A training method of a prediction model, a method and apparatus for information recommendation

Abstract

The specification discloses a training method of a prediction model, a method and device for information recommendation, and specifically discloses the following: obtaining a training sample, inputting historical business data and candidate information into a prediction model to be trained, determining, as a business execution rate corresponding to the candidate information, a probability that a user clicks the candidate information and executes a business corresponding to the candidate information based on an environment in which the user executes the business corresponding to the historical business data through the prediction model, and minimizing a deviation between the business execution rate and label information of the training sample as an optimization target to train the prediction model. In this way, when the candidate information is recommended to the user through the business execution rate predicted by the prediction model, the influence of the environment in which the user is located on the preference of the user is considered, so that the accuracy of the prediction model is improved.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a method for training a prediction model, a method for information recommendation, and an apparatus. Background Technology

[0002] In existing recommendation services, when pushing merchants to users, it is usually based on a pre-trained predictive model. This model learns the user's interests and preferences using historical business data, then determines the matching degree between the user's interests and preferences and the information to be recommended. This predicts the user's business execution rate for each piece of information to be recommended, and prioritizes recommending information with high execution rates. A higher execution rate indicates a higher probability that the user will continue to place an order after clicking on the recommended information.

[0003] However, when the aforementioned prediction model recommends information to users, it only considers whether the user's interests and preferences match the content of historical recommendations, without taking into account the user's environment when recommending information, the impact of the recommendations the user browses and clicks, and the recommendations the user continues to execute after clicking. As a result, the business execution rate predicted by the prediction model has a large deviation, and the information ultimately recommended to the user may not match the user's actual habits. Summary of the Invention

[0004] This specification provides a method for training a prediction model, a method for recommending information, and an apparatus to partially solve the aforementioned problems existing in the prior art.

[0005] The following technical solution is adopted in this specification:

[0006] This manual provides a method for training a prediction model, including:

[0007] Obtain training samples, which contain the user's historical business data and candidate information to be recommended to the user;

[0008] The historical business data and the candidate information are input into the prediction model to be trained. Based on the environment in which the user is in when executing the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information, which is used as the business execution rate corresponding to the candidate information.

[0009] The prediction model is trained with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples.

[0010] Optionally, the prediction model includes a first network;

[0011] Based on the environment in which the user executes the service corresponding to the historical service data, the prediction model determines the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information, which is taken as the service execution rate corresponding to the candidate information. Specifically, this includes:

[0012] Based on the historical business data, the historical recommendation information that the user clicked before recommending the candidate information to the user is determined as the first historical recommendation information, and the historical recommendation information corresponding to the business performed by the user is determined as the second historical recommendation information.

[0013] For each first historical recommendation, determine the service loss characteristics when the user performs the service corresponding to the first historical recommendation; and for each second historical recommendation, determine the service loss characteristics when the user performs the service corresponding to the second historical recommendation.

[0014] The service loss characteristics corresponding to each first historical recommendation information, the service loss characteristics corresponding to each second historical recommendation information, and the environmental data of the user's environment when recommending the candidate information to the user are input into the first network, so as to determine the service execution rate corresponding to the candidate information based on the first network and the environment in which the user is located when executing the service corresponding to the historical service data.

[0015] Optionally, the service loss characteristics corresponding to each first historical recommendation information, the service loss characteristics corresponding to each second historical recommendation information, and the environmental data of the user's environment when recommending the candidate information to the user are input into the first network, so as to determine the service execution rate corresponding to the candidate information based on the first network, based on the environment in which the user is in when executing the service corresponding to the historical service data, specifically including:

[0016] For each first historical recommendation, determine the similarity between the environment in which the user clicked the first historical recommendation and the environment corresponding to the environment data; and for each second historical recommendation, determine the similarity between the environment in which the user executed the second historical recommendation and the environment corresponding to the environment data.

[0017] Based on the business loss characteristics corresponding to the first historical recommendation information whose similarity meets preset conditions, and the business loss characteristics corresponding to the second historical recommendation information whose similarity meets preset conditions, the business execution rate corresponding to the candidate information is determined based on the environment in which the user is in when executing the business corresponding to the historical business data.

[0018] Optionally, the service loss characteristics corresponding to each first historical recommendation information, the service loss characteristics corresponding to each second historical recommendation information, and the environmental data of the user's environment when recommending the candidate information to the user are input into the first network, so as to determine the service execution rate corresponding to the candidate information based on the first network, based on the environment in which the user is in when executing the service corresponding to the historical service data, specifically including:

[0019] For each second historical recommendation, determine the correlation between the second historical recommendation and each first historical recommendation;

[0020] Based on the correlation between the second historical recommendation information and each first historical recommendation information, the business loss characteristics corresponding to the second historical recommendation information are adjusted.

[0021] Based on the business loss characteristics corresponding to each first historical recommendation, the adjusted business loss characteristics corresponding to each second historical recommendation, and the environmental data, the business execution rate corresponding to the candidate information is determined based on the environment in which the user is in when executing the business corresponding to the historical business data.

[0022] Optionally, the business loss characteristics include those used to characterize the amount a user spends at the merchant to which the historical recommendation information belongs and / or the distance between the merchant to which the historical recommendation information belongs and the user when the historical recommendation information is recommended to the user.

[0023] Optionally, the prediction network further includes: a second network;

[0024] The historical business data and the candidate information are input into a prediction model to be trained. Based on the environment in which the user executes the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information. This probability is used as the business execution rate corresponding to the candidate information. Specifically, this includes:

[0025] Based on the historical business data, the historical recommendation information that the user clicked before recommending the candidate information to the user is determined as the first historical recommendation information, and the historical recommendation information corresponding to the business performed by the user is determined as the second historical recommendation information.

[0026] For each first historical recommendation, the distance between the user and the merchant to which the first historical recommendation belongs is determined as the distance corresponding to the first historical recommendation. For each second historical recommendation, the distance between the user and the merchant to which the second historical recommendation belongs is determined as the distance corresponding to the second historical recommendation.

[0027] Based on the distance corresponding to each first historical recommendation and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information is determined, which is used as the business execution rate corresponding to the candidate information.

[0028] Optionally, based on the distance corresponding to each first historical recommendation and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information is determined as the service execution rate corresponding to the candidate information, specifically including:

[0029] When recommending the candidate information to the user, the distance between the user and the merchant to which the candidate information belongs is used as the target distance.

[0030] Based on the similarity between the target distance and the distance corresponding to each first historical recommendation, and the similarity between the target distance and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information is determined, and is used as the service execution rate corresponding to the candidate information.

[0031] Optionally, based on the similarity between the target distance and the distance corresponding to each first historical recommendation, and the similarity between the target distance and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information is determined as the service execution rate corresponding to the candidate information, specifically including:

[0032] Based on the similarity between the target distance and the distance corresponding to each first historical recommendation, the attention weight corresponding to each first historical recommendation is determined; and based on the similarity between the target distance and the distance corresponding to each second historical recommendation, the attention weight corresponding to each second historical recommendation is determined.

[0033] Based on the attention weight corresponding to each first historical recommendation, the distances corresponding to each first historical recommendation are weighted to obtain the weighted distances corresponding to each first historical recommendation. Based on the attention weight corresponding to each second historical recommendation, the distances corresponding to each second historical recommendation are weighted to obtain the weighted distances corresponding to each second historical recommendation.

[0034] Based on the weighted distance corresponding to each first historical recommendation and each second historical recommendation, the probability that the user clicks on the candidate information and executes the corresponding business is determined, which is used as the business execution rate corresponding to the candidate information.

[0035] Optionally, based on the similarity between the target distance and the distance corresponding to each first historical recommendation, and the similarity between the target distance and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information is determined as the service execution rate corresponding to the candidate information, specifically including:

[0036] Using a preset sliding window, each first sub-distance sequence is determined from the first distance sequence composed of the distances corresponding to each first historical recommendation information, and each second sub-distance sequence is determined from the second distance sequence composed of the distances corresponding to each second historical recommendation information;

[0037] For each first sub-distance sequence, the distance feature between the target distance and the first sub-distance sequence is determined based on the similarity between the target distance and the distance corresponding to each first historical recommendation information contained in the first sub-distance sequence; and for each second sub-distance sequence, the distance feature between the target distance and the second sub-distance sequence is determined based on the similarity between the target distance and the distance corresponding to each second historical recommendation information contained in the second sub-distance sequence.

[0038] Based on the distance features corresponding to each first sub-distance sequence and each second sub-distance sequence, the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information is determined, and is used as the business execution rate corresponding to the candidate information.

[0039] Optionally, the historical business data and the candidate information are input into a prediction model to be trained. Based on the environment in which the user executes the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information, which is then used as the business execution rate corresponding to the candidate information. Specifically, this includes:

[0040] The historical business data and the candidate information are input into the prediction model to be trained, so that the prediction model determines the probability that the user clicks on the candidate information based on the environment in which the user performs the business corresponding to the historical business data, which is the predicted click-through rate of the candidate information; and the probability that the user performs the candidate information after the user clicks on the candidate information, which is the business conversion rate of the candidate information.

[0041] Based on the predicted click-through rate and the business conversion rate, the business execution rate corresponding to the candidate information is determined.

[0042] Optionally, before training the prediction model, with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples, the method further includes:

[0043] Based on the environment in which the user is in when performing the business corresponding to the historical business data, the prediction model predicts the probability of the user clicking on the candidate information, which is then used as the predicted click-through rate corresponding to the historical recommendation information.

[0044] The prediction model is trained with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples, specifically including:

[0045] The prediction model is trained with the optimization objectives of minimizing the deviation between the predicted click-through rate and the click-through rate label, and minimizing the deviation between the business execution rate and the business execution rate label.

[0046] This specification provides a method for information recommendation, including:

[0047] Obtain the candidate information that needs to be recommended to the user, as well as the user's historical business data;

[0048] For each candidate information, the candidate information and the historical business data are input into a pre-trained prediction model, so that the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the environment in which the user is in when the user executes the business corresponding to the historical business data. This probability is used as the business execution rate corresponding to the candidate information. The prediction model is trained by the above method.

[0049] Based on the business execution rate corresponding to each candidate information, information is recommended to the user.

[0050] This specification provides a training apparatus for a prediction model, comprising:

[0051] The acquisition module is used to acquire training samples, which contain the user's historical business data and candidate information recommended to the user;

[0052] The business execution rate prediction module is used to input the historical business data and the candidate information into the prediction model to be trained, so that, based on the environment in which the user is in when executing the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information, and uses the probability as the business execution rate corresponding to the candidate information.

[0053] The training module is used to train the prediction model with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples.

[0054] This specification provides an information recommendation device, comprising:

[0055] The acquisition module is used to acquire the candidate information that needs to be recommended to the user and the user's historical business data;

[0056] The business execution rate prediction module is used to input each candidate information and the historical business data into a pre-trained prediction model, so that the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the environment in which the user executes the business corresponding to the historical business data. This probability is used as the business execution rate corresponding to the candidate information. The prediction model is trained by the above method.

[0057] The recommendation module is used to recommend information to the user based on the business execution rate corresponding to each candidate information.

[0058] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the training method for the above-described prediction model and the information recommendation method.

[0059] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the training method for the aforementioned prediction model and the information recommendation method.

[0060] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects:

[0061] In the training method of the prediction model and the information recommendation method provided in this specification, training samples are obtained, and historical business data and candidate information are input into the prediction model to be trained. Based on the environment in which the user performed the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on candidate information and performs the corresponding business, which is taken as the business execution rate corresponding to the candidate information. The optimization objective is to minimize the deviation between the business execution rate and the label information of the training samples. Then, when recommending information to the user, for each candidate information to be recommended, the candidate information and the user's historical business data are input into the pre-trained prediction model. This allows the prediction model to determine the probability that the user clicks on the candidate information and performs the corresponding business, based on the environment in which the user performed the business corresponding to the historical business data, which is taken as the business execution rate corresponding to that candidate information. Information is then recommended to the user based on the business execution rates corresponding to each candidate information.

[0062] As can be seen from the above method, when recommending candidate information to users based on the business execution rate predicted by the prediction model, this method takes into account the user's preferences in the user's environment, thereby improving the accuracy of the prediction model. Attached Figure Description

[0063] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings:

[0064] Figure 1 This is a schematic diagram of the predictive model involved in this specification;

[0065] Figure 2 This is a flowchart illustrating a training method for a prediction model described in this specification.

[0066] Figure 3 This is a flowchart illustrating one method of information recommendation in this specification;

[0067] Figure 4 This is a schematic diagram of a training device for a prediction model provided in this specification;

[0068] Figure 5 A schematic diagram of an information recommendation device provided in this specification;

[0069] Figure 6 The corresponding information provided in this specification Figure 1 or Figure 3 A schematic diagram of an electronic device. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0071] To address the issue that the prediction model's prediction of service execution rate has significant deviations because it fails to consider the user's environment when recommending information, thus affecting the information the user browses and clicks, and the information the user continues to access after clicking, this specification provides a prediction model. This model includes an interest subnetwork, a first network, a second network, and a prediction subnetwork. See [link to documentation]. Figure 1 When training the prediction model, training samples are obtained, which contain the user's historical business data and candidate information recommended to the user. The user's historical business data and candidate information are then input into the prediction model to be trained. Based on the environment in which the user performed the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and performs the business corresponding to the candidate information. This probability is taken as the business execution rate corresponding to the candidate information. Finally, the prediction model is trained with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples.

[0072] In determining the business execution rate, the prediction model inputs the user's historical business data, candidate information, and environmental data of the user's environment when the candidate information was recommended to the user into the interest sub-network to determine the user's basic preference characteristics for the candidate information. Simultaneously, the user's historical business data, the candidate information, and the environmental data are input into the first network to determine the business loss characteristics assuming the user executes the business corresponding to the first historical recommendation information (i.e., the historical recommendation information that the user actually clicked before recommending the candidate information), and the business loss characteristics assuming the user executes the business corresponding to the second historical recommendation information (i.e., the historical recommendation information corresponding to the business that the user executed before recommending the candidate information). Furthermore, the user's historical business data is input into the second network to determine the distance corresponding to each first historical recommendation information and the distance corresponding to each second historical recommendation information. Finally, based on the outputs of the interest sub-network, online cost preference network, and offline cost preference network, the probability of a user clicking on the candidate information and executing the corresponding business is predicted and used as the business execution rate corresponding to the candidate information. The prediction model is then trained with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples.

[0073] Thus, the prediction model not only considers the degree of matching between the user's basic preferences and candidate information, but also the influence of the user's environment on the user's preferences when recommending candidate information to the user, thereby improving the accuracy of the prediction model.

[0074] O2O (Online to Offline) business is a business model that combines online and offline operations. When a user places an order, they typically select products online and then visit the merchant offline to purchase and complete the order. Thus, the user's costs include both the online payment cost and the offline travel cost. Therefore, because the aforementioned predictive model considers the distance between the user and the merchant when recommending candidate information, it can be applied to O2O (Online to Offline) businesses to predict the execution rate of services corresponding to clicked on candidate information, thereby informing information recommendations to the user.

[0075] The training scheme for the prediction model and the information recommendation scheme provided in this specification will be described in detail below with reference to the embodiments.

[0076] Figure 2 This is a flowchart illustrating a training method for a prediction model described in this specification, which includes the following steps:

[0077] Step S200: Obtain training samples, which contain the user's historical business data and candidate information recommended to the user.

[0078] The training methods for the prediction models and the information recommendation methods provided in this manual can be implemented by terminal devices such as desktop computers, or by servers or business platforms that provide business support for information recommendation services. For ease of description, the following explanation will focus solely on the business platform as the implementing entity for the training methods for the prediction models and the information recommendation methods provided in this manual.

[0079] In practice, the business platform first acquires training samples, which contain users' historical business data and candidate information to recommend to users. This candidate information can include various types of data such as advertisements, merchants, products, and user reviews to be recommended to users.

[0080] The user's historical transaction data records the transaction history of all transactions the user has performed. This transaction history includes the first historical recommendation information that the user has clicked and viewed, the distance between the user and the merchant associated with each first historical recommendation information, the average spending per person for each merchant associated with each first historical recommendation information, the group purchase amount associated with each merchant associated with each first historical recommendation information, the second historical recommendation information that the user has clicked and performed transactions with, the amount the user spent at each merchant associated with each second historical recommendation information, the average spending per person for each merchant associated with each second historical recommendation information, the average spending per person for each merchant associated with each second historical recommendation information, and the group purchase amount associated with each merchant associated with each second historical recommendation information.

[0081] In addition, the training sample may also contain user attribute information, which includes information that reflects the user's basic personal characteristics, such as the user's age, gender, city, and place of origin, etc., which will not be described in detail here.

[0082] The training sample will also include environmental data about the user's environment when the information is recommended to the user. This environmental data may include: weather information of the user's location (including the current temperature (e.g., 35°C) and weather conditions (e.g., rain, snow, strong winds, etc.)), the city code of the user's city, and the device model of the terminal device used by the user.

[0083] It should also be noted that this candidate information is also historical recommendation information recommended to users in the past. Based on whether the user clicked on the candidate information and whether the user executed the corresponding business after clicking on the candidate information, the label information of the training sample containing the candidate information can be determined.

[0084] Step S202: Input the historical business data and the candidate information into the prediction model to be trained, so that the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the environment in which the user executes the business corresponding to the historical business data, and uses the probability as the business execution rate corresponding to the candidate information.

[0085] This specification describes a prediction model comprising four networks: an interest subnetwork, a first network, a second network, and a prediction subnetwork. In practice, the business platform inputs the user's historical business data and the candidate information into the prediction model to be trained. The interest subnetwork determines the user's basic preference features for the candidate information; the first network determines the business loss features when the user executes the business corresponding to the historical recommendation information; and the second network determines the distance to the historical recommendation information where the user has executed the business. Finally, the basic preference features, business loss features, and the distance to the historical recommendation information where the user has executed the business are input into the prediction subnetwork to predict the probability that the user clicks on the candidate information and executes the business corresponding to that candidate information, which is then used as the business execution rate corresponding to that candidate information.

[0086] The process of determining the business execution rate corresponding to this candidate information will be explained in detail below.

[0087] First, the business platform needs to determine the user's basic preference characteristics through an interest subnetwork. Specifically, the business platform inputs the candidate information to be recommended to the user, the user's historical business data, user attribute information, and environmental data of the user's environment when the candidate information was recommended into the interest subnetwork. This interest subnetwork encodes the historical business data and user attribute information to obtain the user profile features U, encodes the candidate information to obtain the recommendation information features S, and encodes the environmental data to obtain the environmental features C corresponding to the user's environment when the candidate information was recommended. Then, the interest subnetwork concatenates the user profile features U, the recommendation information features S, and the environmental features C, and extracts features through a fully connected layer to obtain the user's interest preference features b for the candidate information. u,s Among them, user interest preference feature b for this candidate information. u,s It can be characterized as: b u,s =ELU(MLP(concat(U,S,C))), where ELU() is the activation function and MLP() indicates that the input data is processed using a multilayer perceptron.

[0088] It should be noted that in this specification, when encoding continuous data, the continuous data needs to be discretized first, and then encoded. However, for discrete data, the encoding process can be performed directly.

[0089] At the same time, the business platform also needs to use the first network to determine the business loss characteristics corresponding to the historical recommendation information that the user clicked before recommending the candidate information to the user, and the business loss characteristics corresponding to the historical recommendation information that the user performed.

[0090] In practice, the business platform first determines, based on historical business data, the historical recommendation information clicked by the user before recommending candidate information to the user, as the first historical recommendation information. For each first historical recommendation information, the platform determines the business loss characteristics assuming the user would execute the business corresponding to that first historical recommendation information. Simultaneously, the business platform determines the historical recommendation information corresponding to the business executed by the user before recommending candidate information to the user, as the second historical recommendation information. For each second historical recommendation information, the platform determines the business loss characteristics assuming the user would execute the business corresponding to that second historical recommendation information. Then, the business platform inputs the determined business loss characteristics for each first historical recommendation information, the business loss characteristics for each second historical recommendation information, and the environmental data of the user's environment when recommending candidate information to the user into the first network of the prediction model. Based on this first network, the platform determines the business execution rate corresponding to the candidate information, based on the environment in which the user was in when executing the business corresponding to the historical business data.

[0091] The business loss characteristics of historical recommendation information include features representing the amount a user spends at the merchant associated with the historical recommendation and / or the distance between the merchant and the user when the historical recommendation was recommended. In actual business, the amount a user spends at the merchant associated with the historical recommendation in the first historical recommendation information business loss can be the statistically calculated average order value, the assumed cost the user would have paid when performing the transaction, etc. The amount a user spends at the merchant associated with the historical recommendation information in the second historical recommendation information business loss is the actual amount the user spends.

[0092] Thus, the business loss characteristics when a user executes the business corresponding to historical recommendation information can be characterized as follows:

[0093]

[0094] This represents the merchant identifier of the merchant to which the i-th historical recommendation belongs when the user executes the service corresponding to the i-th historical recommendation.

[0095] This represents the distance between the merchant to which the i-th historical recommendation belongs and the user when the user performs the service corresponding to the i-th historical recommendation.

[0096] This characterizes the amount a user spends at the merchant associated with the i-th historical recommendation when executing the service corresponding to that recommendation.

[0097] Thus, the business loss characteristics corresponding to each first historical recommendation can be determined using the above formula. Business loss characteristics corresponding to each second historical recommendation information Next, the business loss characteristics corresponding to each first historical recommendation information will be analyzed. Business loss characteristics corresponding to each second historical recommendation information The information is input into a first network, and based on this first network, the execution rate of the corresponding business information is determined based on the environment in which the user executes the business corresponding to the historical business data.

[0098] Before determining the business execution rate corresponding to the candidate information based on the first network and the business loss characteristics, the business platform can further filter the first historical recommendation information and the second historical recommendation information to determine the first historical recommendation information that matches the environmental data of the user's environment when recommending the candidate information to the user, and the second historical recommendation information that matches the environmental data of the user's environment when recommending the candidate information to the user. Then, based on the filtered first historical recommendation information and second historical recommendation information, the business execution rate corresponding to the candidate information is determined.

[0099] In practice, the business platform can determine the similarity between the environment in which the user clicked on each first historical recommendation and the environment corresponding to the corresponding environmental data for each first historical recommendation. Then, it filters out the business loss features corresponding to the first historical recommendation that meet preset similarity criteria. Simultaneously, for each second historical recommendation, it determines the similarity between the environment in which the user executed the second historical recommendation and the environment corresponding to the corresponding environmental data for each second historical recommendation, and again filters out the business loss features corresponding to the second historical recommendation that meet preset similarity criteria. Next, based on the business loss features corresponding to the first and second historical recommendations that meet preset similarity criteria, it determines the business execution rate corresponding to the candidate information, based on the environment in which the user executed the business corresponding to the historical business data.

[0100] Specifically, determining the similarity between the environment in which the user clicks on the first historical recommendation and the environment corresponding to the environmental data, as well as the similarity between the environment in which the user executes the second historical recommendation and the environment corresponding to the environmental data, can both be achieved through a sparse self-attention mechanism.

[0101] Specifically, the formula for this sparse self-attention mechanism is as follows:

[0102]

[0103] In the above formula, topk(·) represents taking the k historical recommendation information that is most relevant to the environmental data of the user's environment when the candidate information is recommended to the user;

[0104] Q can be used to characterize query features;

[0105] K T It can be used to represent the transpose of the queried key;

[0106] V can be used to represent the value of the key being queried;

[0107] It can be used to characterize the square root of a vector dimension.

[0108] Thus, the business loss feature corresponding to the first historical recommendation information whose similarity meets the preset conditions is represented as:

[0109]

[0110] Among them, Q c =σ(W c ×C+b c The query condition represents the environmental characteristics of the user's environment when the candidate information is recommended to the user;

[0111] The key of the query is composed of the business loss characteristics corresponding to each first historical recommendation information;

[0112] This represents the value of the query key, which is composed of the business loss characteristics corresponding to each first historical recommendation.

[0113] W c , And, b c , It is a learnable parameter matrix;

[0114] H c This refers to the business loss characteristics when a user performs a service corresponding to the first historical recommendation information clicked in the past.

[0115] Thus, the output of Sparse attention is the business loss of the first historical recommendation information corresponding to the environmental features of the user's environment when recommending candidate information to the user.

[0116] Similarly, the business loss feature corresponding to the first historical recommendation information whose similarity meets the preset conditions is represented as:

[0117]

[0118] Among them, Q c =σ(W c ×C+b c The query condition represents the environmental characteristics of the user's environment when the candidate information is recommended to the user;

[0119] The key representing the query is composed of the business loss characteristics corresponding to each second historical recommendation information;

[0120] This represents the value of the query key, which is composed of the business loss characteristics corresponding to each second historical recommendation.

[0121] W c , And, b c , This is the parameter matrix that needs to be learned;

[0122] H o The business loss characteristics when corresponding to the second historical recommendation information of the business that the user has performed in the past.

[0123] Thus, the output of Sparse attention is the business loss of the business corresponding to the several second historical recommendation information that are most relevant to the environmental features of the user's environment when recommending candidate information to the user.

[0124] Thus, the business loss feature corresponding to the first historical recommendation information whose similarity meets the preset conditions can be characterized as: in, The service loss characteristic is used to characterize the service performed by a user when executing the service corresponding to the i-th historical recommendation in the first historical recommendation information whose similarity meets a preset condition. The service loss characteristic corresponding to the second historical recommendation information whose similarity meets a preset condition can be characterized as follows: Characterized in, The business loss characteristic is used to characterize the business corresponding to the i-th first historical recommendation information in the first historical recommendation information where the user performs the service with similarity satisfying the preset conditions.

[0125] In this way, by using only the business loss features corresponding to the first historical recommendation information that meets the preset similarity conditions, and the business loss features corresponding to the first historical recommendation information with similarity, the business execution rate corresponding to the candidate information can be determined, which can reduce the amount of resources used. At the same time, the extracted business loss features are closer to the business loss of the current user's actual preferences, thus improving the accuracy of the prediction model.

[0126] Furthermore, in actual business operations, there is a certain conversion pattern between the first historical recommendation information clicked by the user and the second historical recommendation information of the business actually performed by the user. In this specification, the first network will be used to learn the correlation between the first historical recommendation information clicked by the user and the second historical recommendation information of the business performed by the user, so as to help determine the business execution rate corresponding to the candidate information and improve the accuracy of the prediction model.

[0127] In practice, the business platform determines the correlation between each second historical recommendation and each first historical recommendation for each second historical recommendation. Then, based on the correlation between the second historical recommendation and each first historical recommendation, it adjusts the business loss characteristics corresponding to the second historical recommendation. Finally, based on the business loss characteristics corresponding to each first historical recommendation, the adjusted business loss characteristics corresponding to each second historical recommendation, and environmental data, it determines the business execution rate corresponding to the candidate information based on the environment in which the user executes the business corresponding to the historical business data.

[0128] In practice, the business platform can use the following two methods to adjust the business loss characteristics corresponding to the second historical recommendation information.

[0129] Method 1: For each second historical recommendation, determine the correlation between the second historical recommendation and each first historical recommendation.

[0130] In practice, for each second historical recommendation, the business platform determines its correlation with each first historical recommendation based on the corresponding business loss characteristics and the business loss characteristics between the second and first historical recommendations. Then, based on this correlation, it determines an adjustment coefficient for each first historical recommendation relative to the second historical recommendation. Finally, for each first historical recommendation, the platform weights its corresponding business loss based on the adjustment coefficient, and then adjusts the business loss characteristics of the second historical recommendation based on the weighted business loss characteristics of each first historical recommendation, resulting in the adjusted business loss characteristics for that second historical recommendation.

[0131] The formula for the adjustment coefficient of the i-th first historical recommendation information for the j-th second historical recommendation information is as follows:

[0132]

[0133] in, This represents the business loss feature corresponding to the i-th first historical recommendation information in the first historical recommendation information whose similarity meets the preset conditions;

[0134] This represents the business loss feature corresponding to the j-th second historical recommendation information in the second historical recommendation information whose similarity meets the preset conditions;

[0135] This represents the correlation between the business loss feature corresponding to the i-th first historical recommendation clicked by the user and the business loss feature corresponding to the j-th second historical recommendation executed by the user.

[0136] This is the weight matrix to be learned.

[0137] Furthermore, the formula for determining the adjusted business loss characteristics corresponding to the j-th second historical recommendation is as follows:

[0138]

[0139] in, This represents the business loss feature corresponding to the i-th first historical recommendation information in the first historical recommendation information whose similarity meets the preset conditions;

[0140] This represents the business loss feature corresponding to the j-th second historical recommendation information in the second historical recommendation information whose similarity meets the preset conditions;

[0141] This represents the correlation between the business loss feature corresponding to the i-th first historical recommendation clicked by the user and the business loss feature corresponding to the j-th second historical recommendation executed by the user.

[0142] This represents the sum of the correlations between the business loss characteristics corresponding to each first historical recommendation clicked by the user and the business loss characteristics corresponding to each second historical recommendation executed by the user;

[0143] The correlation between the business loss feature corresponding to the i-th first historical recommendation information clicked by the user and the business loss feature corresponding to the j-th second historical recommendation information executed by the user, and the proportion of the correlation between the business loss features corresponding to each first historical recommendation information clicked by the user and the business loss features corresponding to each second historical recommendation information executed by the user, is called the adjustment coefficient of the i-th first historical recommendation information for the j-th second historical recommendation information.

[0144] This represents the weighted business loss feature corresponding to the i-th first historical recommendation information, obtained by using the adjustment coefficient of the i-th first historical recommendation information for the j-th second historical recommendation information and weighting the business loss corresponding to the i-th first historical recommendation information.

[0145] This means that the weighted business loss feature corresponding to each first historical recommendation is used to adjust the business loss feature corresponding to the j-th second historical recommendation that the user has executed, so as to obtain the adjusted business loss feature corresponding to the j-th second historical recommendation.

[0146] The above method comprehensively considers the overall correlation between each first merchant and each second merchant. The following will also provide the features corresponding to each type of merchant information within the transaction cost feature to determine the correlation between the first merchant and the second merchant under this type of merchant information.

[0147] Method 2: For each second feature in the business loss corresponding to each second historical recommendation information, determine the correlation between the second feature and each second feature in the business loss corresponding to each first historical recommendation information.

[0148] In practice, the business platform determines the correlation between each second sub-feature in the business loss features corresponding to the second historical recommendation information and each first sub-feature in the same business loss features. Then, based on this correlation, it determines an adjustment coefficient for each first sub-feature relative to each second sub-feature. Finally, for each first sub-feature, it weights it according to the adjustment coefficient for that second sub-feature, and then adjusts the second sub-feature based on each weighted first sub-feature to obtain the adjusted second sub-feature.

[0149] Among them, the business loss characteristics corresponding to the first historical recommendation information include three first sub-features, namely: in, This represents the merchant identifier feature in the business loss features corresponding to each first historical recommendation information. This represents the feature in the business loss characteristics corresponding to each first historical recommendation information that characterizes the distance between the merchant to which the historical recommendation information belongs and the user when recommending historical recommendation information to the user. This refers to the feature in the business loss features corresponding to each first historical recommendation information that represents the amount of money a user spends at the merchant to which the historical recommendation information belongs.

[0150] The business loss characteristics corresponding to the second historical recommendation information also include three second sub-features, namely: in, This represents the merchant identifier feature in the business loss features corresponding to each first historical recommendation information. This indicates that the feature corresponding to the business loss feature of each first historical recommendation is used to characterize the distance between the merchant to which the historical recommendation information belongs and the user when recommending historical recommendation information to the user. This refers to the feature in the business loss features corresponding to each first historical recommendation information that represents the amount of money a user spends at the merchant to which the historical recommendation information belongs.

[0151] Thus, for the second sub-feature of the ctxi dimension in the business loss feature corresponding to the second historical recommendation information, the formula for determining the correlation between this ctxi dimension feature and the first sub-feature of the ctxj dimension in the business loss feature corresponding to the first historical recommendation information is as follows:

[0152]

[0153] in, This represents the first sub-feature in the ctxi dimension of the business loss feature corresponding to the first historical recommendation information;

[0154] This represents the second sub-feature in the ctxi dimension of the business loss feature corresponding to the second historical recommendation information;

[0155] This represents the correlation between the second sub-feature of the ctxi dimension in the business loss feature corresponding to the second historical recommendation information and the first sub-feature of the ctxi dimension in the business loss feature corresponding to the first historical recommendation information.

[0156] The weight matrix to be learned.

[0157] Furthermore, the formula for determining the transaction cost characteristics of the second merchant after adjustment is as follows:

[0158]

[0159] in, This represents the first sub-feature in the ctxi dimension of the business loss feature corresponding to the first historical recommendation information;

[0160] This represents the second sub-feature of the ctxj-th dimension in the business loss feature corresponding to the second historical recommendation information;

[0161] This represents the correlation between the first sub-feature of the first dimension ctxi in the business loss feature corresponding to the first historical recommendation information and the second sub-feature of the second dimension ctxj in the business loss feature corresponding to the second historical recommendation information.

[0162] This represents the sum of correlations between each dimension of the first sub-feature in the business loss feature corresponding to the first historical recommendation information and the second sub-feature of the ctxj-th dimension in the business loss feature corresponding to the second historical recommendation information;

[0163] The correlation between the first sub-feature of dimension ctxi in the business loss feature corresponding to the first historical recommendation information and the second sub-feature of dimension ctxj in the business loss feature corresponding to the second historical recommendation information is the proportion of the correlation between each first sub-feature of dimension ctxi in the business loss feature corresponding to the first historical recommendation information and the second sub-feature of dimension ctxj in the business loss feature corresponding to the second historical recommendation information. This is called the adjustment coefficient of the first sub-feature of dimension ctxi for the second sub-feature of dimension ctxj.

[0164] This indicates that the adjustment coefficient of the first sub-feature in the ctxi dimension is used to weight the second sub-feature in the ctxj dimension;

[0165] This means that the first sub-feature after each weighting is weighted and summed, and the second sub-feature in dimension ctxj is also weighted.

[0166] For example, This indicates the distance between the user and the merchants associated with the historical recommendations they clicked, and the correlation between the distance between the user and the merchants associated with the historical recommendations they have performed transactions with. This indicates the distance between the user and the merchant associated with the historical recommendations they clicked, and the correlation between this information and the amount spent by the user at the merchant associated with the historical recommendations they made.

[0167] In this process, the business loss features corresponding to the first historical recommendation information can be either the business loss features determined by the business platform from historical business data for each first historical recommendation information, or the business loss features corresponding to the first historical recommendation information whose similarity meets preset conditions. Similarly, the business loss features corresponding to the second historical recommendation information can be either the business loss features determined by the business platform from historical business data for each second historical recommendation information, or the business loss features corresponding to the second historical recommendation information whose similarity meets preset conditions.

[0168] In actual business operations, the business platform can concatenate the adjusted business loss features corresponding to the second historical recommendation information (adjusted through the two methods mentioned above) with each adjusted second sub-feature to obtain the adjusted business loss features corresponding to each second historical recommendation information. Then, the obtained features are pooled, and based on the pooled features, the business execution rate corresponding to the candidate information is determined, taking into account the environment in which the user executed the business corresponding to the historical business data.

[0169] In addition, the business platform will also use the second network to extract the distance between the user and the merchant to which the first historical recommendation information belongs when the user executes the business corresponding to the first historical recommendation information, and the distance between the user and the merchant to which the second historical recommendation information belongs when the user executes the business corresponding to the second historical recommendation information. Based on the extracted distances, the platform will determine the business execution rate corresponding to the candidate information based on the environment in which the user was in when executing the business corresponding to the historical business data.

[0170] In specific implementation, the business platform determines, based on historical business data, the historical recommendation information clicked by the user before recommending candidate information to the user, as the first historical recommendation information, and the historical recommendation information corresponding to the business performed by the user, as the second historical recommendation information. Then, for each first historical recommendation information, the distance between the user and the merchant to which the first historical recommendation information belongs, assuming the user performs the business corresponding to the first historical recommendation information, is determined as the distance corresponding to the first historical recommendation information. Similarly, for each second historical recommendation information, the distance between the user and the merchant to which the second historical recommendation information belongs, is determined as the distance corresponding to the second historical recommendation information. Finally, based on the distances corresponding to each first and second historical recommendation information, the probability that the user clicks on candidate information and performs the business corresponding to the candidate information is determined as the business execution rate corresponding to the candidate information.

[0171] Specifically, the business platform can determine the distance between the user and the merchant to which the candidate information belongs when recommending the candidate information to the user, as the target distance. Then, based on the similarity between the target distance and the distance corresponding to each first historical recommendation information, and the similarity between the target distance and the distance corresponding to each second historical recommendation information, the probability of the user clicking on the candidate information and executing the business corresponding to the candidate information is determined.

[0172] Furthermore, the business platform can determine the attention weight for each first historical recommendation based on the similarity between the target distance and the distance corresponding to each first historical recommendation, and determine the attention weight for each second historical recommendation based on the similarity between the target distance and the distance corresponding to each second historical recommendation. Then, based on the attention weight for each first historical recommendation, the distances corresponding to each first historical recommendation are weighted to obtain a weighted distance for each first historical recommendation, and similarly, based on the attention weight for each second historical recommendation, the distances corresponding to each second historical recommendation are weighted to obtain a weighted distance for each second historical recommendation. Finally, based on the weighted distances for each first and second historical recommendation, the probability that a user clicks on a candidate recommendation and executes the corresponding business function is determined, which serves as the business execution rate for the candidate recommendation.

[0173] In actual business operations, the business platform uses a preset sliding window to determine each first sub-distance sequence from the first distance sequence composed of the distances corresponding to each first historical recommendation information, and to determine each second sub-distance sequence from the second distance sequence composed of the distances corresponding to each second historical recommendation information. Then, for each first sub-distance sequence, the distance feature between the target distance and the first sub-distance sequence is determined based on the similarity between the target distance and the distances corresponding to each first historical recommendation information contained in the first sub-distance sequence. Similarly, for each second sub-distance sequence, the distance feature between the target distance and the second sub-distance sequence is determined based on the similarity between the target distance and the distances corresponding to each second historical recommendation information contained in the second sub-distance sequence. Finally, the business execution rate corresponding to the candidate information is determined based on the distance features corresponding to each first sub-distance sequence and each second sub-distance sequence.

[0174] The distance between the merchant and the user associated with the historical recommendation information can be denoted as d = Embedding(s) dis If the length of the sliding window is set to ws, the first distance sequence is divided according to the length of the sliding window to obtain several first sub-distance sequences, denoted as ws. And the second distance sequence is divided to obtain several second sub-distance sequences, denoted as...

[0175] Then, the similarity between the i-th first sub-distance sequence and the target distance can be characterized as:

[0176]

[0177] Correspondingly, the distance features between the first distance sequences can be represented as:

[0178] The similarity between the i-th sub-second sub-distance sequence and the target distance can be characterized as:

[0179]

[0180] Correspondingly, the distance features between the second distance sequences can be represented as:

[0181] Finally, based on the distance features between the first distance sequences and the distance features between the second distance sequences, the business execution rate corresponding to the historical recommendation information is determined.

[0182] Step S204: Train the prediction model with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples.

[0183] In practice, the business platform inputs historical business data and candidate information into the prediction model to be trained. The prediction model determines the probability that a user clicks on candidate information based on the environment in which the user performs the business corresponding to the historical business data, serving as the predicted click-through rate (CTR) for that candidate information. It also determines the probability that the user executes the candidate information after clicking it, serving as the business conversion rate. The predicted CTR and the conversion rate are then multiplied to determine the business execution rate for that candidate information. Finally, the prediction model is trained with the optimization objective of minimizing the deviation between the predicted business execution rate and the label information of the training samples.

[0184] Specifically, when a user clicks on a candidate and executes the corresponding business, the training sample's label for the business execution rate is 1; otherwise, it is 0.

[0185] In actual business operations, besides training the prediction model based on the deviation between the business execution rate and the label information of the training samples, it can also be trained based on the deviation between the predicted click-through rate and the label information of the training samples. In this case, if the user does not click on the candidate information, the label information of the training samples for the predicted click-through rate is 0, and the label information of the training samples for the business execution rate is also 0. If the user clicks on the candidate information but does not execute the corresponding business, the label information of the training samples for the predicted click-through rate is 1, and the label information of the training samples for the business execution rate is also 0. If the user clicks on the candidate information and executes the corresponding business, the label information of the training samples for the predicted click-through rate is 1, and the label information of the training samples for the business execution rate is also 1.

[0186] Through the above steps, when recommending information to users, the user's preferences in the user's environment and the correlation between the recommended information and the recommended information clicked by the user are taken into account, thereby improving the accuracy of the prediction model.

[0187] It should be noted that the business loss characteristics corresponding to the first historical recommendation information and the business loss characteristics corresponding to the second historical recommendation information whose similarity meets the preset conditions are all sequences obtained by sorting them in a certain order (such as the order of occurrence). For example, middle, This represents the business overhead characteristic when a user executes the service corresponding to the first historical recommendation. This represents the business loss characteristics when a user performs the service corresponding to the second historical recommendation, ... This represents the business loss characteristic when a user performs the service corresponding to the j-th first historical recommendation.

[0188] Based on the prediction model trained using the above training method, this manual also provides instructions on how to use the prediction model, as follows.

[0189] Figure 3 This is a flowchart illustrating one method of information recommendation in this specification, which specifically includes the following steps:

[0190] Step S300: Obtain the candidate information to be recommended to the user and the user's historical business data.

[0191] In practice, when a business platform needs to recommend information to a user, it first obtains several candidate information items to be recommended, as well as the user's historical business data. These candidate information items and historical recommendation information can include advertisements, merchants, products, user reviews, etc.

[0192] In actual business operations, the user's historical business data records the business records of all the business transactions the user has performed. These records include the first historical recommendation information that the user has clicked and viewed, the distance between the user and the merchant associated with each first historical recommendation information, the average spending per person for each merchant associated with each first historical recommendation information, the group-buying amount associated with each merchant associated with each first historical recommendation information, the second historical recommendation information that the user has clicked and performed on, the amount the user spent at each merchant associated with each second historical recommendation information, the average spending per person for each merchant associated with each second historical recommendation information, the average spending per person for each merchant associated with each second historical recommendation information, and the group-buying amount associated with each merchant associated with each second historical recommendation information.

[0193] In addition, the data input into the prediction model can also include user attribute information, which includes information that reflects the user's basic personal characteristics, such as: user age, user gender, user city, user place of origin, etc.

[0194] In addition, the data input into the prediction model by the business platform in this specification will also include environmental data of the user's environment when recommending information to the user. This environmental data may include: weather information of the user's location (including the current temperature (e.g., 35°C) and weather conditions (e.g., rain, snow, strong winds, etc.)), the city code of the user's city, and the device model of the terminal device used by the user.

[0195] Step S302: For each candidate information, the candidate information and the historical business data are input into a pre-trained prediction model, so that the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the environment in which the user executes the business corresponding to the historical business data, and uses this probability as the business execution rate corresponding to the candidate information.

[0196] The prediction model is obtained by training using the prediction model training method described above.

[0197] In practice, for each candidate information, the business platform inputs historical business data and the information into a prediction model. The prediction model then determines the user's basic preference characteristics, the business loss characteristics corresponding to the first historical recommendation information clicked by the user before recommending the candidate information, and the business loss characteristics corresponding to the second historical recommendation information corresponding to the business executed by the user. Based on the business loss characteristics corresponding to each first historical recommendation information and each second historical recommendation information, the environmental data of the user's environment when recommending the candidate information is input into a first network. Based on the first network, the business execution rate corresponding to the candidate information is determined based on the environment in which the user was in when executing the business corresponding to the historical business data.

[0198] Specifically, based on historical business data, the system determines the first set of historical recommendations that the user clicked before recommending candidate information, and the second set of historical recommendations corresponding to the services the user performed. Then, for each first set of historical recommendations, the system determines the business loss characteristics assuming the user performed the service corresponding to that first historical recommendation. Similarly, for each second set of historical recommendations, the system determines the business loss characteristics assuming the user performed the service corresponding to that second historical recommendation. Furthermore, for each first set of historical recommendations, the system determines the similarity between the environment in which the user clicked the first set of historical recommendations and the corresponding environment in the environmental data. For each second set of historical recommendations, the system determines the similarity between the environment in which the user performed the second set of historical recommendations and the corresponding environment in the environmental data. This process filters out the business loss characteristics corresponding to the first set of historical recommendations and the second set of historical recommendations whose similarity meets preset conditions.

[0199] Subsequently, for each second historical recommendation, the correlation between the second historical recommendation and each first historical recommendation is determined. Then, based on the correlation between the second historical recommendation and each first historical recommendation, the business loss characteristics corresponding to the second historical recommendation are adjusted. Finally, based on the business loss characteristics corresponding to each first historical recommendation, the adjusted business loss characteristics corresponding to each second historical recommendation, and environmental data, the business execution rate corresponding to the candidate information is determined based on the environment when the user executes the business corresponding to the historical business data.

[0200] At the same time, for each first historical recommendation, the business platform determines the distance between the user and the merchant to which the first historical recommendation belongs when the user performs the business corresponding to the first historical recommendation, as the distance corresponding to the first historical recommendation. For each second historical recommendation, the platform determines the distance between the user and the merchant to which the second historical recommendation belongs when the user performs the business corresponding to the second historical recommendation, as the distance corresponding to the second historical recommendation. The platform also determines the distance between the user and the merchant to which the candidate information belongs when recommending the candidate information to the user, as the target distance.

[0201] Then, through a preset sliding window, each first sub-distance sequence is determined from the first distance sequence composed of the distances corresponding to each first historical recommendation information, and each second sub-distance sequence is determined from the second distance sequence composed of the distances corresponding to each second historical recommendation information. For each first sub-distance sequence, the distance feature between the target distance and the first sub-distance sequence is determined based on the similarity between the target distance and the distances corresponding to each first historical recommendation information contained in the first sub-distance sequence. Similarly, for each second sub-distance sequence, the distance feature between the target distance and the second sub-distance sequence is determined based on the similarity between the target distance and the distances corresponding to each second historical recommendation information contained in the second sub-distance sequence. Based on the distance features between the target distance and the first and second sub-distance sequences, the business execution rate corresponding to the candidate information is determined. For a detailed explanation of this process, please refer to the prediction model training process; it will not be elaborated upon here.

[0202] Step S304: Recommend information to the user based on the business execution rate corresponding to each candidate information.

[0203] In practice, after the business platform determines the business execution rate corresponding to each piece of information, it can sort the information in descending order according to the size of the business execution rate. Then, based on the sorting results, it selects the target information to recommend to the user and recommends the target information to the user according to the arrangement of the display positions on the page where the information is displayed to the user.

[0204] Furthermore, the business platform can sort the information in descending order based on the predicted purchase rate corresponding to each piece of information. Then, according to the sorting results, information with a business execution rate greater than a set purchase rate threshold is selected as target information to recommend to users. This target information is then recommended to users based on its placement within the information display page. Of course, there are other methods for selecting target information, which will not be illustrated here.

[0205] It should be noted that all actions involving the acquisition of signals, information, or data in this manual are performed in accordance with the relevant data protection laws and regulations of the country where the device is located, and with the authorization of the owner of the relevant device.

[0206] The above describes one or more embodiments of the prediction model training method and information recommendation method provided in this specification. Based on the same idea, this specification also provides corresponding prediction model training devices and information recommendation devices, such as... Figure 4 and Figure 5 As shown.

[0207] Figure 4 This specification provides a schematic diagram of a training device for a prediction model, which specifically includes:

[0208] The acquisition module 400 is used to acquire training samples, which contain the user's historical business data and candidate information recommended to the user;

[0209] The business execution rate prediction module 401 is used to input the historical business data and the candidate information into the prediction model to be trained, so that, based on the environment in which the user is in when executing the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information, and uses the probability as the business execution rate corresponding to the candidate information.

[0210] The training module 402 is used to train the prediction model with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples.

[0211] Optionally, the prediction model includes a first network;

[0212] The business execution rate prediction module 401 is specifically used to determine, based on the historical business data, the historical recommendation information clicked by the user before recommending the candidate information to the user as first historical recommendation information, and the historical recommendation information corresponding to the business executed by the user as second historical recommendation information; for each first historical recommendation information, determine the business loss characteristics assuming the user executes the business corresponding to the first historical recommendation information, and for each second historical recommendation information, determine the business loss characteristics assuming the user executes the business corresponding to the second historical recommendation information; input the business loss characteristics corresponding to each first historical recommendation information, the business loss characteristics corresponding to each second historical recommendation information, and the environmental data of the user's environment when recommending the candidate information to the user into the first network, so as to determine, based on the first network, the business execution rate corresponding to the candidate information based on the environment in which the user was in when executing the business corresponding to the historical business data.

[0213] Optionally, the business execution rate prediction module 401 is specifically used to determine, for each first historical recommendation, the similarity between the environment in which the user clicks on the first historical recommendation and the environment corresponding to the environment data, and for each second historical recommendation, the similarity between the environment in which the user executes the second historical recommendation and the environment corresponding to the environment data; and based on the business loss characteristics corresponding to the first historical recommendation that meets the preset similarity conditions, and the business loss characteristics corresponding to the business loss of the second historical recommendation that meets the preset similarity conditions, to determine the business execution rate corresponding to the candidate information based on the environment in which the user executes the business corresponding to the historical business data.

[0214] Optionally, the business execution rate prediction module 401 is specifically used to determine the correlation between each second historical recommendation and each first historical recommendation for each second historical recommendation; adjust the business loss characteristics corresponding to the second historical recommendation based on the correlation between the second historical recommendation and each first historical recommendation; and determine the business execution rate corresponding to the candidate information based on the environment in which the user is in when executing the business corresponding to the historical business data, based on the business loss characteristics corresponding to each first historical recommendation, the adjusted business loss characteristics corresponding to each second historical recommendation, and the environmental data.

[0215] Optionally, the business loss characteristics include those used to characterize the amount a user spends at the merchant to which the historical recommendation information belongs and / or the distance between the merchant to which the historical recommendation information belongs and the user when the historical recommendation information is recommended to the user.

[0216] Optionally, the prediction network further includes: a second network;

[0217] The business execution rate prediction module 401 is specifically used to determine, based on the historical business data, the historical recommendation information clicked by the user before recommending the candidate information to the user, as first historical recommendation information, and the historical recommendation information corresponding to the business executed by the user, as second historical recommendation information; for each first historical recommendation information, determining the distance between the user and the merchant to which the first historical recommendation information belongs, assuming the user executes the business corresponding to the first historical recommendation information, as the distance corresponding to the first historical recommendation information; and for each second historical recommendation information, determining the distance between the user and the merchant to which the second historical recommendation information belongs, assuming the user executes the business corresponding to the second historical recommendation information, as the distance corresponding to the second historical recommendation information; and based on the distance corresponding to each first historical recommendation information and the distance corresponding to each second historical recommendation information, determining the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information, as the business execution rate corresponding to the candidate information.

[0218] Optionally, the business execution rate prediction module 401 is specifically used to determine the distance between the user and the merchant to which the candidate information belongs when recommending the candidate information to the user, as the target distance; and to determine the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the similarity between the target distance and the distance corresponding to each first historical recommendation information, and the similarity between the target distance and the distance corresponding to each second historical recommendation information, as the business execution rate corresponding to the candidate information.

[0219] Optionally, the business execution rate prediction module 401 is specifically configured to: determine the attention weight corresponding to each first historical recommendation based on the similarity between the target distance and the distance corresponding to each first historical recommendation; determine the attention weight corresponding to each second historical recommendation based on the similarity between the target distance and the distance corresponding to each second historical recommendation; weight the distances corresponding to each first historical recommendation based on the attention weights corresponding to each first historical recommendation to obtain a weighted distance corresponding to each first historical recommendation; weight the distances corresponding to each second historical recommendation based on the attention weights corresponding to each second historical recommendation to obtain a weighted distance corresponding to each second historical recommendation; and determine the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the weighted distances corresponding to each first historical recommendation and each second historical recommendation, using this probability as the business execution rate corresponding to the candidate information.

[0220] Optionally, the business execution rate prediction module 401 is specifically used to determine each first sub-distance sequence from the first distance sequence composed of the distances corresponding to each first historical recommendation information through a preset sliding window, and to determine each second sub-distance sequence from the second distance sequence composed of the distances corresponding to each second historical recommendation information; for each first sub-distance sequence, to determine the distance feature between the target distance and the first sub-distance sequence based on the similarity between the target distance and the distances corresponding to each first historical recommendation information contained in the first sub-distance sequence; and for each second sub-distance sequence, to determine the distance feature between the target distance and the second sub-distance sequence based on the similarity between the target distance and the distances corresponding to each second historical recommendation information contained in the second sub-distance sequence; and to determine the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the distance features corresponding to each first sub-distance sequence and the distance features corresponding to each second sub-distance sequence, as the business execution rate corresponding to the candidate information.

[0221] Optionally, the business execution rate prediction module 401 is specifically used to input the historical business data and the candidate information into a prediction model to be trained, so as to determine, based on the environment in which the user is in when executing the business corresponding to the historical business data, the probability that the user clicks on the candidate information, as the predicted click-through rate of the candidate information, and the probability that the user executes the candidate information after assuming that the user clicks on the candidate information, as the business conversion rate of the candidate information; and to determine the business execution rate of the candidate information based on the predicted click-through rate and the business conversion rate.

[0222] Optionally, the device further includes:

[0223] Click-through rate prediction module 403 is used to predict the probability of a user clicking on the candidate information based on the environment in which the user performs the business corresponding to the historical business data before training the prediction model, with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples. This prediction model is used as the predicted click-through rate corresponding to the historical recommendation information.

[0224] The training module 402 is specifically used to train the prediction model with the optimization objectives of minimizing the deviation between the predicted click-through rate and the click-through rate label, and minimizing the deviation between the business execution rate and the business execution rate label.

[0225] Figure 5 A schematic diagram of an information recommendation device provided in this specification specifically includes:

[0226] The acquisition module 500 is used to acquire the candidate information to be recommended to the user and the user's historical business data;

[0227] The business execution rate prediction module 501 is used to input each candidate information and the historical business data into a pre-trained prediction model, so that the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the environment in which the user executes the business corresponding to the historical business data, and uses this probability as the business execution rate corresponding to the candidate information. The prediction model is trained by the prediction model training method described above.

[0228] The recommendation module 502 is used to recommend information to the user based on the business execution rate corresponding to each candidate information.

[0229] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 2 The training method of the provided prediction model or the above Figure 3 The information provided is recommended by the method.

[0230] This instruction manual also provides Figure 6 The diagram shows a schematic structural representation of the electronic device. Figure 6 At the hardware level, the electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for the business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 2 The training method of the prediction model or the above Figure 3 The method of information recommendation described herein. Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. In other words, the execution subject of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0231] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0232] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0233] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0234] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0235] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0236] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0237] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0238] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0239] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0240] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0241] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0242] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0243] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0244] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0245] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0246] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for training a prediction model, characterized in that, The prediction model includes a first network, and the method includes: Obtain training samples, which contain the user's historical business data and candidate information to be recommended to the user; The historical business data and the candidate information are input into the prediction model to be trained. Based on the environment in which the user is in when executing the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information, which is used as the business execution rate corresponding to the candidate information. The prediction model is trained with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples; Specifically, based on the environment in which the user executes the service corresponding to the historical service data, the prediction model determines the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information, which is used as the service execution rate corresponding to the candidate information. This includes: Based on the historical business data, the historical recommendation information that the user clicked before recommending the candidate information to the user is determined as the first historical recommendation information, and the historical recommendation information corresponding to the business performed by the user is determined as the second historical recommendation information. For each first historical recommendation, determine the service loss characteristics when the user performs the service corresponding to the first historical recommendation; and for each second historical recommendation, determine the service loss characteristics when the user performs the service corresponding to the second historical recommendation. The service loss characteristics corresponding to each first historical recommendation information, the service loss characteristics corresponding to each second historical recommendation information, and the environmental data of the user's environment when recommending the candidate information to the user are input into the first network, so as to determine the service execution rate corresponding to the candidate information based on the first network and the environment in which the user is located when executing the service corresponding to the historical service data.

2. The method as described in claim 1, characterized in that, The service loss characteristics corresponding to each first historical recommendation information, the service loss characteristics corresponding to each second historical recommendation information, and the environmental data of the user's environment when recommending the candidate information to the user are input into the first network. Based on the first network, the service execution rate corresponding to the candidate information is determined based on the environment in which the user is in when executing the service corresponding to the historical service data. Specifically, this includes: For each first historical recommendation, determine the similarity between the environment in which the user clicked the first historical recommendation and the environment corresponding to the environment data; and for each second historical recommendation, determine the similarity between the environment in which the user executed the second historical recommendation and the environment corresponding to the environment data. Based on the business loss characteristics corresponding to the first historical recommendation information whose similarity meets preset conditions, and the business loss characteristics corresponding to the second historical recommendation information whose similarity meets preset conditions, the business execution rate corresponding to the candidate information is determined based on the environment in which the user is in when executing the business corresponding to the historical business data.

3. The method as described in claim 1 or 2, characterized in that, The service loss characteristics corresponding to each first historical recommendation information, the service loss characteristics corresponding to each second historical recommendation information, and the environmental data of the user's environment when recommending the candidate information to the user are input into the first network. Based on the first network, the service execution rate corresponding to the candidate information is determined based on the environment in which the user is in when executing the service corresponding to the historical service data. Specifically, this includes: For each second historical recommendation, determine the correlation between the second historical recommendation and each first historical recommendation; Based on the correlation between the second historical recommendation information and each first historical recommendation information, the business loss characteristics corresponding to the second historical recommendation information are adjusted. Based on the business loss characteristics corresponding to each first historical recommendation, the adjusted business loss characteristics corresponding to each second historical recommendation, and the environmental data, the business execution rate corresponding to the candidate information is determined based on the environment in which the user is in when executing the business corresponding to the historical business data.

4. The method as described in claim 1, characterized in that, The business loss characteristics include those used to characterize the amount a user spends at the merchant to which the historical recommendation information belongs and / or the distance between the merchant to which the historical recommendation information belongs and the user when the historical recommendation information is recommended to the user.

5. The method as described in claim 1, characterized in that, The prediction model also includes: Second network; The historical business data and the candidate information are input into a prediction model to be trained. Based on the environment in which the user executes the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information. This probability is used as the business execution rate corresponding to the candidate information. Specifically, this includes: Based on the historical business data, the historical recommendation information that the user clicked before recommending the candidate information to the user is determined as the first historical recommendation information, and the historical recommendation information corresponding to the business performed by the user is determined as the second historical recommendation information. For each first historical recommendation, the distance between the user and the merchant to which the first historical recommendation belongs is determined as the distance corresponding to the first historical recommendation. For each second historical recommendation, the distance between the user and the merchant to which the second historical recommendation belongs is determined as the distance corresponding to the second historical recommendation. Based on the distance corresponding to each first historical recommendation and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information is determined, which is used as the business execution rate corresponding to the candidate information.

6. The method as described in claim 5, characterized in that, Based on the distance corresponding to each first historical recommendation and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information is determined, which is used as the service execution rate corresponding to the candidate information. Specifically, this includes: When recommending the candidate information to the user, the distance between the user and the merchant to which the candidate information belongs is determined as the target distance; Based on the similarity between the target distance and the distance corresponding to each first historical recommendation, and the similarity between the target distance and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information is determined, and is used as the service execution rate corresponding to the candidate information.

7. The method as described in claim 6, characterized in that, Based on the similarity between the target distance and the distance corresponding to each first historical recommendation, and the similarity between the target distance and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information is determined, which is used as the service execution rate corresponding to the candidate information. Specifically, this includes: Based on the similarity between the target distance and the distance corresponding to each first historical recommendation, the attention weight corresponding to each first historical recommendation is determined; and based on the similarity between the target distance and the distance corresponding to each second historical recommendation, the attention weight corresponding to each second historical recommendation is determined. Based on the attention weight corresponding to each first historical recommendation, the distances corresponding to each first historical recommendation are weighted to obtain the weighted distances corresponding to each first historical recommendation. Based on the attention weight corresponding to each second historical recommendation, the distances corresponding to each second historical recommendation are weighted to obtain the weighted distances corresponding to each second historical recommendation. Based on the weighted distance corresponding to each first historical recommendation and each second historical recommendation, the probability that the user clicks on the candidate information and executes the corresponding business is determined, which is used as the business execution rate corresponding to the candidate information.

8. The method as described in claim 6, characterized in that, Based on the similarity between the target distance and the distance corresponding to each first historical recommendation, and the similarity between the target distance and the distance corresponding to each second historical recommendation, the probability that the user clicks on the candidate information and executes the service corresponding to the candidate information is determined, which is used as the service execution rate corresponding to the candidate information. Specifically, this includes: Using a preset sliding window, each first sub-distance sequence is determined from the first distance sequence composed of the distances corresponding to each first historical recommendation information, and each second sub-distance sequence is determined from the second distance sequence composed of the distances corresponding to each second historical recommendation information; For each first sub-distance sequence, the distance feature between the target distance and the first sub-distance sequence is determined based on the similarity between the target distance and the distance corresponding to each first historical recommendation information contained in the first sub-distance sequence; and for each second sub-distance sequence, the distance feature between the target distance and the second sub-distance sequence is determined based on the similarity between the target distance and the distance corresponding to each second historical recommendation information contained in the second sub-distance sequence. Based on the distance features corresponding to each first sub-distance sequence and each second sub-distance sequence, the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information is determined, and is used as the business execution rate corresponding to the candidate information.

9. The method as described in claim 1, characterized in that, The historical business data and the candidate information are input into a prediction model to be trained. Based on the environment in which the user executes the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information. This probability is used as the business execution rate corresponding to the candidate information. Specifically, this includes: The historical business data and the candidate information are input into the prediction model to be trained, so that the prediction model determines the probability that the user clicks on the candidate information based on the environment in which the user performs the business corresponding to the historical business data, which is the predicted click-through rate of the candidate information; and the probability that the user performs the candidate information after the user clicks on the candidate information, which is the business conversion rate of the candidate information. Based on the predicted click-through rate and the business conversion rate, the business execution rate corresponding to the candidate information is determined.

10. The method as described in claim 9, characterized in that, Before training the prediction model, with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples, the method further includes: Based on the environment in which the user is in when performing the business corresponding to the historical business data, the prediction model predicts the probability of the user clicking on the candidate information, which is then used as the predicted click-through rate of the candidate information. The prediction model is trained with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples, specifically including: The prediction model is trained with the optimization objectives of minimizing the deviation between the predicted click-through rate and the click-through rate label, and minimizing the deviation between the business execution rate and the business execution rate label.

11. A method for information recommendation, characterized in that, include: Obtain the candidate information that needs to be recommended to the user, as well as the user's historical business data; For each candidate information, the candidate information and the historical business data are input into a pre-trained prediction model, so that the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the environment in which the user is in when the user executes the business corresponding to the historical business data, and uses the probability as the business execution rate corresponding to the candidate information. The prediction model is trained by the method described in any one of claims 1 to 10 above. Based on the business execution rate corresponding to each candidate information, information is recommended to the user.

12. A training device for a prediction model, characterized in that, The prediction model includes a first network, and the device includes: The acquisition module is used to acquire training samples, which contain the user's historical business data and candidate information recommended to the user; The business execution rate prediction module is used to input the historical business data and the candidate information into the prediction model to be trained, so that, based on the environment in which the user is in when executing the business corresponding to the historical business data, the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information, and uses the probability as the business execution rate corresponding to the candidate information. The training module is used to train the prediction model with the optimization objective of minimizing the deviation between the business execution rate and the label information of the training samples; Specifically, the business execution rate prediction module is used to determine, based on the historical business data, the historical recommendation information that the user clicked before recommending the candidate information to the user, as the first historical recommendation information, and the historical recommendation information corresponding to the business that the user executed, as the second historical recommendation information. For each first historical recommendation, determine the service loss characteristics when the user performs the service corresponding to the first historical recommendation; and for each second historical recommendation, determine the service loss characteristics when the user performs the service corresponding to the second historical recommendation. The service loss characteristics corresponding to each first historical recommendation information, the service loss characteristics corresponding to each second historical recommendation information, and the environmental data of the user's environment when recommending the candidate information to the user are input into the first network, so as to determine the service execution rate corresponding to the candidate information based on the first network and the environment in which the user is located when executing the service corresponding to the historical service data.

13. An information recommendation device, characterized in that, include: The acquisition module is used to acquire the candidate information that needs to be recommended to the user and the user's historical business data; The business execution rate prediction module is used to input the candidate information and the historical business data into a pre-trained prediction model for each candidate information, so that the prediction model determines the probability that the user clicks on the candidate information and executes the business corresponding to the candidate information based on the environment in which the user executes the business corresponding to the historical business data, and uses the probability as the business execution rate corresponding to the candidate information. The prediction model is trained by the method described in any one of claims 1 to 10 above. The recommendation module is used to recommend information to the user based on the business execution rate corresponding to each candidate information.

14. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 11.

15. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1 to 11.

Images