In a feasible embodiment, the target matching degree is determined according to text similarity, behavior similarity, and geographic similarity between the first relevant data and the second relevant data.

[0116] In a feasible embodiment, the first related data includes first text data, first behavioral data and first geographic data, and the second related data includes second text data, second behavioral data and second geographic data, wherein text similarity is determined based on the first text data and the second text data; behavioral similarity is determined based on the first behavioral data and the second behavioral data; and geographic similarity is determined based on the first geographic data and the second geographic data.

[0117] In a feasible embodiment, the first related data includes first behavior data, and the second related data includes the average click-through rate and the average conversion rate of the information to be pushed, wherein the platform activity of the target object in the target activity is determined based on the feature value corresponding to the feature included in the first behavior data; the platform activity of the information to be pushed in the target activity is determined based on the average click-through rate and the average conversion rate of the information to be pushed.

[0118] In a feasible embodiment, the acquisition unit 701, before the processing unit 702 generates the predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data, is also used to: acquire training data, the training data including the first historical relevant data, the second historical relevant data and the matching score, the first historical relevant data being the historical relevant data corresponding to the target push information, the second historical relevant data being the historical relevant data corresponding to the target object, and the matching score being determined based on the click-through rate and conversion rate of the target object to the target push information; the processing unit 702 is also used to perform model training on the initial model based on the training data to obtain a matching score prediction model after the training is completed; in terms of generating the predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data, the processing unit 702 is specifically used to: input the first relevant data and the second relevant data into the matching score prediction model to obtain the predicted matching score corresponding to the information to be pushed.

[0119] In a feasible embodiment, the first relevant data includes first behavior data, and the matching score prediction model includes the historical platform activity of the target object, and the historical platform activity of the target object is determined based on the historical relevant data corresponding to the target object, wherein the platform activity of the target object in the target activity is determined based on the feature value corresponding to the feature included in the first behavior data, including: the platform activity of the target object in the target activity is determined based on the feature value corresponding to the feature included in the first behavior data, and the historical platform activity.

[0120] In a feasible embodiment, the predicted matching score is the sum of multiple sub-predicted matching scores, and the multiple sub-predicted matching scores are generated based on information push data from multiple different platforms.

[0121] In a feasible embodiment, in terms of determining the push strategy of the information to be pushed based on the predicted matching score, the push unit 703 is specifically used for: when the predicted matching score is higher than the preset score, determining the push frequency of the information to be pushed according to the predicted matching score, the higher the predicted matching score, the higher the push frequency; and pushing the information to be pushed to the target object according to the push frequency.

[0122] It can be understood that since the method embodiment and the device embodiment are different presentation forms of the same technical concept, the content of the method embodiment part in this application should be synchronously adapted to the device embodiment part and will not be repeated here.

[0123] In the case of an integrated unit, such as Figure 8 As shown, Figure 8 This is a block diagram of the functional units of another information push device 70 provided in an embodiment of the present application. Figure 8 In the embodiment, the information push device 70 includes: a processing module 812 and a communication module 811. The processing module 812 is used to control and manage the actions of the information push device 70, such as the steps of the acquisition unit 701 and the processing unit 702, and / or other processes for executing the technology described herein. The communication module 811 is used to support the interaction between the information push device 70 and other devices, such as the steps of the push unit 703. Figure 8 As shown, the information pushing device 70 may further include a storage module 813 , and the storage module 813 is used to store program codes and data of the information pushing device 70 .

[0124] Among them, the processing module 812 can be a processor or a controller, for example, a central processing unit (CPU), a general processor, a digital signal processor (DSP), an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It can implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements a computing function, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The communication module 811 can be a transceiver, an RF circuit or a communication interface, and the like. The storage module 813 can be a memory.

[0125] Among them, all relevant contents of each scenario involved in the above method embodiment can be referred to the functional description of the corresponding functional module, and will not be repeated here. Figure 2 The information push method shown.

[0126] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When a computer instruction or computer program is loaded or executed on a computer, a process or function according to an embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0127] Fig. 9 This is a structural block diagram of an electronic device provided in an embodiment of the present application. Fig. 9 As shown, the electronic device 900 may include one or more of the following components: a processor 901, a memory 902 and a communication interface 903. The processor 901, the memory 902 and the communication interface 903 are interconnected and perform communication work with each other, wherein the memory 902 may store one or more computer programs, and the one or more computer programs may be configured to implement the methods described in the above embodiments when executed by one or more processors 901.

[0128] The processor 901 may include one or more processing cores. The processor 901 uses various interfaces and lines to connect the various parts of the entire electronic device 900, and executes various functions and processes data of the electronic device 900 by running or executing instructions, programs, code sets or instruction sets stored in the memory 902, and calling data stored in the memory 902. Optionally, the processor 901 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 901 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processor (Graphics Processing Unit, GPU) and a modem. It can be understood that the above-mentioned modem may not be integrated into the processor 901, and can be implemented separately through a communication chip.

[0129] The memory 902 may include a random access memory (RAM) or a read-only memory (ROM). The memory 902 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 902 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc. The data storage area may also store data created by the electronic device 900 during use, etc.

[0130] It is understandable that the electronic device 900 may include more or fewer structural elements than those in the above structural block diagram, for example, a power module, physical buttons, a WiFi (Wireless Fidelity) module, a speaker, a Bluetooth module, a sensor, etc., which are not limited here.

[0131] The electronic device 900 may be a server or a part of a server.

[0132] An embodiment of the present application provides a computer-readable storage medium, wherein program data is stored in the computer-readable storage medium, and when the program data is executed by a processor, the program data is used to execute part or all of the steps of any one of the information push methods recorded in the above method embodiments.

[0133] The present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps of any information push method described in the above method embodiment. The computer program product may be a software installation package.

[0134] It should be noted that, for the method embodiments of any of the aforementioned information push methods, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by this application.

[0135] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "one" does not exclude multiple situations. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0136] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the method embodiments of any of the above-mentioned information push methods can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable memory, and the memory can include: a flash drive, a read-only memory (English: Read-Only Memory, abbreviated as: ROM), a random access memory (English: Random Access Memory, abbreviated as: RAM), a disk or an optical disk, etc.

[0137] The embodiments of the present application are introduced in detail above. Specific examples are used in this article to illustrate the principle and implementation method of a method and device for pushing information of the present application. The description of the above embodiments is only used to help understand the method and its core idea of ​​the present application. At the same time, for general technical personnel in this field, according to the idea of ​​a method and device for pushing information of the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of the methods, hardware products, and computer program products of the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0141] It can be understood that any product that is controlled or configured to execute the processing method of the flowchart described in the method embodiment of an information push method of the present application, such as the terminal and computer program product in the above flowchart, falls within the scope of the related products described in the present application.

[0142] Obviously, those skilled in the art can make various changes and modifications to the information push method and device provided by this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include these modifications and variations.

Claims

1. An information push method, characterized in that: The method comprises: Acquire information push data, wherein the information push data includes first relevant data corresponding to the target object and second relevant data corresponding to the information to be pushed; Generate a predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data, wherein the predicted matching score is associated with a target matching degree and a target activity degree, wherein the target matching degree is used to characterize the matching degree between the information to be pushed and the target object, and the target activity degree is used to characterize the platform activity degree of the information to be pushed and the target object; A push strategy for the information to be pushed is determined based on the predicted matching score.

2. The method according to claim 1, characterized in that The target matching degree is determined according to text similarity, behavior similarity, and geographic similarity between the first relevant data and the second relevant data.

3. The method according to claim 2, characterized in that The first related data includes first text data, first behavior data and first geographic data, and the second related data includes second text data, second behavior data and second geographic data, wherein: The text similarity is determined based on the first text data and the second text data; The behavior similarity is determined based on the first behavior data and the second behavior data; The geographic similarity is determined based on the first geographic data and the second geographic data.

4. The method according to claim 1, characterized in that: The first related data includes first behavior data, and the second related data includes an average click rate and an average conversion rate of the information to be pushed, wherein: The platform activity of the target object in the target activity is determined based on a feature value corresponding to a feature included in the first behavior data; The platform activity of the information to be pushed in the target activity is determined based on the average click rate and average conversion rate of the information to be pushed.

5. The method according to claim 1, characterized in that Before generating a predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data, the method further includes: Acquire training data, the training data including first historical related data, second historical related data, and a matching score, the first historical related data being historical related data corresponding to the target push information, the second historical related data being historical related data corresponding to the target object, and the matching score being determined based on a click rate and a conversion rate of the target object to the target push information; Performing model training on the initial model based on the training data to obtain a matching score prediction model after training; Generating the predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data includes: The first relevant data and the second relevant data are input into the matching score prediction model to obtain a predicted matching score corresponding to the information to be pushed.

6. The method according to claim 5, characterized in that The first related data includes first behavior data, the matching score prediction model includes the historical platform activity of the target object, and the historical platform activity of the target object is determined based on the historical related data corresponding to the target object, wherein: The platform activity of the target object in the target activity is determined based on a feature value corresponding to a feature included in the first behavior data, including: The platform activity of the target object in the target activity is determined based on a feature value corresponding to a feature included in the first behavior data and the historical platform activity.

7. The method according to claim 1, characterized in that The predicted matching score is the sum of multiple sub-prediction matching scores, and the multiple sub-prediction matching scores are generated based on information push data from multiple different platforms.

8. An information push device, characterized in that: The device comprises: An acquisition unit, configured to acquire information push data, wherein the information push data includes first relevant data corresponding to the target object and second relevant data corresponding to the information to be pushed; a processing unit, configured to generate a predicted matching score corresponding to the information to be pushed according to the first relevant data and the second relevant data, wherein the predicted matching score is associated with a target matching degree and a target activity degree, wherein the target matching degree is used to characterize a matching degree between the information to be pushed and the target object, and the target activity degree is used to characterize a platform activity degree of the information to be pushed and the target object; A pushing unit is used to determine a pushing strategy for the information to be pushed based on the predicted matching score.

9. An electronic device, characterized in that: The device comprises: A processor, a memory, and a communication interface, wherein the processor, the memory, and the communication interface are connected to each other and perform communication work between them; The memory stores executable program code, and the communication interface is used for wireless communication; The processor is used to call the executable program code stored in the memory to execute the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: A computer program for electronic data exchange is stored, wherein the computer program enables a computer to execute the method according to any one of claims 1 to 7.

Images